TABLE OF CONTENTS - Research Explorer

TABLE OF CONTENTS - Research Explorer
Localisation Of Wireless Sensor Nodes In Confined Industrial Processes
A thesis submitted to the University of Manchester for the degree of
Doctor of Philosophy (PhD)
in the Faculty of Engineering and Physical Sciences
2012
Mr Michalis Antoniou
School of Electrical and Electronic Engineering, Microwave and
Communication Systems Group
Table of contents
Table of contents................................................................................................................................... 2
List of figures........................................................................................................................................ 6
List of tables ....................................................................................................................................... 14
List of Abbreviations .......................................................................................................................... 15
Abstract .............................................................................................................................................. 17
Declaration ......................................................................................................................................... 18
Copyright Statement ........................................................................................................................... 18
Dedication........................................................................................................................................... 19
Acknowledgement .............................................................................................................................. 19
List of Publications ............................................................................................................................. 20
CHAPTER 1 INTRODUCTION ..................................................................................................... 21
1.1 Introduction .................................................................................................................................. 21
1.2 Overview of Wireless Sensor Networks ....................................................................................... 22
1.3 WSN applications ......................................................................................................................... 23
1.3.1 Environmental monitoring ......................................................................................................... 23
1.3.2 Military applications .................................................................................................................. 25
1.3.3 Wireless sensor networks for industrial processes ..................................................................... 26
1.4 Localisation .................................................................................................................................. 30
1.5 Thesis motivation and objectives .................................................................................................. 31
1.6 Related work ................................................................................................................................. 33
1.6.1 Grain monitoring using acoustic signals .................................................................................... 33
1.6.2 Grain monitoring using narrow band signals ............................................................................. 34
1.7 Thesis layout ................................................................................................................................. 38
CHAPTER 2 BASIC RANGING AND LOCALISATION METHODS....................................... 40
2.1 Introduction .................................................................................................................................. 40
2.2 Basic ranging techniques .............................................................................................................. 40
2.2.1 Time-based ranging ................................................................................................................... 40
2.2.2 Received signal strength (RSS)-based ranging........................................................................... 43
2.3 Basic localisation techniques ........................................................................................................ 45
2.3.1 Trilateration ............................................................................................................................... 46
2.3.2 Multilateration ........................................................................................................................... 47
2.4 Well-known WSN localisation algorithms .................................................................................... 49
2.4.1 Ad Hoc localisation system ........................................................................................................ 49
2.4.2 DV-Hop algorithm ..................................................................................................................... 50
.
2
2.4.3 DV-Distance algorithm .............................................................................................................. 52
2.4.4 Iterative localisation algorithms ................................................................................................. 53
2.4.5 Centroid algorithm ..................................................................................................................... 55
2.4.6 Weighted centroid localisation (WCL) algorithm ...................................................................... 56
2.4.7 Ecolocation algorithm ................................................................................................................ 57
2.5 Summary....................................................................................................................................... 60
CHAPTER 3 UWB TECHNOLOGY FOR RANGING AND LOCALISATION ....................... 61
3.1 Introduction .................................................................................................................................. 61
3.2 UWB technology .......................................................................................................................... 61
3.3 UWB antenna and their impact on ranging and localisation ......................................................... 64
3.3.1 Antenna orientation .................................................................................................................... 67
3.4 UWB-based ranging and localisation review ................................................................................ 68
3.4.1 Basic features of UWB-based ranging and localisation applications ......................................... 68
3.4.2 UWB-based ranging and localisation applications ..................................................................... 69
3.5 Summary....................................................................................................................................... 80
CHAPTER 4 PROTOTYPE UWB-BASED LOCALISATION METHOD ................................. 81
4.1 Introduction .................................................................................................................................. 81
4.2 Basic localisation system .............................................................................................................. 81
4.3 Distance difference estimation ...................................................................................................... 83
4.4 Localisation algorithm .................................................................................................................. 86
4.5 Algorithm implementation ............................................................................................................ 88
4.6 The localisation system ................................................................................................................. 90
4.6.1 Electronic systems ..................................................................................................................... 90
4.6.2 Mechanical system ..................................................................................................................... 93
4.7 Experimental investigation ........................................................................................................... 97
4.7.1 Analysis of experimental results .............................................................................................. 100
4.8 Summary..................................................................................................................................... 102
CHAPTER 5 ADAPTIVE LEADING EDGE DETECTION ...................................................... 104
5.1 Introduction ................................................................................................................................ 104
5.2 Basic advanced LE detection method ......................................................................................... 105
5.3 Improved advanced LE detection method ................................................................................... 109
5.4 Evaluation of the BALED method .............................................................................................. 112
5.4.1 Analysis of BALED results ...................................................................................................... 114
5.5 Evaluation of the IALED method ............................................................................................... 117
5.5.1 Analysis of IALED results ....................................................................................................... 118
5.6 Summary..................................................................................................................................... 121
.
3
CHAPTER 6 NOVEL LOCALISATION METHODS ................................................................ 123
6.1 Introduction ................................................................................................................................ 123
6.2 Theoretical distance difference profile analysis .......................................................................... 123
6.2.1 Analysis of 2D theoretical distance difference profiles ............................................................ 123
6.2.2 Analysis of 3D theoretical distance difference profiles ............................................................ 125
6.2.3 Properties of distance difference profiles ................................................................................. 128
6.3 Novel localisation algorithms ..................................................................................................... 135
6.4 Basic profile positioning method ................................................................................................ 135
6.4.1 MATLAB algorithm ................................................................................................................ 136
6.4.2 BMSF experimental evaluation................................................................................................ 138
6.4.3 BMLS experimental evaluation ............................................................................................... 140
6.4.4 BMLSD experimental evaluation............................................................................................. 142
6.5 Centroid based localisation ......................................................................................................... 145
6.5.1 MATLAB algorithm ................................................................................................................ 147
6.5.2 BCL experimental evaluation .................................................................................................. 147
6.5.3 ACL experimental evaluation .................................................................................................. 149
6.6 Comparative positioning method ................................................................................................ 152
6.6.1 MATLAB implementation ....................................................................................................... 153
6.6.2 Experimental evaluation .......................................................................................................... 154
6.7 Analytical localisation method .................................................................................................... 156
6.7.1 Experimental evaluation .......................................................................................................... 159
6.8 Summary..................................................................................................................................... 161
CHAPTER 7 NETWORK-BASED LOCALISATION ................................................................ 164
7.1 Introduction ................................................................................................................................ 164
7.2 WSN4IP nodes............................................................................................................................ 165
7.2.1 WSN4IP hardware and software .............................................................................................. 165
7.3 Node radiation characteristics ..................................................................................................... 166
7.3.1 Antenna response in air ............................................................................................................ 167
7.3.2 Antenna response inside grain ................................................................................................. 169
7.3.3 Antenna response whilst inside mortar shells........................................................................... 171
7.4 The effect of antenna orientation ................................................................................................ 172
7.5 The effect of WSN4IP antenna orientation ................................................................................. 174
7.6 Networking experiments in grain ................................................................................................ 176
7.6.1 Experimental configuration ...................................................................................................... 176
7.6.2 Networking results ................................................................................................................... 179
7.6.3 Localisation results .................................................................................................................. 181
.
4
7.7 Summary..................................................................................................................................... 183
CHAPTER 8 DISCUSSION AND FUTURE WORK .................................................................. 184
8.1 Discussion ................................................................................................................................... 184
8.2 Thesis summary .......................................................................................................................... 184
8.3 Future work on UWB-based localisation .................................................................................... 187
8.3.1 UWB-receivers ........................................................................................................................ 187
8.3.2 Placement of receiving antennas .............................................................................................. 189
8.4 Future work on network-based localisation ................................................................................ 191
8.5 Future work on the node‘s power supply .................................................................................... 193
REFERENCES ............................................................................................................................... 195
Total word count: 44.366
.
5
List of figures
Figure 1. WSN architecture ..................................................................................................... 22
Figure 2.GlacsWeb sensor network deployment ..................................................................... 24
Figure 3. Antenna characterisation inside biomass ................................................................. 27
Figure 4. RSS over distance inside silage ............................................................................... 28
Figure 5.WSNs embedded in a silage stack ............................................................................ 29
Figure 6. WSN to monitor grain silo ...................................................................................... 32
Figure 7. Attenuation measurements inside grain at 915MHz ............................................... 36
Figure 8. Attenuation at 915MHz ........................................................................................... 36
Figure 9. Grain tests experimental setup.................................................................................. 37
Figure 10. Experimental RSSI over distance .......................................................................... 38
Figure 11. Experimental cross-correlation .............................................................................. 42
Figure 12.Theoretical and experimental RSS ......................................................................... 44
Figure 13.Trilateration diagram ............................................................................................... 46
Figure 14. Spherical multilateration example .......................................................................... 47
Figure 15. Atomic multilateration............................................................................................ 49
Figure 16. Iterative multilateration .......................................................................................... 50
Figure 17. DV-Hop localisation ............................................................................................... 51
Figure 18. Simulated DV-Distance localisation ...................................................................... 52
Figure 19. Centroid localisation ............................................................................................... 56
Figure 20. WCL localisation .................................................................................................... 57
Figure 21. Ecolocation position estimate ................................................................................ 59
Figure 22. 0.5ns monocycle ....................................................................................................59
Figure 23.Train of 0.5ns pulses................................................................................................ 61
Figure 24. Wide-band time spectrum....................................................................................... 62
Figure 25.Narrow-band and UWB spectrum coverage ........................................................... 63
.
6
Figure 26. Monopole antenna magnetic field ......................................................................... 65
Figure 27.Cable transmitted UWB pulse.................................................................................66
Figure 28.Wireless transmission .............................................................................................. 66
Figure 29. UWB antenna ringing effect ................................................................................... 67
Figure 30. Orientation effect on UWB pulses ......................................................................... 68
Figure 31.Received signal in free space...................................................................................69
Figure 32.Indoors received signal ............................................................................................ 69
Figure 33.Block diagram of UWB ranging tests ..................................................................... 70
Figure 34.UWB positioning system ........................................................................................ 71
Figure 35.PAL localisation system ......................................................................................... 73
Figure 36.Low cost UWB positioning experimental setup ..................................................... 73
Figure 37.Ranging tests at various scenarios .......................................................................... 75
Figure 38. LE detection algorithms ........................................................................................ 77
Figure 39.Noise floor in a received UWB pulse ..................................................................... 79
Figure 40.UWB-based localisation layout ............................................................................... 82
Figure 41.Ideal received pulse train.........................................................................................82
Figure 42.Real received UWB pulse ....................................................................................... 84
Figure 43. Gated LE at reference antenna ............................................................................... 84
Figure 44. Distance difference diagram ................................................................................... 85
Figure 45. Localisation using circle intersection ..................................................................... 86
Figure 46. Circular radii intersection ...................................................................................... 88
Figure 47. Variance minimisation Vs assumed distances ....................................................... 89
Figure 48. UWB positioning in a grain silo ............................................................................ 90
Figure 49. Pulse generator circuit ........................................................................................... 91
Figure 50. Ring monopole transmit and receiving antennas.................................................... 91
Figure 51. Transmitting antenna for tests in the large vessel .................................................. 92
.
7
Figure 52. DSO UWB pulse trail ............................................................................................. 93
Figure 53. Experimental rig ..................................................................................................... 94
Figure 54. Small scale experimental setup .............................................................................. 95
Figure 55. Single receiving antenna on a rotating rig .............................................................. 96
Figure 56. Small vessel receiving antenna on a cross formation ............................................. 96
Figure 57. Large silo and receiving antenna ............................................................................ 97
Figure 58. X axis position estimates from prototype method .................................................. 98
Figure 59. AE for experiments across X axis .......................................................................... 98
Figure 60. Z axis position estimates from prototype method .................................................. 99
Figure 61. AE for experiments across Z axis .......................................................................... .99
Figure 62.Distance difference X axis.......................................................................................99
Figure 63.Distance difference Z axis.......................................................................................99
Figure 64.Raw data at reference (0cm)..................................................................................101
Figure 65.Gated LE at reference (0cm) ................................................................................. 101
Figure 66.Raw signal at reference at 10cm ...........................................................................102
Figure 67.LE at 10cm from the reference .............................................................................. 102
Figure 68.Noise floor identification ....................................................................................... 105
Figure 69. Histogram of received signal noise floor.............................................................. 106
Figure 70. LE at X=10cm ...................................................................................................... 107
Figure 71. BALED-based localisation method ...................................................................... 108
Figure 72.Raw received signal at 60cm.................................................................................109
Figure 73. Detected LE at 60cm ............................................................................................ 109
Figure 74.Raw received signal at 40cm.................................................................................109
Figure 75.Detected LE at 40cm ............................................................................................. 109
Figure 76. LE minima and maxima points............................................................................. 110
.
8
Figure 77. Raw received signal at 40cm................................................................................110
Figure 78. Detected LE at 40cm ............................................................................................ 110
Figure 79. 3ns Pulse limit example ........................................................................................ 111
Figure 80. IALED localisation method flow chart ................................................................ 112
Figure 81.Measured and estimated X axis positions ............................................................. 113
Figure 82. AE for X axis positions ........................................................................................ 113
Figure 83.Measured and estimated Z axis positions .............................................................. 113
Figure 84. AE for Z position estimates .................................................................................. 114
Figure 85. Distance difference X axis....................................................................................114
Figure 86. Distance difference Z ........................................................................................... 114
Figure 87. Distance difference profiles at X=53cm, Y=89cm (experiment number 3) ......... 115
Figure 88.profiles at X=42cm, Y=72cm (experiment number 3) .......................................... 116
Figure 89.Profiles at Z=44cm, Y=45cm (experiment number 16) ........................................ 116
Figure 90.Measured and estimated X axis positions ............................................................. 117
Figure 91. AE for X axis position estimates .......................................................................... 117
Figure 92. Measured and estimated Z axis positions ............................................................. 117
Figure 93. AE for measurements across Z axis ..................................................................... 118
Figure 94.Distance difference X axis.....................................................................................118
Figure 95.Distance difference Z axis ..................................................................................... 118
Figure 96.Profiles at Z=44cm, Y=45cm (experiment number 16) ........................................ 119
Figure 97.Profile at X=70cm, Y=30cm (experiment number 11) ......................................... 119
Figure 98.Reference received signal......................................................................................120
Figure 99. LE at reference antenna ........................................................................................ 120
Figure 100.Raw received signal at 60cm...............................................................................120
Figure 101.Detected LE at 60cm ........................................................................................... 120
.
9
Figure 102.Raw signal at X=28, Y=89cm..............................................................................121
Figure 103.Raw signal at X=17, Y=150cm ........................................................................... 121
Figure 104.2D distance difference diagram ........................................................................... 124
Figure 105. Profiles at Xt=28cm............................................................................................124
Figure 106. Profiles at Xt=72cm............................................................................................ 124
Figure 107.Profile at Xt=50cm .............................................................................................. 125
Figure 108.3D distance difference profile on a plane ............................................................ 126
Figure 109.Distance difference profile on a plane ................................................................. 126
Figure 110.Profile at X=20,Y=50,Z=80cm...........................................................................127
Figure 111.Profile at X=80,Y=50,Z=20cm ......................................................................... ..127
Figure 112.Profile at X=80,Y=50,Z=20cm ...........................................................................127
Figure 113.Profile at X=50,Y=10,Z=50cm ........................................................................... 127
Figure 114.2D profile for transmitter at (X,Y)=(50,50) ........................................................ 128
Figure 115. Contour plot for (Xt,Zt)=(50cm,50cm) .............................................................. 131
Figure 116.Maximum distance difference Vs depth .............................................................. 133
Figure 117.Illustration of distance profile uniqueness ........................................................... 133
Figure 118. Theoretical profiles ............................................................................................. 136
Figure 119. Profile for Xt=44cm, Yt= 30cm .......................................................................... 137
Figure 120. Profile for X=44, Y=30.......................................................................................138
Figure 121.Profile for X=45, Y=30 ....................................................................................... 138
Figure 122.BMSF positions along X axis .............................................................................. 138
Figure 123.X axis coordinates error....................................................................................... 139
Figure 124.BMSF positions along Z axis .............................................................................. 139
Figure 125. Z axis coordinates error ...................................................................................... 139
Figure 126.BMLS measured and estimated X axis positions ................................................ 140
.
10
Figure 127.X axis coordinates error...................................................................................... 141
Figure 128.BMLS measured and estimated Z axis positions ................................................ 141
Figure 129.Z axis coordinates error ....................................................................................... 141
Figure 130. BMLD measured and estimated X axis positions .............................................. 142
Figure 131.X axis coordinates error....................................................................................... 143
Figure 132.BMLD measured and estimated Z axis positions ................................................ 143
Figure 133.Z axis coordinates error ....................................................................................... 143
Figure 134.Basic localisation Z=65, Y=30cm ....................................................................... 144
Figure 135.Equidistance receiving antenna positions ............................................................ 145
Figure 136. BCL plots............................................................................................................ 146
Figure 137.Measured and estimated X axis positions ........................................................... 148
Figure 138.X axis coordinates error....................................................................................... 148
Figure 139.Measured and estimated Z axis positions ............................................................ 148
Figure 140. Z axis coordinates error ...................................................................................... 149
Figure 141.Measured and estimated X axis positions ........................................................... 150
Figure 142. ACL X axis coordinates error............................................................................. 150
Figure 143.Measured and estimated Z axis positions ............................................................ 150
Figure 144. ACL Z axis coordinates error ............................................................................. 151
Figure 145.Basic centroid localisation Zt=43cm, Yt=92cm................................................... 152
Figure 146.Comparative positioning diagram ....................................................................... 153
Figure 147.Measured and estimated X axis positions ........................................................... 154
Figure 148. X axis coordinates error...................................................................................... 154
Figure 149. Measured and estimated Z axis positions ........................................................... 155
Figure 150. Z axis coordinates error ...................................................................................... 155
Figure 151. Receive and transmit position coordinates ......................................................... 156
Figure 152.Measured and estimated X axis positions ........................................................... 159
.
11
Figure 153.X axis coordinates error....................................................................................... 159
Figure 154.Measured and estimated Z axis positions ............................................................ 160
Figure 155.Z axis coordinates error ....................................................................................... 160
Figure 156. Errors in profiles Vs errors in Yt ........................................................................ 162
Figure 157.WSN4IP sensor node integrated parts ................................................................ 165
Figure 158.WSN4IP sensor node........................................................................................... 166
Figure 159.Errors in profiles Vs errors in Yt ......................................................................... 167
Figure 160. Quarter wave monopole antenna ........................................................................ 167
Figure 161.Elevation ZY radiation pattern ............................................................................ 168
Figure 162.Azimuth XY radiation pattern ............................................................................. 168
Figure 163. Antenna response in air ..................................................................................... 169
Figure 164. Antenna response in air and in grain .................................................................. 170
Figure 165. Tuned antenna inside grain ................................................................................. 171
Figure 166.Antenna inside the protective sphere .................................................................. 171
Figure 167.Antenna response inside the protective sphere ................................................... 172
Figure 168.Antenna orientation positions .............................................................................. 173
Figure 169.Antenna tests inside grain and in air.................................................................... 173
Figure 170.WSN4IP node Vs dipole in anechoic chamber starting at 0o .............................. 174
Figure 171. Radiation pattern for the WSN4IP node at 0o .................................................... 175
Figure 172. WSN4IP antenna 3D radiation pattern ............................................................... 175
Figure 173. Nodes coordinates inside silo ............................................................................. 177
Figure 174. Experimental apparatus ...................................................................................... 178
Figure 175. Network connectivity ......................................................................................... 179
Figure 176. RSS Vs distance for nodes 3 and 8 ..................................................................... 180
Figure 177. RSS based centroid localisation ......................................................................... 182
Figure 178. Ecolocation algorithm results ............................................................................. 183
.
12
Figure 179. Alternative receiving antenna placements .......................................................... 189
Figure 180. Random antenna orientation ............................................................................... 191
Figure 181. Angular antennas around a sensor node ............................................................ 192
.
13
List of tables
Table 1: Prototype localisation MAE and STD in cm ............................................................. 99
Table 2: RGLED distance difference profiles along X axis .................................................. 100
Table 3:RGLED distance difference profile errors ................................................................ 101
Table 4: BALED MAE and STD ........................................................................................... 114
Table 5: BALED distance difference profile errors ............................................................... 115
Table 6: IALED localisation method MAE and STD ............................................................ 118
Table 7: MAE for IAELED distance difference profile errors .............................................. 119
Table 8: MAE for all position estimates ................................................................................ 121
Table 9: Distance difference MAE ........................................................................................ 122
Table 10: MAE and STD for the BMSF ................................................................................ 140
Table 11: MAE and STD for the BMLS method ................................................................... 142
Table 12: MAE and STD for the BMLSD ............................................................................. 144
Table 13: MAE and STD for the BCL method ...................................................................... 149
Table 14: MAE and STD for the ACL method...................................................................... 151
Table 15: MAE and STD for the comparative method .......................................................... 155
Table 16: MAE and STD for the Analytical method ............................................................. 160
Table 17: MAEs for all localisation methods in the small vessel ......................................... 161
Table 18: MAEs for all localisation methods in the large vessel .......................................... 163
Table 19: Sub-sampling MAEs.............................................................................................. 188
Table 20: Depth accuracy ...................................................................................................... 190
.
14
List of Abbreviations
2D-Two Dimensional
3D- Three Dimensional
AHLoS -Ad Hoc Localization System
APS -Ad hoc Positioning
ADC- Analogue to Digital Converter
ACL- Advanced centroid localisation
BALED-Basic Adaptive Leading Edge Detection
BCL- Basic centroid localisation
BMLS- Basic Method Least Squares
BMLSD- Basic Method Least Squares Interpolation Derivative
BMSF- Basic Method Straight Forward
DARPA -Defence Advanced Research Project Agency
DV- Distance Vector
DSO- Digital Sampling Oscilloscope
GPS- Global Positioning System
IALED-Improved Adaptive Leading Edge Detection
ISM -Industrial, Scientific and Medical
LE- Leading edge
LMS -Least Mean Square
LNA- Low Noise Amplifier
LOS- Line Of Sight
LS- Least Squares
ML- Maximum Likelihood
.
15
NLOS- Non Line Of Sight
PC- Personal Computer
RF- Radio frequency
RGLED- Range Gate Leading Edge Detection
RSS -Received Signal Strength
RTT - Round-trip-time
SMA- Sub-Miniature version A
SNR- Signal to Noise Ratio
SHM- Self-Healing Minefield
TDoA- Time Difference of Arrival
ToA- Time of Arrival
ToF – Time of Flight
UWB -Ultra Wide Band
VNA- Vector Network Analyser
WCL- Weighted Centroid Localization
WSN -Wireless Sensor Network
WSN4IP- Wireless Sensor Networks for Industrial Processes
.
16
Abstract
Work described in this thesis is concerned with localisation techniques, for determining the
position, of wireless sensors whilst immersed in confined industrial processes, such as those
occurring in the chemical, pharmaceutical and food processing industries.
Two different approaches to localisation were investigated. The first approach employed an
existing hardware system that used ultra wide band (UWB) signals whist the second approach
used a network localisation method based on information from narrow-band received signals.
A prototype UWB-based localisation algorithm processed experimental received UWB
pulses to detect their leading edges (LE) that were used to derive Time Difference of Arrival
(TDoA) data. In turn TDoA data were translated into distances and used to compute the
locations of the sensor nodes. Nevertheless, the process of detecting the LEs caused
significant errors in the localisation process. To deal with this problem new automated
adaptive LE detection methods were derived that succeeded in reducing localisation errors by
half, compared to the prototype method, reaching accuracies of ±2cm.
A thorough analysis of TDoA profiles revealed that these follow specific trends depending on
the positions of the sensor nodes. A number of properties of TDoA profiles are proved
mathematically and incorporated into seven localisation algorithms. These algorithms were
examined using experimental TDoA data and shown to achieve average localisation errors up
to 3cm.
Network-based localisation was examined at a later stage of this research since complexities
of large scale measurements and difficulties with equipment, delayed acquiring experimental
data. The deployed network consisted of a number of nodes whose positions were known
(anchors) that were used to estimate the positions of sensor nodes whose positions where
considered to be unknown. Localisation was based on received signal strength (RSS) data, at
every node to be localised, in anticipation that RSS could provide distance information that
could be used in the localisation procedure. Nevertheless, fluctuations in RSS only allowed
using localisation algorithms that associated RSS to the positions of anchors. The average
localisation error in the network-based localisation algorithms was between 30cm to 100cm.
.
17
Declaration
No portion of the work referred to in this thesis has been submitted in support of an
application for another degree or qualification of this or any other university or other institute
of learning.
Copyright Statement
1. The author of this thesis (including any appendices and/or schedules to this thesis) owns
certain copyright or related rights in it (the ―Copyright‖) and s/he has given The University of
Manchester certain rights to use such Copyright, including for administrative purposes.
2. Copies of this thesis, either in full or in extracts and whether in hard or electronic copy,
may be made only in accordance with the Copyright, Designs and Patents Act 1988 (as
amended) and regulations issued under it or, where appropriate, in accordance with licensing
agreements which the University has from time to time. This page must form part of any such
copies made.
3. The ownership of certain Copyright, patents, designs, trademarks and other intellectual
property (the ―Intellectual Property‖) and any reproductions of copyright works in the thesis,
for example graphs and tables (―Reproductions‖), which may be described in this thesis, may
not be owned by the author and may be owned by third parties. Such Intellectual Property
and Reproductions cannot and must not be made available for use without the prior written
permission of the owner(s) of the relevant Intellectual Property and/or Reproductions.
4. Further information on the conditions under which disclosure, publication and
commercialisation of this thesis, the Copyright and any Intellectual Property and/or
Reproductions described in it may take place is available in the University IP Policy (see
http://www.campus.manchester.ac.uk/medialibrary/policies/ intellectual-property.pdf), in any
relevant Thesis restriction declarations deposited in the University Library, The University
Library's regulations (see http://www.manchester.ac.uk/library/aboutus/regulations) and in
The University's policy on presentation of Theses.
.
18
Dedication
I dedicate my thesis to my parents and to my sister for all of their encouragement and
understanding throughout my studies. Special dedication and prayers goes to my
grandparents whose advices and love i will never forget.
Acknowledgement
I would like to take this opportunity to express my sincere gratitude to my supervisor Dr.
Peter N. Green for his guidance and advice throughout my studies. Working together in
developing experimental arrangements and above all in developing novel localisation
methods was one of the most challenging and educational experiences of my life.
I am truly grateful to Dr Graham Parkinson for all of his valuable help with the RF side of my
work. Working together, in setting up the experiments, transferring hundreds of kilos of grain
and developing grain spillage countermeasures, made the whole experience much less
tedious.
I am also thankful to Dominic Crutchley with whom I had the pleasure to work with in
developing ―innovating‖ experimental procedures for tests inside grain silos. Dominic‘s
exceptional programming skills were vital in setting up a successful network of sensor nodes
inside a grain silo.
My acknowledgements goes to Dr Rob Sloan, Peter R. Green and Prof Trevor York for the
very constructive meetings and discussions we had throughout the duration of the WSN4IP
project. In addition, I would like to acknowledge Keith Khan and Vincet James, from the
electronics workshops, for the constructive debates we had on several aspects of electronics
and communication principles during laboratory demonstrations.
Lastly but not least i express my deepest thanks to my close friends in Manchester: Antonis
Phasouliotis, Rajasmita Gossuami, Jiaping Lu and Mohamed Babahani. Without them life in
Manchester wouldn‘t have been as rich as it was.
.
19
List of Publications
Accepted and Published
[1] G. Parkinson, D. Crutchley, P. M. Green, M. Antoniou, M. Boon, P. N. Green, P. R.
Green, R. Sloan, T. York, ―Enviromental Monitoring in Grain‖, IEEE, I2MTC, 2010.
[2] M. Antoniou, M.C.Boon, P.N.Green, P.R.Green and T.A.York, ―Wireless Sensor
Networks for Industrial Processes‖, Sensors Applications Symposium (SAS), IEEE,
USA, February,2009.
[3] M.Antoniou, P.N. Green, ―Adaptive leading edge detection for WSN localisation inside
industrial processes‖, IET, WSS2012, June 2012
Submitted
[1] M. Antoniou, P.N.Green ―Empirical Localisation Method for Wireless Sensor Nodes In
Confined Industrial Processes‖, 10th Workshop on Positioning, Navigation and
Communication, WPNC 2013, IEEE, Dresden, Germany, March 2013
[2] M. Antoniou, P.N.Green ―Analytical Localisation Method for Wireless Sensor Nodes
Embedded In Industrial Processes‖, 10th Workshop on Positioning, Navigation and
Communication, WPNC 2013, IEEE, Dresden, Germany, March 2013
.
20
CHAPTER 1
INTRODUCTION
1.1 Introduction
―Where am I?‖, ―Where is it?‖ These are two fundamental question humans have always
been asking. Positioning using the stars in the sky, geomorphologic fixtures, a compass and a
map have been widely used to allow travel all over the planet. In one of his voyages, captain
James Cook proclaimed that his aim was to go beyond anyone has ever been before, yet to go
as far as it was humanly possible to go [6]. Nevertheless, the continuous desire to understand
the surrounding cosmos led to the development of new technologies that allowed exploring
areas that once were considered out of reach such as the outer space.
Over the past 40 to 50 years a variety of advanced electronic positioning systems have been
developed, typically driven by the needs of the military and the civil aviation and maritime
sectors, culminating in the satellite-based Global Positioning System (GPS) [1, 4, 20,21]..
However, the ability to determine position is important in other contexts than navigation.
With the advent of wireless sensors (discussed in more detail in Section 1.2), it has become
possible to obtain measurements of parameters of interest over an area by simply distributing
the sensors over the region. A key motivating factor for the use of wireless sensors is ease of
deployment over potentially rough terrain or inaccessible locations. A consequence of this
kind if deployment is that the position of a node i.e. the position at which its measurements
are taken is typically known. However, it is often important for position to be associated with
a sensor measurement and hence there is considerable interest in the issue of positioning
wireless sensors.
The work reported in this thesis is concerned with localisation of sensor nodes that monitor
industrial processes inside confined spaces. A representative application from the food
industry was considered where sensors are used to monitor grain stored inside silos. To this
extent, localisation methods and practises presented in this thesis are concerned, but not
limited, to sensors embedded inside grain storage silos.
.
21
1.2 Overview of Wireless Sensor Networks
A Wireless Sensor Network (WSN) consists of numerous sensor nodes, scattered over an area
of concern with the aim of monitoring conditions that can improve analysis and further
understanding of that area [1]. A characteristic example of a WSN was the deployment of
wireless sensor nodes to investigate the nesting behaviour of seabirds, on a remote island,
were continues human presence was undesirable [2]. Data from the sensor nodes were
transmitted over the internet through a base station that had the capability of satellite
communications thus allowed researchers to process real-time readings remotely [2].
Each node in a WSN is equipped with a radio transceiver, a power source, usually a battery,
as well as several sensors and a microprocessor that allows gathering and processing data
locally [1].
The nodes in a WSN cooperate to create and maintain the network without any supporting
infrastructure such as base stations, switching centres or any other wired equipment; instead
nodes once deployed can establish communication links among them and act as routers to
forwarded information from one node to the next along an established route as seen in figure
1 [1, 3-5]. Nodes form links with neighbours that are in radio range and set inter-node
connections that define the network topology. Instantaneous network formation without any
preplanning is known as ad hoc topology.
Multi-hop network
Base station
Single-hop network
Figure 1. WSN architecture
.
22
Most WSNs include one or more base stations or gateways that provide a sink for the data
gathered by the network and an interface between the WSN and other systems, typically the
Internet as seen in figure 1. If the network has a single base station and all nodes have a direct
link to it, then the network has a star topology and is an example of a single-hop network
(figure 1) where the term hop means a direct radio link. However, in many WSNs, nodes
might be out of range of the base station and so if they wish to transmit data to it, it must be
sent via a number of intermediate nodes. Such a network is called a multi-hop network
(figure 1) and the task of finding a path between a node and the base station is known as
multi-hop routing [1, 3-5].
The infrastructure-less characteristics of ad hoc networks reduces the cost and increases the
flexibility of the network to adapt in different application scenarios [5].
In general, the design of WSNs is governed by constraints such as small size and low cost
[1]. However the most challenging constraint by far is sustainable low power consumption in
order to prolong the life-span of the nodes for as long as possible. This is especially important
in case were sensor are deployed to monitor environments that are not easily accessible [1, 2]
and so batteries cannot be replaced.
Data collection and processing as well as the employed communication methods determine,
to a great extent, the operational period of the nodes and as such various method have been
developed to deal with these issues. To further relax node‘s data processing constraints nodes
in a network can pass their information from node to node and finally to a base station with
higher processing capability which is able to broadcast information to users far away [1].
1.3 WSN applications
WSNs find applications in areas such as the military, the environment, health, space
exploration, and in industry. Some of these application areas are discussed next.
1.3.1 Environmental monitoring
Environmental monitoring has been a popular application for WSNs. Research in this area
includes deploying sensor networks in a remote island to monitor habitat patterns of seabirds
[2], to monitor underwater life [7], and even monitoring the harsh environment of Icelandic
glaciers [8-10].
.
23
The deployment of WSN to study glaciers (figure 2) in Iceland was an initiative by a research
team as part of the Glacsweb project [8-10]. This research is a good example of a practical
deployment of a WSN and so will be presented in some detail next.
The objective of this project was to provide a better understanding of the effect of
environmental changes at the base of the glaciers and how this affects glacial displacement
[9-11]. A number of sensor nodes were placed inside the glacier and one base station outside
on the top of the glacier to collect all the data, as shown in figure 2.
Base station
glacier ice
lower base
station receiver
Sensor
nodes
sediment
Figure 2. GlacsWeb sensor network deployment
The nodes were equipped with a number of sensors to observed parameters such as
conductivity, heat, temperature, pressure and light variations. Further, information on the tilt
and roll of the sensor was derived using 3-axis (x, y, z) tilt sensors. The node electronics were
enclosed within a polyester hull and placed inside ice using hot water drills.
Communication among nodes and the base station was achieved using transceivers operating
at 173MHz; since communication at higher frequencies usually used in WSNs2 results in high
levels of signal attenuation that degrades the process of data collection from the sensors. The
antennas used for communication were quarter wavelength helical antennas. However, it was
found necessary to ‗tune‘ the antennas (reducing their length) to obtain satisfactory
performance, since the signals propagated through ice and not air.
2
Frequencies in the unlicensed Industrial, Scientific and Medical (ISM) band are often used in WSNs. Thse
include 433, 868 and 2400 MHz.
.
24
Nodes near the top of the glacier transmitted their data directly to the base station in a single
hop approach which was proven to be power inefficient whilst nodes deep inside the glacier
broadcasted their data one to another in a multi-hop approach which proven to be beneficial
in terms of energy consumption.
The nodes were coordinated by software running on a battery powered microcontroller. To
save power, the microcontroller was placed in standby (sleep) mode for the most o for time
and was activated only once a day to transmit the collected data to the base station. The
purpose of the base station was to collect all the data from the sensor nodes for further
processing and to transmit data via a combination of GSM radio links and Internet, to the
research team at Southampton.
1.3.2 Military applications
Many military applications of WSNs have been proposed ranging from enemy and asset
surveillance to target tracking and elimination [11-13]. An interesting application is the
―clever‖ antitank landmines project, undertaken in the United States. After deployment, the
landmines developed in this work could communicate with each other forming ad-hoc WSN,
and can even change position [11]. This system of landmines is known as the Self-Healing
Minefield (SHM) and is so adaptable to the movements of the enemy that if part of the
minefield is breached it can be physically reorganised by moving part of the mines to cover
the breached area.
Each tank mine is equipped with a short range radio transceiver with a wide-beam antenna
and once it is placed on a mine field it communicates with other mines to form an ad-hoc
network. In addition, four rocket thrusters that are attached on each mine allow them to jump
into their new position if and when this is required.
Once deployed, mines transmit signals regularly to other mines in the network; an absence of
a signal at any time from one or more mines inside the network is interpreted as being caused
by a hostile activity thus initiating the process of relocation that can take place within 10
seconds. The SHM network established a robust method to deal with enemy attempts to jam
its radio signals. In any interference attempt, to disrupt mine‘s communication amongst the
mines, these could hop channels between acoustic and spread-spectrum signals [11].
.
25
1.3.3 Wireless sensor networks for industrial processes
In industrial settings, WSNs can be used to monitor facilities, to predict possible equipment
failures and restrict access to unauthorized personnel thus contributing in improving quality
and productivity and reducing operating costs.
A project known as Actuated Acoustic Sensor Networks for Industrial Processes (AASN4IP)
was concerned with developing technology to support the monitoring of liquid-based
industrial processes by immersed WSNs [14-17]. The project developed a demonstrator
system to monitor conditions within nuclear storage ponds. Nodes were mobile small-scale
autonomous vehicles, with a spherical hull, propulsion and control systems.
Nodes could communicate with each other and with the base stations outside the pond using
acoustic techniques. In the upper areas of the pond, communication between a node and a
base station was performed using single hop links. However, deep in the pond, it is believed
that multi-hop techniques would be required to propagate data up to the base station.
The communication system supported a localisation system that enabled nodes to establish
their position within the pond. This was an important aspect of the project, since readings
within the pond are of limited value unless the location of the sensors is known.
In a related work to AASN4IP, reported in [123], a localisation system was developed to
locate sensors inside a cylindrical vessel filled with 250 litres of water. Analysis of the
localisation methods, using data from acoustic transducers attached on sensor nodes, shown
that these could achieve accuracies up to ±5cm [123]. Of interest were the observations that
the number of the transmitting transducers, attached on each sensor, as well as their
orientation with respect to the receivers have significant impact on localisation [123]. To be
more specific it was observed that an increased number of transducers (8) with a specific
separation between them could alleviate issues arose from random orientations of the sensors
as these rotate inside the water thus improving localisation accuracies.
Research in [18, 19] concerned the deployment of WSNs inside storage biomass products
such as silage4, in order to support early identification of potentially hazardous situations.
4
Silage here refers to animal food made from a mixture of seeds (i.e. grain or corn) and chopped grass all
ensiled in airtight conditions within plastic wrapping for better preservation.
.
26
Specifically, the variation of temperature and humidity were monitored to identify the early
stages of decomposition.
A preliminary study in [18] examined the impact of biomass on antenna resonance frequency
and on transmission range. As in the GlacsWeb project, discussed in Section 1.3.1, the fact
that the nodes were to be buried in a medium that was not air was expected to affect the
resonance frequency of the antennas and their transmission range.
Tests were carried out using two 433MHz antenna, a homemade loop antenna and a
commercially available helix antenna both enclosed in a plastic box to prevent direct contact
with the silage. A vector network analyser (VNA) and a spectrum analyser (SA) were used to
measure the resonance frequency of the antennas and the transmission range. The
experimental setup is shown in figure 3.
Figure 3. Antenna characterisation inside biomass [18]
Results showed a shift in the resonance frequency for both the antennas, whilst inside silage.
However, ―detuning‖ was more intense for the helix antenna (424MHz in air, 345MHz in
grain) rather than for the loop (434MHz in air, 429MHz in grain) antenna. The reason for
this difference lies on the fact that the near field of the loop antenna is mostly magnetic thus
influenced by the permeability of the silage rather than electric as in the case of the helix
antenna that depends on the permittivity (dielectric constant) [18, 43]. As therefore, given the
permeability6 of silage is similar to that of air [18, 44] and that the dielectric constant
6
.
Permeability is the degree of magnetisation of a material
27
(permittivity) was about 35 [18], discrepancies in the resonance shifts between the two
antennas were expected.
Transmission range measurements were performed using the SA to measure RSS over
distance in the experimental setup seen in figure 3. The results are shown in figure 4.
Figure 4. RSS over distance inside silage [18]
Results in figure 4 demonstrate that the two helix antenna at 345 and 424 MHz can reach
longer transmission range up to 8m whilst the RSS for the loop antenna dropped below noise
floor of 105dBm at just 2m [18]. These results are a direct consequence of the
characterisation of the near field of the antennas discussed above. The loop antenna had most
of its signal reflected back to it rather than radiated thus its transmission range was
significantly reduced [18]. On the other hand, helix antenna had most of their signal radiated
and as such could reach longer transmission ranges.
Figure 4 revealed another interest observation. From the two helix antenna used, the 345MHz
antenna had much higher RSS than the 424MHz antenna over all distance measurements.
This was justified on the difference between wavelengths of the two antennas. The 345MHz
antenna have larger wavelength compared to the 424MHz one, as indicated in equation (1)
thus signals out of 345MHz antenna are susceptible to increased diffraction due to scattering
(size of material in silage much smaller than wavelength ~87cm thus signals can bend around
obstacles and travel further). As a consequence signals out of the 345MHz antenna have
better penetration inside the silage [18].

c
f
(1)
where c is speed of propagation, f is the frequency of the signal, and λ is the wavelength of
the propagated signal
.
28
As a future work, the authors of [18] proposed using broadband antennas instead of
narrowband ones in order to avoid antenna detuning and optimise signal propagation through
silage stacks.
Research in [19] examined the impact on the ability of WSNs to broadcast their signals whilst
inside a biomass as seen in figure 5. Sensor units (transceiver and batteries) were enclosed
(airtight sealed) within a spherical housing to protect them from pressures developing inside
stored silages and to prevent any moisture or methane gases from coming in contact with any
electrical part of the sensor. Nodes were buried within the silage stack at two depths, 25cm
and 50cm. The network had a star topology with nodes transmitting to gateway node located
outside the stack and the network
Fermented
grass
sensors
Air-tight cover
Gateway
Transceiver
Figure 5.WSNs embedded in a silage stack [18]
Tests in [19] shown that the sensors could operate for a period of fifty three days, following a
specific sleep-wake up approach, whilst transmitting at maximum power of 10mW. This was
particularly important given that one of the objectives was to use WSNs to remotely monitor
silage stacks, over a long periods, without the need to periodically extract samples from the
silage to observe their condition, a practise commonly followed up to date [18, 19]. The
maximum distance over which reliable communications, between any one sensor and the
gateway, could be achieved was 5m.
As future work, the authors of [19] proposed networking of the sensor nodes, i.e. transferring
data from one node to another and then to the gateway, to extend battery life given that
individual nodes will not need to transmit at high power levels to reach the gateway directly.
.
29
1.4 Localisation
A key aim in many WSN deployments is to enable nodes to be distributed in a casual manner
without the need for careful placement [8, 11-13]. This is motivated by a desire to overcome
difficulties imposed by the deployment area (war conflicts, toxic substances) as well as to
reduce deployment times and costs. However, in many applications, collecting data without
information on the position at which sensors operate, it is not beneficial. In these cases some
form of localisation procedure is needed. Since the work reported in this thesis is concerned
with localisation in WSNs, the topic is introduced here, and will be discussed in detail in later
chapters.
A widely used localisation method is the Global Positioning System (GPS) [1, 4, 20, 21].
GPS is a worldwide radio-navigation system formed from a constellation of twenty four
satellites and their ground stations. It uses the satellites as reference points to effectively
calculate the positions of ground nodes. However GPS is not always suitable for WSNs since
GPS receivers are expensive with high power consumption [1, 4, 20,21] and therefore not
suitable for use on numerous small and disposable nodes especially in cases were sensors are
static thus their position needs to be recorded only once. Further, GPS technology is not
suitable for all applications since it does not work well indoors or in the presence of obstacles
in the line of sight between the satellite and the receiver [1, 4, 20, 21].
The localisation problem, in WSNs, is typically formulated in terms of few nodes whose
positions are known, referred to in literature as anchors and a majority of unknown nodes
whose positions are to be determined by the localisation process. Anchors may obtain their
coordinates via manual configuration when placed in an area of concern in predefined
positions or by other means such as GPS.
Localisation in WSNs typically relies on network communication, and so is sometimes called
network-based localisation. It can be performed with or without the use of dedicated
hardware. Techniques that use dedicated hardware are range based [1, 22], those that do not
are usually, but not exclusively, range-free. Range-based techniques acquire distance
information by analysing the strength or time of arrival of received signals [1, 23]. On the
other hand, range-free localisation allows nodes to estimate their position using the
coordinates of known nodes transmitted to them [24, 25].
.
30
1.5 Thesis motivation and objectives
This research was closely allied to an EPSRC-funded project known as Wireless Sensor
Networks for Industrial Processes (WSN4IP) [26-28] at the University of Manchester and
shared its motivations. The goal of the WSN4IP project was to investigate the use of wireless
sensor networks, immersed within industrial processes, for the purpose of monitoring internal
conditions. The term ‗industrial processes‘ was interpreted in a broad sense and included
processes occurring in the chemical, pharmaceutical and in food industries. For the purposes
of simplicity, the focus of WSN4IP was restricted to processes involving particulate solids.
The follow-on AASN4IP project (see section 1.3.3) considered liquid-based processes.
The aim of immersing a WSN in a process is to increase the amount of data that is available
to operators about the internal state of the process so that, for example, undesirable conditions
can be identified (see section 1.3.3 and below). In other situations the information could be
used to reduce energy consumption, the usage of raw materials and the amount of waste
produced.
The WSN4IP project investigated issues such as node hardware, network communication and
localisation. An important part of the project involved the development of a representative
demonstrator system. The chosen application was the monitoring of grain stored in silos.
An ever increasing population dictates a constant and high quality food supply, where grain
and its products are ranked at the very top of demand [31]. Therefore there is a growing
interest in ensuring, not only the production, but also the storage of grain in a safe
environment [26-41]. Monitoring grain storage conditions or indeed the storage of any type
of crops is important in a number of aspects. Storage of crops in an enclosed space, over long
periods, can lead to a build up of heat and humidity ideal conditions for speeding up
decomposition and the formation of carcinogenic toxins. The fraction of such toxins existing
within a body of grain is governed by strict legislations. If the fraction is too high, the grain
cannot be used for human consumption or even for animal feeding given that carcinogenic
toxins can reach humans even with animal by-products such as milk [35].
A number of companies are specialized in providing equipment to observe grain conditions, in
particular temperature and humidity [32-34]. Nevertheless, up to date, the employed
.
31
equipment is wired and although reliable, it is costly and lacks flexibility. However it is
difficult to measure internals of granular flow with cabled sensors. To address issues from
cabled sensors, the WSN4IP project proposed using WSNs for insertion into a model grain
silo as illustrated in figure 6.
Figure 6. WSN to monitor grain silo [27]
Node hardware and software was developed within the project and ultimately experiments on
networking where carried out using an immersed network of ten wireless nodes. Details of the
WSN4IP nodes and the in-silo experiments can be found in Chapter 7.
As indicated above, localisation was an important part of the WSNIP project that aimed at
improving the usefulness of sensing data by determining the spatial variations of key
processes enclosed in confined spaces. Researchers working on the WSN4IP project proposed
an UWB localisation system a prototype which is discussed in Chapter 4. However the
performance of this system was not thoroughly evaluated and from the limited amount of
results that had been made available from the rest of the WSNIP team, it appeared that
significant room for improvement existed.
The main objective of this research was to build on the knowledge acquired from the
WSN4IP project, to implement improvements and develop new methods that will allow for
accurate network-based and UWB-based localisation systems for WSNs inside industrial
processes.
.
32
1.6 Related work
Research in the area of monitoring grain stored in silos, using wireless systems, has not
previously examined the issue of localisation. Instead research is primarily focused on the
effect of grain and of the silo building material on the propagated signals, in establishing
whether communication through grain is possible and at what distances. Various signal types
have been considered for propagation inside grain; those include acoustic, narrow band and
UWB signals. The use of narrow-band and acoustics signals will be discussed in the
following sections whilst the use of UWB technology will be discussed in more detail in
chapter 3.
1.6.1 Grain monitoring using acoustic signals
Sound waves have very low frequency (20Hz-20kHz) and can propagate through solids or
liquids for longer distances and with lower attenuation than in air [14-17]. For these reasons
sound waves are widely used in underwater communications [14-17] and have also been
investigated for possible applications in grain monitoring inside silos [36-38].
The use of acoustic signals was practically examined in [36, 38] with the aim of detecting
harmful insects penetrated the silo causing damage to the grain that might not be easily
observable by eye [36].
The authors of [36] identified that sounds produced by larvae and adult insects whilst in
motion or eating have unique time signature thus allowing identifying them. An experimental
investigation was carried out using a cylindrical tube filled with 1kg of grain and two
piezoelectric transducers immersed in grain. Part of the used grain was infested with larvae
and was placed at the middle of the distance in between the two transducers whilst adult
insects were thrown inside grain and left to move around freely.
The method of identifying the presence of insects and also filtering out noise signals caused
by random grain settling (and other sounds coming from outside the silo) involved of an offline phase and a real time phase. In the off-line phase, hundred sound samples were collected,
half from larvae and another half from adult insects and another hundred samples were taken
to characterise noise. In one of the real time measurements, 15 minutes of sound data were
recorded that include a mixture of all possible sound signals that can be detected by the
.
33
piezoelectric sensors. Once real time data had been recorded these were compared with the
data from the off-line phase in order to detect similarities in the patterns thus identifying the
presence of the insects.
Work in [37] presented a more thorough analysis of sound propagation inside grain than in
[36] with the aim of establish a reference starting point for future implementation of acoustic
detection of insects inside grain. An experimental procedure was carried using a 2.5 x 1.3 x
1.6 m container filled with about 5 m3 of wheat and by placing a single speaker and two
microphones arranged in a horizontal line 50cm below the surface of the grain. Initially the
authors examined whether sound waves actually propagate from grain kernel to grain kernel
or whether sound waves travel through the air gabs between grain kernels.
To determine the propagation path of the sound, experiments were carried out using a piezoelectric microphone that initially was situated inside grain and then moved outside 1mm
above the grain surface. In both cases, a strike on the outside surface of the tank at about
50cm from the piezo-electric microphones generated sound vibrations detected by both
microphones (inside and outside grain) with very small differences in between them. Based
on all observations the authors of [37] concluded that sound propagated through the air gaps
in between grain kernels. Thereafter, the authors of [37] investigated the effect of frequency
on the attenuation of the sound waves. Numerous tests at different frequencies revealed that
grain attenuates high frequencies thus it was concluded that the frequency of weak sound
waves, generated by insects at 1m from a receiving microphone, should lie within a range
lower than 1 kHz in order to be reliably detected [37].
Overall, experimental work presented in [36-38] revealed that it is possible to use acoustic
signals to monitor grain inside silo. Nevertheless, it was concluded that grain is a major
attenuator that severely limits the distances at which acoustic sensors could operate. Thus, the
use of acoustic sensors is not an efficient method to monitor grain since a large number of
acoustic sensors will be required to provide effective monitoring of small area of stored grain.
1.6.2 Grain monitoring using narrow band signals
Work presented in [39-41] pointed out that WSNs can be a valuable asset in grain
monitoring. Sensors deployed within grain silos can form ad-hoc networks thus allowing
.
34
delivering data over long distances, to a central receiver, hence coping with cases were
individual sensors will be out of range of the central receiver. To this extent, extensive
research was carried out in order to establish the maximum distance within which two
battery-operated nodes could communicate between each other.
Reference [39] introduced the idea of monitoring grain in silos using a buried WSN. It was
pointed out that multi-hop network would be necessary since individual sensors buried deep
within the silo would most likely to be out of range of a base station. [39] reported an effort
to establish the maximum distance over which two battery-operated nodes could
communicate with each other when immersed in grain.
Work on propagation of RF signals through grain [39, 42] suggested that attenuation of radio
frequencies in grain can be estimated by equation (2), [39], given ε"<< ε'
α=
8.686πε"
λ ε'
(2)
where α represents attenuation in dB per meter of path length, (dB/m),  is the free-space
wavelength for the transmitted frequency, (m) and ε' and ε" are the dielectric constant and
loss factors respectively.
Experiments reported in [39] revealed that the dielectric constants depend on the moisture
content of the grain and the frequency of the propagated signals. For a fixed quantity of grain
with constant moisture content, and propagated signals in the range 900MHz to 10GHz ε'
values of ε' between 3 to 2.5 and ε'' values from 0.3 to 0.25 were obtained[39, 42].
As indicated in section 1.3.1, the unlicensed ISM bands are often used in WSNs, in particular
the 868MHz and 2.4GHz bands. Based on equation (2), a significant amount of attenuation
should be expected in these bands. For example, at 900MHz equation (2) gives attenuation
between 11 to 15dBm/m with attenuation at 2.4GHz reaching 30 to 40dBm/m. In order to
take advantage of lower attenuation at lower frequencies, the operating frequency in [39] was
set around 900MHz.
To practically investigate the maximum distance over which two nodes can communicate
with each other, an experimental procedure was carried out in, a grain filled, steel container
with four commercially available WSN nodes (Mica motes) as seen in figure 7.
.
35
Figure 7. Attenuation measurements inside grain at 915MHz [39]
One of the nodes was used as a gateway mote placed at the top of the container with the other
three nodes being situated inside the grain at different depths. The operating frequency for the
motes was 915MHz transmitted at 1mW. According to [39], the node‘s batteries would
enable them to operate for up to four years, depending on usage.
Experimental data depicting the maximum and minimum attenuation values as well as the
average attenuation over distance are shown in figure 8.
(b)
(a)
Figure 8. Attenuation at 915MHz [39]
Figure 8a shows the maximum and minimum attenuation values, recorded at the same
distances, where it can be seen that these deviate considerably from the average values.
Deviations in attenuation were attributed to the multipath effect. Multipath effects arise when
.
36
signals follow multiple paths in their way from the transmitter to the receiver thus creating
multiple versions of the same signal. The main cause of multipath in [39] was considered to
be the metallic structure of the container, within which tests took place, which caused
increased reflections amongst the propagated signals.
Results in figure 8b show that signal attenuation is relatively constant in distances less than
50cm that lies within the near-field3 region of the 915MHz antennas (=33cm). In near-field
region, energy decays very rapidly with distance. On the other hand in the far field region4
(>50cm between transmitter and receiver), it was observed that signal attenuation increases
with distance (figure 8b). In addition it was observed that the maximum communication
distance between transmitter and receiver was about 2m [39].
Analysis of the results, using equation (2), reveals discrepancies between experimental and
predicted theoretical attenuation values. The attenuation at 915MHz based on equation (2) is
-15dBm/m thus considering the results in figure 8 and a baseline of -53dBm, the attenuation
at 1m will be –68 dBm. Nevertheless, the average experimental attenuation at 1m was
-73dBm [39]. These results highlighted the need to base any radio models relating distance
to RSS on experimental data rather than theoretical.
A more comprehensive investigation of grain monitoring using WSN is discussed in [40, 41].
The experimental setup consisted of three ZigBee based nodes and steel container filled with
grain. An overview of the experimental setup is shown in figure 9.
Stainless steel
vessel
180cm
RS-232
Grain surface
Sensor node
30cm
Figure 9. Grain tests experimental setup
3
4
.
Near-field region is defined as the area where the distance from the antenna is well below 1
Far-field region is the area where the distance from the antenna significantly exceeds 1
37
In order to establish the maximum distance over which two nodes can communicate, tests
were contacted to measure the received signal strength (RSS) at the receiving node at
different receiver-transmitter separation distances [40]. This was found to be about 1.15m,
with any further increase in the separation resulting in a rapid fall in RSS reaching noise floor
levels. The experimental results are graphically shown in figure 10.
As far as the lifetime of the ZigBee based nodes is concerned, experiments shown that these
can remain functional for more than a year, given the specific sleep and wake-up program
that was followed.
Figure 10. Experimental RSS over distance
Overall the WSN system in [40] shown to meet the requirements of low power consumption
and of reliability in collecting and broadcasting the collected data. As future work, the
authors identified the need for a better organisation of the network by keeping distances
among nodes as small as possible whilst investigating methods to extend the transmission
range and to optimise the routing of data from the sensor nodes to the control centre. These
aim at reducing energy consumption for individual nodes thus increasing the useful operating
time of the network. Further it was proposed that the container with grain could be housed in
a room equipped with an air-conditioning system that could adjust humidity and temperature
based on readings from the sensors inside grain.
1.7 Thesis layout
The rest of the thesis is organised as follows. The second chapter reviews basic ranging and
localisation methods widely used in the literature and a number of which is also used in the
localisation methods examined in the rest of the chapters.
.
38
The third Chapter introduces UWB technology giving emphasis on parameters of interest
such as signal processing, distance estimation and localisation as these were practically
examined in the literature.
Chapter 4 presents a prototype UWB-based localisation method, developed for the WSN4IP
project, and reviews the algorithm and the experimental procedures followed. Localisation
results were thoroughly examined and observations on the impact of received signals on
localisation accuracy were recorded.
Work in Chapter 5 builds on the knowledge acquired from Chapter 4 and proposes a number
of signal processing techniques contributing to more accurate position estimates as opposed
to the results in Chapter 4.
Chapter 6 introduces novel localisation methods deviating from the prototype localisation
method discussed in the last two chapters. These methods were based on analysis of
theoretical and experimental distance data that shown a repeating pattern of specific
observations. These patterns were materialised into localisation algorithms and evaluated
using the experimental data derived as discussed in Chapter 4.
Chapter 7 discusses network based localisation for sensors immersed inside a large scale silo.
There is detailed explanation of the experimental procedures and of the impact of grain on the
propagated signals. In addition the results from the implementation of two network-based
localisation algorithms using preliminary experimental data are presented.
An overall discussion of the work presented in this thesis as well as proposed future work is
presented in Chapter 8.
.
39
CHAPTER 2
BASIC RANGING AND LOCALISATION METHODS
2.1 Introduction
This chapter presents a survey of basic ranging and positioning methods widely discussed in
the literature related to WSNs localisation. A number of these methods have been utilised in
the practical work discussed in this thesis presented in chapter 4 onwards.
At the beginning of this chapter, a number of fundamental methods of determining distance
(known as ranging) and position are presented. Thereafter a number of localisation
algorithms are discussed, some of which utilised the basic principles discussed at the start of
this chapter.
2.2 Basic ranging techniques
In this section, the two ranging techniques most commonly used in WSNs are introduced.
The first technique is based on time measurements and includes Time of-Arrival (ToA) and
Time-Difference-of-Arrival (TDoA) approaches, whilst the other is based on RSS.
2.2.1 Time-based ranging
There are two basic time-based ranging techniques, ToA and TDoA [49-51]. The ToA
method gives range estimates based on a measure of the time it takes for a signal to arrive
from a transmitter at a receiver. In TDoA, a number of receiving antennas are considered to
allow for range estimates to be computed. To be more specific range estimates, in TDoA, are
derived by measuring the difference in the arrival time of the signals at a reference antenna to
the signals arrived at the rest of the antennas in the system under concern. ToA and TDoA
can be applied using different kinds of signals, such as RF, acoustic or ultrasound [49-51].
There are two basic techniques through which ToA ranging can be implemented. One
technique is based on time synchronisation between transmitter and receiver and is known
simply as the ToA technique. The other technique is known as Round-trip-time (RTT) and it
can give range estimates without the need for time synchronisation [15].
.
40
In the ToA technique, transmitting nodes broadcast their information along with the time at
which the signal was sent. Once a node receives the signal it can estimate its distance from
the transmitter based on the time taken for the signal to arrive [15]. The receiver calculates
ToA by subtracting the time of transmission from the time of receipt and then it calculates its
distance from the transmitter as follows:
d ij  ToA c
(3)
where, dij is the distance from transmitter i to receiver j (m), c is the propagation speed
(m/s).
The main disadvantage of ToA ranging is the need for accurate time synchronisation between
timers in the transmitter and the receiver. Further, depending on the medium within which
signals are propagated, multipath effects and noise can have significant impact on the
precision of ToA, and techniques to overcome such issues increase the complexity of ToA
ranging [38, 15, 8].
TDoA data are computed using the difference in ToA between a reference receiving
node/antenna and a number of non-receiving nodes/antenna [38, 41, 15, 33].
A common method used to derive TDoA is cross-correlation between received signals.
Cross-correlation is used to find the time required to shift received signals towards the signals
at the reference antenna in order to maximise the integral of the product of those two signals
[46-48]:
T
Cora.j τ    X a t   X j t  τ   dt
(4)
0
where Xa is the reference signal arrived at time t and Xj is the signal from subsequent
receiving antenna delayed by time η.
A characteristic example of cross-correlation of received signals with the signal at the
reference antenna is shown in figure 11. This shows the correlated signals from four antennas
located at 10, 20 and 30cm from the reference antenna, located at X=0cm.
.
41
Figure 11. Experimental cross-correlation [27]
TDoA data are converted into distance using an adaptation of equation (3). TDoA data
Δtij =(ti-tj) that is the difference of ToA at reference receiver i to the rest of receivers j in
equation (3) result in distance differences, Δd (di-dj) between a transmitter to receiver i and j
as follows:
Δt ij  c=Δdij
(5)
Equation (5) demonstrates that ranging based on TDoA does required knowledge of at least
one distance from the transmitter to the receivers and this can be seen as a drawback
compared to ToA based ranging. However there are techniques that remove this requirement
as will be discussed latter in Chapter 4. Overall, TDoA-based ranging is advantageous when
compared to ToA since with TDoA there is no need for clock synchronisation between the
transmitters and the receivers.
The Round-trip-time (RTT) is a technique that can overcome this synchronisation problem.
In RTT, the sender transmits a signal and a receiver echoes it back to the original sender
whilst receiver nodes record the time taken by a signal to travel from the transmitter to a
receiver and back again (RTT)[95]. The distance d ij between transmitter i and receiver j, for a
given RTT, T, is given as follows [95]:
dij  err  T  c
(6)
where err is an error factor attributed to errors in estimating RTT
.
42
2.2.2 Received signal strength (RSS)-based ranging
The RSS-based ranging is widely used for localisation in WSNs [52-58]. This approach is
popular because most radio transceivers used in WSN nodes enable measurements of RSS to
be obtained and hence no additional hardware is needed in order to compute range estimates.
However, it should be noted that such RSS measurements are typically prone to errors.
The RSS radio model employs a path loss equation that allows derivation of the distance
between a transmitter and a receiver based on knowledge of the transmitted power of the
signal and a measurement of the received power at the receiving node.
An empirical model generally used to characterise RSS as a function of distance in RF
channels is the log-normal shadowing model [33, 43, 44]. This model has been confirmed
through extensive experimental measurements by numerous researchers and it can provide
adequate representation of the variation of the received power with respect to the distance
both in indoor and outdoor situations [33, 44]. Shadowing here refers to the random
variations experienced in the received signal power at a given distance, caused by obstacles
in the path of the signals from the transmitter to the receiver. The model shows that for a
given transmission power, there is an exponential fall in the mean RSS with increasing
distance between the transmitter and the receiver. Shadowing effects are accounted for in the
model through a zero mean Gaussian term. The log-normal law is given by:
PRSS (dBm)=Pt (dBm)-PLd (dB)
(7)
In equation (7) the received power (PRSS) is equal to the transmitted power (Pt) minus the
pathloss power PLd, where pathloss is given by equation (8)
d 
PLd (dB)=PLd0 +10n p log   +Xσ
 d0 
(8)
where np is the path loss exponent measuring power attenuation relative to distance and Xζ is
a random variable with zero mean and ζ2 variance, in RSS measurements, due to log-normal
shadowing, PLd0 is the pathloss at a reference distance d0
.
43
Hence if the transmission power Pt is known and the log-normal parameters d0, PL(d0), np and
ζ for a particular environment as well as the PRSS are measured then equation (7) can be used
to obtain an estimate if the distance between a transmitter and a receiver.
The pathloss at a reference distance d0 (usually taken at 1m) can, in principle, be computed
by employing the free space path loss equation as follows:
PLd 0 (dB)=10log
 G G λ2 
Pt
=-10log  t 2r 2 
PRSS
  4π  d 0 
(9)
where Gt and Gr are the gains of the transmitting and receiving antennas respectively and λ is
the wavelength in meters of the propagated signal.
Figure 12 exemplifies the differences between theoretical and experimental RSS results.
Theoretical RSS were derived using the free space model and random Gaussian noise whilst
experimental RSS were recorded using ChipconCC1010 transceiver [57].
Figure 12.Theoretical and experimental RSS [57]
Figure 12 shows that although RSS decreases with an increase in distance (as expected) the
theoretical RSS values deviate substantially from the experimental RSS. This observation
strengthens the need to use experimental RSS values in order to derive an empirical model
that will allow deriving distance estimates out of RSS data.
Equations (7) and (8) can be arranged into equation (10) that is suitable for range estimation:
.
44
d  d 0 10 
where  
(10)
Pt  PRSS  PL d 0  X
10n p
As indicated above, the log-normal law is an empirical relationship and as such it is based on
experimental measurements of RSS. It has been found that parameters PL(d0), nd and ζ
depend on the environment in which the experimental measurements were taken. Values of
these parameters obtained in one environment are not typically applicable in another [56]. In
fact, it has been observed that especially in indoor spaces, it can be difficult to obtain a
consistent model relating distance to the signal strength because of increase amount of
multipath fading and shadowing effects [33, 42, 44]. Further, in many cases is difficult to
ensure efficient calibration of the transceiver modules used in low cost nodes. Calibration
deficiencies can introduce significant errors in RSS measurements [44]. In addition, antenna
orientation between a receiving and a transmitting node can have a significant impact on RSS
measurements [52].
2.3 Basic localisation techniques
Consider a sensor node whose coordinates are unknown. If the distances (ranges) from this
node to a set of reference points (anchors), with known coordinates, can be determined, then
the position of the unknown node can be found. In the simple case of an unknown node in
two-dimension (2D) scenario, distances from the unknown to three anchors are required. In a
three-dimension (3D) case, the distances from the unknown to four anchors are needed.
The basic procedure for performing position determination based on range estimates is
known as trilateration. A similar approach based on measures of the angles formed between
an unknown node and a set of anchors points is known as triangulation. This is less
commonly used in WSNs that trilateration and the more advanced technique of trilateration
known as multilateration and as such it is not considered in this research.
The basic principles of trilateration and multilateration localisation techniques are discussed
in the following two sections.
.
45
2.3.1 Trilateration
Trilateration in 2D employs the known locations of at least three anchors and the distances
between an unknown node to the anchors to compute the position of the unknown node [49,
51, 59, 60].
Figure 13, depicts an overview of the trilateration process with three anchors AN1(X1, Y1),
AN2(X2, Y2) and AN3 (X3, Y3) and an unknown node UN (X, Y). Here it is assumed that the
distances Ri between UN and ANi are known. This means that UN must lie on a circle
centred on ANi having a radius Ri. Hence the unknown node lies at the point of interception
of the three circled centred around the respective anchors as shown in figure 13.
AN3 (X3, Y3)
R3
UN(x,y)
R2
AN2 (X2, Y2)
R1
AN1 (X1, Y1)
Figure 13. Trilateration diagram
The relation between the coordinates of the circles to the coordinates of the unknown node, at
the point of intersection is given as follows [60]:
 X  X 2  Y  Y 2 
1
1

R 1  
  
2
2
R 2    X  X 2   Y  Y2  

R 3  
2
2
 X  X 3   Y  Y3  


(11)
The location of UN in figure 13 can be obtained from equation (11) when solving for X and
Y as follows [59]:
.
46
X 2(X1  X 3 ) 2(Y1  Y3 ) 
Y  2(X  X ) 2(Y  Y )
2
3
2
3 
  
1
X12  X 32  Y`12  Y32  R 32  R 12 


X 22  X 32  Y22  Y32  R 32  R 22 
(12)
Extending the analysis to 3D is straight forward. In this case, the unknown point is given by
the intersection of four spheres centred on the anchor‘s positions with radii equal to the
distance of each anchor from the unknown. An extended version of this analysis appears in
the next section.
2.3.2 Multilateration
When attempting to apply range-based localisation to real-world problems, difficulties often
arise because of errors in the range estimates and these can significantly impair the accuracy
of the position estimates. A technique known as multilateration can be applied to enhance
localisation accuracy [49, 59-61]. Multilateration requires that the distance between the
unknown and more than three anchors in 2D (3D) are known.
Figure 14, depicts a 3D multilateration process where spheres are centred around four
anchors AN1(X1,Y1,Z1), AN2(X2,Y2,Z2), AN3(X3,Y3,Z3), AN4(X4,Y4,Z4) with all the spheres
crossing each other at the position of an unknown node UN(X,Y,Z) (yellow star).
(a) Side view
(b) Top view
Figure 14. Spherical multilateration example
Let R1, R2, R3, R4 be the radii of the spheres, seen in figure 14, then the relation between the
coordinates of the anchors and the coordinates of the crossing point with respect to the four
radii is given as follows [60]:
.
47
2
2
2 

 R 1   X  X 1   Y  Y1   Z  Z1  
  


  

R n  
2
2
2






X

X

Y

Y

Z

Z
n
n
n


(13)
The location of the unknown in figure 14 can be deduced from equation (13) when solving
for X, Y and Z through a set of equations formed when subtracting Rn-1-Rn in order to
eliminate X or Z parameters accordingly. For example, subtraction of R2 from R1 will result
in the following:
R 2n 1  R 2n  X  X n 1 2  Y  Yn 1 2  Z  Z n 1 2  X  X n 2  Y  Yn 2  Z  Z n 2
(14)
Equation system (14) can be expressed in the form A  X  b where
X 2n  X12  Yn2  Y12  Z 2n  Z12  R 2n  R12

(Yn  Y1 )
( Z n  Z1 ) 
 (X n  X 1 )




A  2 





 b
 2
2
2
2
2
2
2
2 
(X n  X n 1 ) (Yn  Yn 1 ) ( Z n  Z n 1 )
X n  X n 1  Yn  Yn 1  Zn  Z n 1  R n  R n 1 
X 
X= Y 
 Z 
(15)
The system in equation (15) is over-determined, that is there are more equations than
unknowns, and it may be solved by least square estimation (LS) method as discussed in [49,
59-61]. In the above example, the LS method gives the location of the unknown node as
follows:

X  A AT  A

1
 AT  b
(16)
The LS method can reduce the error in the position estimate since it provides the best fit for
which the sum of the square of errors between an estimated and a real position attains a
minimum value [49, 59-61].
In general, multilateration is considered to be more accurate than trilateration but it involves a
higher computational overhead.
.
48
2.4 Well-known WSN localisation algorithms
In the previous section, some of the basic concepts of localisation were introduced. This
section contains a brief description of a number of well-known range-based and range-free
localisation algorithms.
2.4.1 Ad Hoc localisation system
The authors of the Ad Hoc Localisation System method (AHLoS) [59, 61] proposed a
distributed localisation technique where unknown nodes can find their position using a
limited fraction of anchor nodes. Localisation is based on range information however
extensive experiments were carried before deciding whether to use RSS range information
from radio frequency signals or ToA ranging data from ultrasound signals. The experiments
showed that RSS-based ranging was less accurate than ToA-based ranging as therefore only
the latter was used in the localisation process.
In the AHLoS algorithm, nodes that are within communication range of each other, referred
as neighbouring nodes, first estimate the distance among them using ToA. Then those
unknown nodes, that have established a direct communication link with at least three anchors,
can compute their position using the basic multilateration technique discussed in section 2.3.
In the AHLoS method this is known as atomic multilateration an example of which is shown
in figure 15.
Figure 15. Atomic multilateration
Once the unknown nodes, which were able to find their position, have done so, they become
anchors they were used in the localisation process to allow for other unknown nodes to find
.
49
their position. This ―chain reaction‖ is continued until all nodes in the network have
established their position. The ―chain reaction‖ is described in the literature as iterative
multilateration is depicted in figure 16 whereas the unknown node UN1 is in direct contact
with 4 anchor nodes AN1, AN2, AN3 and AN4 and therefore its position can be estimated
using atomic multilateration. On the other hand the unknown node UN2 is in contact with
only two anchors AN3 and AN4 and the unknown node UN1. After UN1 has been localised, it
can act as an anchor enabling UN2 to estimate its own position using iterative multilateration.
AN1
AN2
AN3
AN4
UN1
UN2
Figure 16. Iterative multilateration
Simulation results have shown that when 45 unknown nodes and 5 anchors are placed in a
square grid of side 15 meters, the estimated positions based on ultrasound signals can be
within 20cm of the actual positions [59, 61].
AHLoS is a localisation scheme that is prone to errors due to the fact that it employs
unknown nodes that become anchors through the localisation procedure. However the authors
of AHLoS claim that this error accumulation can be controlled and reduced due to the
accuracy of the ranging technique used which as indicated above is based on ToA with
ultrasound signals [59, 61].
2.4.2 DV-Hop algorithm
The DV-Hop algorithm is a range-free method that allows estimating the position of the
unknown nodes based on the number of hops between all nodes and the Euclidean distances
between anchors [54,62,120-122]. In DV-Hop localisation, anchor nodes broadcast their
locations through the network and compute the minimum number of hops along the routes
between each of the anchors. Then the Euclidean distance amongst the anchors is determined
.
50
and the average distances per hop for each of the routes among the anchors are evaluated [54,
62, 120-122].
The average hop-size (CF) between anchor i with coordinates (Xi,Yi) and anchor j (Xj,Yj) is
estimated by the anchors as follows:
 X
CFi 
j i
i
Xj
2  Yi  Y j 2
h
(17)
ij
j i
where hi is the number of hops between the anchors
The average hop-size, CFi, is then multiplied by the number of hops between the anchors and
unknowns to provide an estimate of the distances between the anchors and the unknowns.
These distances along with the coordinates of the anchors are then used within multilateration
to find the positions of the unknown nodes.
A simulation result from the implementation of the DV-Hop localisation algorithm is shown
figure 17. In this case the average hop size was estimated using Disjkstra algorithm that gives
the minimum hop count between anchors [120-122].
Figure 17. DV-Hop localisation
What can be seen in figure 17 and indeed in numerous other examples in the literature [120122] is that unknown nodes with the highest number of hops to the anchors result in position
estimates with the highest degree of error compare to those with lower number of hops. This
is justified by the fact that DV-Hop distances are only an approximation of the straight line
distances between anchors and unknowns and the higher the number of hops the higher the
.
51
built up error in DV-Hop distances resulting in significant degrading of the localisation
results[120-122].
2.4.3 DV-Distance algorithm
The DV-Distance algorithm is closely related to DV-Hope method but it uses the sum of
neighbour-to-neighbour distances in place of hop count to derive range information between
anchors and the unknowns [61, 62, 121-122].
The average DV-Distance size Dv, between anchor i with coordinates (Xi,Yi) and anchor j
(Xj,Yj) is estimated by the anchors as follows:
 X
Dv ij 
j i
i
Xj
2  Yi  Y j 2
d
(18)
ij
j i
where
d
j i
ij
is the sum of the Euclidean distances between all nodes on the route from one
anchor to another
Once anchors compute the DV-Distance size they broadcast this information to the unknown
nodes. This allows unknown nodes to use DV-Distance size and the coordinates of the
anchors within multilateration to find their positions. A simulated example of a DV-Distance
method is shown in figure 18.
Figure 18. Simulated DV-Distance localisation
Figure 18 shows similar trends as in the DV-Hop algorithm whereas; the further an unknown
is from an anchor, the highest the deviation of the position estimates from the true positions.
Analogous results were observed through simulations in [61, 62,120-122] with the authors
.
52
pointing out that DV-Distance localisation is highly susceptible to the accuracy of the
methods used to derive range estimates.
2.4.4 Iterative localisation algorithms
The authors of [55, 56] propose a method for node positioning based on Maximum
Likelihood (ML) estimation. Here the ML technique is used to allow for the computation of
position estimates given pair-wise distances amongst the nodes, which are derived from RSS
measurements.
The algorithm discussed in [55, 56] is based on distance estimates derived from an empirical
radio model derived through an extensive experimentation. Equation (19) next is a modified
version of the log-normal shadowing equation presented in section 2.2.2 and shows how the
RSS distance dˆi , j is derived:
P0 -Pi,j
d̂i,j =d 0 ×10
where
10n P
Xσ
10n P
(19)
=di,j×10
d0 is the reference distance in meters (1m), P0 is the received power in free space
(watts), Pi,j is the RSS power at i received from transmitter j (watts), np is the path loss
exponent, Xζ is a random variable with zero mean and ζ2 variance representing the standard
deviation in RSS measurements due to log-normal shadowing, Xζ=N(0,ζ2), d i,j is the
theoretical Euclidean distance between a transmit point j(Xt,Yt,Zt) and a receive point i
(Xj,Yj,Zj) given as follows:
d ij 
X t  X j 2  Yt  Yj 2  Z t  Z j 2
(19)
The ML relative location estimation ˆR x, y  for the RSS case is given by equation (20):
ˆ R  arg min
X
mn
 d̂ i, j
 ln
 d2
i, j
jH (i , j1) 
 
i 1




2
(20)
where m is the number of reference nodes, n is the number of the unknown nodes, H(i) is the
number of neighbouring nodes that device i detected
.
53
The minimum value of equation (20) is obtained using an iterative conjugate gradient
algorithm. Here, the algorithm takes the true positions of the nodes under concern from which
it derives the Euclidean distances and the distances derived through RSS measurements and
then through an iterative approach it returns the positions at which equation (18) reaches a
predetermined convergence point.
The localisation algorithm in [55, 56] was implemented in practise in a wireless sensor
network test-bed of 56 nodes deployed in an indoor office area 16 m by 14 m. Results
showed the average location error for all 40 unknown devices to be 2.1 meters. Of the 33
devices located inside the rooms, 22 were estimated to be within the correct room with the
other 11 nodes estimated to be either in the immediately neighbouring room or in the hallway
just outside the correct room. The maximum localisation error, in all cases, was found to be
4.2 m and the minimum error 0.12 m.
Another iterative minimisation algorithm is discussed in [12] although, in this case, position
estimates are derived based on distances out of TDoA data. First of all, TDoA data are
converted into distance differences, Δd, using equation (5).
The distance difference Δdij,( Δdij≈di-dj), can also be computed in theory using the Euclidean
distances between a transmitter position at (Xt, Yt, Zt) and the ith and jth receiving antenna by
applying Pythagoras theorem into Cartesian coordinates as follows:
Δdij =
 Xt -Xi  +  Yt -Yi  +  Zt -Zi 
2
2
2
-
 X -X  +  Y -Y  +  Z -Z 
2
t
j
2
t
j
t
j
2
(21)
Both the experimental and the theoretical distance differences profile values are then used
into an objective function:
N 1 N
FX, Y, Z  arg min   f ij2 X, Y, Z
X ,Y , Z
(22)
i 1 ji 1
where f ij2 X, Y, Z  (c  Δt ij  Δd ij ) 2
The values of Xi,j, Yi,j and Zi,j minimising the objective function in (22) are taken as the final
position of the node to be localised. To solve (22), a quasi-Newton algorithm was used with
.
54
initial estimates for Xt, Yt and Zt being the coordinates of the geometrical centre of the room
were the mobile unit is situated.
2.4.5 Centroid algorithm
Centroid algorithms [58, 63] are range-free and compute the coordinates of an unknown
node, UNi, by taking the centroid of the positions of two or more anchors with which the
unknown node can establish a direct communication link. The coordinates of an unknown
node are calculated via equation (23).
UNi (x,y)=
1 n
 AN j (x,y)
n j=1
(23)
where n is the number of anchors with which an unknown has direct link and AN j (x,y)
denotes the coordinates of the anchors.
A parameter of interest in the centroid algorithm is the connectivity metric CMi [4]. Here
each anchor periodically transmits a beacon advertising its position. Further, it is assumed
that unknown nodes know the periodicity of the beacons from each anchor and so the
unknowns can measure the number of beacons received in a fixed time interval [4, 37]. Each
unknown node can compute the connectivity metric CMi for each anchor node i over time
period t using the following:
CMi 
N recv (i, t)
100
N sent (i, t)
(24)
where Nrecv(i, t) is the number of beacons received from anchor i in time t and Nsent(i, t) is the
number of beacons sent by anchor i in time t.
Figure 19 demonstrates a simulation result, generated by the author using MATLAB using
five anchors, depicted with green circles, and three unknown nodes, depicted with yellow
circles. What can be seen in figure 19 is that unknown nodes with number 2 and 3 have links
to three or four anchors and as such their positions are given (red circles with respective
number) as the centroid of those anchors. On the other hand, unknown node 1 has a link only
to one anchor, anchor 2, and as such the position of this unknown node is given at the exact
position of anchor 2, based on equation (23) resulting in higher localisation errors.
.
55
Figure 19. Centroid localisation
The authors of [63] have conducted an experimental evaluation of the centroid algorithm in a
10mx10m outdoor car park using one unknown node and four anchors. The square parking
lot was subdivided into 100 1mx1m smaller grids and measurements were taken by placing
the unknown node at each of the 121 intersection points of the gridded area, whilst the four
anchor nodes were positioned at the four corners of the car park. In the above experimental
set up, anchors transmitted their coordinates every two seconds and all grid points had a link
to at least one anchor with a connectivity metric of 90% or more. The average localisation
error for all 121 points was 1.83m with the minimum error being 0m and the maximum
4.12m [63].
2.4.6 Weighted centroid localisation (WCL) algorithm
The WCL method [58, 64] is an enhancement to the basic centroid algorithm that provides a
more accurate approach to find the position of an unknown node, given the latter can
successfully communicate with two or more anchor nodes. The WCL method uses distances
derived from RSS or TDoA measurements, as weighting factors applied in the basic centroid
calculation.
The algorithm calculates the position of unknown nodes by averaging the weighted sum of
the coordinates of the anchors with which an unknown node has links to. The weighting
factors are related to the estimated distance dij between anchor j and unknown node i as
follows:
n
UN i X, Y  
w
ij
 AN j X, Y 
j1
(25)
n
w
ij
j1
.
56
1
, d ij is the estimated distance between anchor j and unknown node i
d ij
WCL was examined, by the author, using the same simulation scenario of anchors and
where w ij =
unknowns seen in figure 19. A comparison between the results seen in figure 19 and those in
figure 20 revealed that, as long an unknown node has links to more than one anchor, then the
WCL is more accurate than the simple centroid approach.
Figure 20. WCL localisation
The WCL was evaluated both with simulations and experimental measurements in an outdoor
space [58, 64]. In both cases four anchor nodes where placed at the corners of a 43mx43m
area that was subdivided into grids and measurements were taken between the anchors and
the unknowns inside this area. The average localisation error from all the grid points given by
the simulations was 2.6m whilst the experimental error was 5.3m.
The reason why the experimental localisation error was higher than the theoretical value lies
in the fact that the distance weights used in WCL were derived from RSS measurements that
include real time noise parameters (shadowing). On the other hand, in simulations, noise
parameters where introduced randomly and could not account for an accurate representation
of a real noise environment [58, 64].
2.4.7 Ecolocation algorithm
The Error Controlling Localisation (Ecolocation) method, like many other WSN localisation
methods uses RSS, from a number of anchors, at an unknown node to estimate the position of
the unknown node [124]. However, it does not employ the log normal law to associate RSS
with distance as discussed in Section 2.2.2. Instead, each node orders the RSS received from
.
57
in-range anchors and assumes that the anchor producing the highest RSS is nearest, that
producing the next highest RSS is second nearest etc.:
Ri>Rj=> di<dj
(27)
where Ri(j) is the RSS at the unknown node due to a transmission from anchor i(j), at distance
di(j) from the unknown. This ordering is used to compute an α×α measurement matrix M,
where α is the number of in-range anchors at the unknown:
= 1 if Ri<Rj
Mij = 0 if Ri=Rj
(28)
=-1 if Ri>Rj
The echolocation algorithm required that an unknown grid is superimposed upon the
environment in which the WSN is deployed. Since the coordinated of each grid point are
known, and the coordinates of each anchor are known, then the unknown can compute an
α×α constraint matrix C at each grid point:
=1 if di>dj
Cij = 0 if di=dj
(29)
=-1 if di<dj
i.e Cij‘s value is set dependent on the relative distance of the grid point with respect to
anchors i and j. This procedure results in a constraint matrix at every grid point. The
measurement matrix M is then compared, on an element-by-element basis, with each
constraint matrix C and the position on the grid with the highest number of matching
constraints is taken to be the position of the unknown. If several grid points have the same
number of matches, then the centroid of these points is taken to be the location of the
unknown [124].
Both the measurement and constraint matrices encode information about the distance of a
point (the unknown or grid point) to the set of anchors. The location of the unknown is then
chosen on the basis of the grid point whose anchor distance information matches its own
most closely [124]. Hence, although multipath/shadowing effects might cause individual
contradictions the inequality in equation (27) holds. This ‗majority view‘ potentially provides
the algorithm with a degree of robustness to multipath effects. The disadvantage of the
method is its computational cost. As the grid size is reduced to provide finer resolution, the
.
58
effort of computing the constraint matrices, and matching them against the measurement
matrix, increases rapidly [124].
A simulation example of the Ecolocation algorithm is shown in figure 21. In this case, nine
anchors A1 to A9 are placed on a grid 10×10 whilst the position of the unknown node
denoted by the letter P is situated outside the grid (figure 21).
Figure 21. Ecolocation position estimate [124]
In the scenario depicted in figure 21, it was assumed that RSS at the unknown node P, from
each of the anchors A1 to A9, were directly proportional to the distances of the anchors to
the unknown. As a result, unknown node P ranked the anchors, it can communicate with, as
follows: A1,A2,A3,A4,A5,A6,A7,A8,A9 and used this ranking to set the constraints in
equations (28) and (29). As expected, the location of the unknown node P was given by
Ecolocation at point E, indicated by a red star, in figure 21.
The accuracy of Ecolocation was evaluated using experimental data from MICA 2 notes that
were placed in indoors and outdoors spaces. Outdoor measurements took place in a car park
with eleven nodes randomly distributed over a 144 m2 area. At any one time one of the nodes
acted as unknown to record RSS, from the rest of the nodes, that were considered as anchors.
In this case, absolute localisation error, using Ecolocation, varied from 4m up to few cm. On
the other hand, indoor tests took place in an office building 120m2 using twelve MICA nodes
randomly distributed. Implementation of Ecolocation using data from indoors tests gave
errors of more than 1m [124]. Discrepancies in errors were attributed to the amount of
obstructions in the line of side between nodes (more obstructions in indoors space- higher
errors) and to the orientation of the antenna nodes with respect to each other [124].
.
59
Overall it was observed that Ecolocation can provide relative accurate position estimates as
long as fluctuations in RSS do not alter significantly most of the set constraints between
distances and RSS [124].
2.5 Summary
This chapter introduced basic localisation and ranging techniques and has described a number
of localisation algorithms that use range and range-free localisation approaches.
A thorough review of both of ranging and range-based localisation revealed that in all cases
the quality of the received signals determine the accuracy of the results. To start with, if
signals are severely affected by noise this will cause discrepancies in computation of range
estimates. Ranging based on RSS can be highly erroneous especially if measurements are
taken in indoors areas were multipath is more intense and if the nodes assumed orientations
that have negative impact on signal radiation. On the other hand ranging based on ToA and
especially TDoA gives more accurate range estimates since times are not affected by major
fluctuations as in the case of signals‘ amplitude. In the case of TDoA, the use of multiple
receive antennas adds diversity to the receive signals thus improving the probability to
receive signals least affected by noise.
Of particular interest is the observation that as the number of anchors, an unknown node can
communicate with, increases then localisation accuracy increases as well. This is an
important observation that can provide a straight forward solution in cases were unknown
nodes might be far from anchors i.e increase the number of anchors to extend network
coverage thus increase accuracy.
The following chapters present a number of ranging and localisation methods that implement
methods discussed in this chapter.
.
60
CHAPTER 3
UWB TECHNOLOGY FOR RANGING AND LOCALISATION
3.1 Introduction
To a large extent, research reported in this thesis is concerned with UWB-based techniques
for ranging and localisation. This chapter provides relevant background, specifically a review
of UWB technology and its application for ranging and localisation purposes.
The chapter begins with an overview of UWB wireless technology and how this compares
with other forms of wireless systems. This is followed by a brief consideration of the effect
that UWB antennas have on the propagated signals, both at the transmitting and receiving end
as this is relevant to some aspects of the research reported in this thesis. A literature review of
UWB-based ranging and localisation is then presented.
3.2 UWB technology
UWB systems propagate trains of ultra-short pulses (monocycles) that have widths ranging
between 0.5ns to 1.6ns and pulse-to-pulse intervals between 25ns and 1000ns; thus the duty
cycles of UWB signals are very low [99]. A typical 0.5ns single monocycle is shown in
figure 22 whilst a pulse train with 25ns pulse intervals is shown in figure 23. Note that figures
22 and 23 depict ideal pulses, generated in MATLAB, without considering noise.
Figure 22. 0.5ns monocycle
.
Figure 23.Train of 0.5ns pulses
61
The mathematical representation of the monocycle seen in figure 23 is the first derivative of
the Gaussian pulse that given as follows [94]:
t
2
 2 t   
V ( t )   2 e   
 
(30)
were η is the pulse width and t is the period of the pulse
UWB pulses lie within a spectrum between 3.1GHz and 10.6GHz or occupy an absolute
bandwidth (BW) that is greater than 500MHz or have a fractional (BWfrac) greater than 20%
[80-82, 90]. BWfrac, is given as follows:
BWfrac =
2×  BW  2×  f H -f L 
=
f H +f L
f H +f L
(31)
where fH is the highest frequency component in the signal and fL is the lowest frequency
component measured from the -10dB point (figure 24)
The frequency spectrum of a typical 0.5ns UWB pulse is shown in figure 24 where the BW
exceeds 500MHz.
-10dB point
Figure 24. Wide-band time spectrum
The large BW of UWB signals is advantageous compare with Narrow Band (NB) signals
since according to Shannon‘s theorem the higher the BW the higher the channel capacity,
thus larger amounts of data can be transmitted [98, 100]. Based on Shannon‘s theorem UWB
systems are suitable for multimedia and video surveillance applications over short distances
[70-72, 98].
.
62
The complexity and the cost of UWB systems is reduced by the fact that data can be
transmitted directly over UWB signals without the need for an extra hardware to perform
modulation at higher frequencies as in NB systems. Data modulation in UWB systems is
typically carried out using pulse position modulation (PPM) [65]. A further advantage of
UWB radio is that no other communication system operates in the same frequency band thus
minimising interference. This gives rise to a wide range of UWB applications in areas such as
multimedia entertainment and the use of UWB-based sensor networks for maintenance inside
aeroplanes as well as the use of wireless UWB-based sensors for patient monitoring inside
hospitals [71-72, 96].
The lack of interference between UWB and NB systems is emphasised in figure 26. This
shows the power spectral density (PSD) over frequency for UWB and for five common NB
systems, GPS, global system of mobile (GSM) phone network, Wi-Fi, wireless local area
network (WLAN) and ZigBee. What can be seen in figure 25 is that, other than the fact that
UWB systems operate at a different frequency band than the NB systems, the PSD of UWB
is very low that is seen as noise floor to NB communication thus further reducing the
possibility for interferences from UWB [80, 83].
PSD
GPS
GSM
Wi-Fi/WLAN/
ZigBee
-41.3 dBm/MHz
UWB spectrum
1.6GHz
2.1GHz 2.4GHz
3.1GHz
10.6GHz
Frequency (GHz)
Figure 25.Narrow-band and UWB spectrum coverage (adapted from [83])
The short duration of the UWB pulses improves their endurance within a dense multipath
environment since, in contrast to narrow-band signals, UWB pulses do not overlap with each
other [65, 66, 69, 80]. The reception of UWB signals without any overlapping amongst the
signals makes it easier to resolve the time of arrival of the leading edges in a method that will
be discussed in further detail in chapters 4 and 5.
.
63
The properties of UWB transmissions discussed above make it attractive for use in WSNs. To
this extent, the Institute of Electronics and Electrical Engineering (IEEE) developed the IEEE
802.15.4a standard for low rate wireless personal area networks (LR-WPANs) [79-81]. The
objective of the IEEE 802.15.4a standard was to ensure robust data communication over long
distances as well as to provide enhanced positioning capabilities all with minimum
interferences to and from existing NB-systems [79-81].
3.3 UWB antenna and their impact on ranging and localisation
Much of the work in this thesis is concerned with a UWB-based positioning system. A key
aspect of this, and other similar systems, is the detection of the leading edge (LE) of UWB
pulses as they arrive at the receiving antennas. The LEs are considered to be the least affected
by multipath and as such are widely used to allow for accurate ranging and localisation. A
review of UWB systems revealed that the antennas employed have a significant effect on the
propagated UWB pulses, and these effects need to be discussed in order to understand the
form of the received signal whose LE is sought in ranging and localisation applications [8494].
Antennas radiate electromagnetic energy when presented with current at their input
(transmitting) or convert electromagnetic energy to current when used for receiving [89, 94].
A magnetic field (B) is induced by the antenna at point P whilst current i(t) flows through it,
as given by the Biot-Savart law [30]:

dB
 0 it dL sin 
4R 2
(32)
where µ0= 4π×10−7 H/m and denotes the permeability of free space (magnetic constant), R is
a distance of point P from current source, θ is the angle between the point and the antenna
element, B is the magnetic field and dL is a differential element of the antenna of length L as
seen in figure 26.
.
64
Figure 26. Monopole antenna magnetic field [94]
If the current through the antenna varies with time, then the magnetic field varies with time,
giving rise to an electric field via Faraday‘s law. The electric field due to equation (32) is
given as follows [94]:
 L R  L cos  
di t  

c
c


sin    0  dL

dt
dE 
4R 2
(33)
where L is the length of the antenna (seen in figure 26) and c is the speed of propagation of
electrons through the space (assumed same speed of propagation through antenna).
The total generated electric field is directly related to the dimensions and the shape of the
antenna and as such the total electric field induced in a monopole antenna is [94]:

E
0
4R 2
sin 
L
d   L R  L cos   
i t 
 dL 
dt   c
c
 
0


(34)
Equation (34) is representative of an immediate area surrounding the antenna within the near
field. In far field, the radiated electric field is taken as [94]:

E
μ 0c
4πR
2

sin θ  
L R  L cos θ  
R 
  i t  
i t  
cos θ  1  
c
c
c 
 
(35)
What can be seen in equation (35) is that the electric field is proportional to the time
derivative of the current flowing through the antenna. The impact of differentiation in a
narrowband antenna, using continuous signals, such as sine-waves, is limited to phase shifts.
.
65
On the other hand, differentiation of UWB pulses results in significant change in pulse shape
one or more times depending on the types and number of antenna used [66, 84-94].
An example of the impact of the transmitting antenna on UWB signals is shown in figures 27
and 28 (depicting test cases within an anechoic chamber) [76]. Figure 27 shows a Gaussian
pulse transmitted using a cable directly to the oscilloscope without any intermediate antenna
being used. On the other hand, figure 28 shows the received pulse but this time transmitted
wirelessly using antennas.
Figure 27.Cable transmitted UWB pulse [76]
Figure 28.Wireless transmission [76]
The pulse shown in figure 28 is a result of double differentiation of the Gaussian pulse
depicted in figure 27 thus verifying that antennas alter the shape of the propagated pulses.
A direct consequence of differentiation of the signals by the antennas is the loss of reciprocity
between receiving and transmitting antenna especially in the time domain [76, 84-94].
Another issue of concern, due to the alteration in the shape of the signals, is the detection of
the LEs of the received pulses. In cases where a threshold might be used to detect the LEs
(discussed in more detail later in this section and in Chapter 5) the process of deriving the
threshold should carefully considered alterations in the shape and especially in the time of
arrival of the received pulses. An effective threshold and an accurate representation of the
LEs are important parameters that can determine localisation accuracy [128, 104, 107].
The fact that propagated UWB pulses undergo significant change (due to the antenna) and
lose their original shape is not the only issue of concern. Another factor affecting UWB
pulses is the ringing effect that again is attributed to the antennas [90-92]. A major cause of
ringing is the fact that pulses propagating through the antenna, from the feeding point to the
.
66
end of the antenna, are reflected within the antenna numerous times. As a consequence,
signals travel back and forward to the source leading to the formation of standing waves that
in turn causes antenna to radiate for a time period exceeding the period of the original pulse
fed into the antenna. Examples of ringing, from literature, are shown in figure 29.
Figure 29. UWB antenna ringing effect [26, 92]
Mismatches between the antenna and the transmission lines, mechanical disparities in the
construction of the antenna, its dimensions, especially with respect to the ground plane, can
further strengthen ringing effects [90-92].
3.3.1 Antenna orientation
Research on range estimation and on range-based localisation showed that multipath is a
major cause of errors in both cases. In addition research revealed that the orientation of the
receiving antenna with respect to the transmitter‘s antenna can have a considerable impact on
the accuracy of ranging and localisation [73-75].
Work presented in [73] used one transmitting and one receiving UWB antenna at different
angles to examine the effect of antenna orientation on the propagated UWB pulses. A relative
orientation of 0o along the azimuth and elevation plane, with the two antennas facing each
other, was used as a reference to compare pulses resulting from other angles of orientation.
The degree to which received pulses, at a reference orientation of 0o differ from the any other
orientation was examined using cross-correlation. Cross-correlation between the pulse at 0o
and received pulses at orientation 90oin both the elevation and the azimuth plane is depicted
in figure 30.
.
67
Cross-correlation plots, in figure 30, indicate that the received pulses were distorted
according to the relative orientation between the receiving and the transmitting antenna.
Figure 30. Orientation effect on UWB pulses [73]
Analysis presented in this section highlights the need for orientation issues to be examined
carefully when setting up ranging and localisation systems.
3.4 UWB-based ranging and localisation review
UWB technologies have been used for ranging and localisation in areas such as asset
monitoring onboard ships, vehicle sensing, clinical and military applications and the
monitoring of industrial processes inside vessels, with reported accuracies reaching 5cm or
less [65,67,71,76, 96].
Although there is a plethora of research on UWB technologies analysis of relevant systems is
primarily carried out through simulations. This section presents a number of UWB-based
ranging and localisation methods that were implemented and evaluated through experiments
and present aspects closely related to the research discussed in this thesis.
3.4.1 Basic features of UWB-based ranging and localisation applications
The majority of UWB-based ranging and localisation methods employ a similar set of steps.
These include the detection of the LEs, the derivation of ToA or TDoA and the actual ranging
and localisation approaches. In most scenarios ToA and TDoA data are derived through postprocessing of received signals (i.e cross-correlation) by an algorithm running in MATLAB.
.
68
The use of the LEs, to derive ToA and TDoA, has been proven to enhance ranging and
localisation accuracies; since these represent a portion of the received signals that arrive first
at the receiving antenna and that are least affected by noise. Consider for example the
simulated received UWB signal in free space conditions shown in figure 31, the peak of this
signal denotes the LE of the signal and is clearly identifiable. On the other hand, figure 32
shows a received signal (captured using a high-speed sampling oscilloscope), in an indoors
space, using a receiving antenna situated at 12.8m from a transmitter.
Figure 31.Received signal in free space [67]
Figure 32.Indoors received signal [67]
Figure 32 shows that the experimentally received signal is significantly distorted by
multipath, thus making the identification of the LE much more difficult.
The following applications follow similar approaches towards implementing ranging or
localisation (use ToA or TDoA) with differences concentrated on the methods used to derive
the LEs of the received signals.
3.4.2 UWB-based ranging and localisation applications
The authors of [65] developed a prototype UWB system, based on TDoA, for potential use in
navigation sensors attached to vehicles. The experimental system is shown in figure 33.
The antennas used in the tests were modifications of the Vivaldi antenna (figure 33) in terms
of having two extended circular apertures seen in figure 33, rather than long straight radiation
apertures [65] as in conventional Vivaldi antennas. The authors of [65] claimed that the
modified antennas were more efficient since they were able to reduce ringing (discussed in
section 3.3) given the reduction in the antenna‘s area reduced the amount of internal
reflections.
.
69
Vivaldi antenna
Figure 33. Block diagram of UWB ranging tests [65]
The transmitted data was an 8-bit binary pseudo-random code generated by a field
programmable gate array (FPGA). The pseudo-random code was generated at specific clock
cycles and was directed via wire links to both the DSO and to the pulse generator [65]. As
such the pulse generator was triggered to generate 3ns pulses at the times at which the
pseudo-random code was present at its input. The DSO was triggered at specific time
intervals to allow for a sampling window that enabled incoming received signals to be
captured. Received pulses were saved in the oscilloscope for further processing using
MATLAB [65].
Measurements were taken inside an anechoic chamber using one transmitting and one
receiving antenna placed at different distances one from the other with one of these (1m)
being selected as a reference [65]. Signals received at 1m reference distance were correlated
with signals obtained at other separations to obtain TDoA data as discussed in section 2.2.1.
Both correlation and range estimation were carried out using dedicated algorithm running in
MATLAB. At this stage, it is important to note that that since tests were carried out inside an
anechoic chamber there was no significant noise affecting the received signals and as such
there was no further signal processing, instead the complete raw received signals were passed
into the correlation procedure [65].
Measurements were taken at five non-reference distances between the transmitter and the
receiver with resulting in an average error of 0.12cm and STD of 0.251cm [65]. Overall, it
has to be noted that the employed ranging method served as a proof of concept only as it was
.
70
evaluated solely on tests inside an anechoic chamber thus overlooking noise that can exist in
real world applications that can affect range accuracies.
Research presented in [66] used UWB technology to develop a positioning system to find the
locations of soldiers both inside buildings and outdoors. An experimental setup was
established where a number of fixed anchor beacon (anchor) transceivers and one rover
(unknown) transceiver as seen in figure 34.
Figure 34. UWB positioning system [66]
Each transceiver consisted of a low power pulse generator for signal transmission and
analogue components (low noise amplifier (LNA)) and a digital circuit to detect the LEs of
the received signals [66].
This digital circuit was so configured, using a digital phase lock loop, to allow a pulse to go
through it only for a time period equal to the duration (width) of the propagated pulse [66].
Further, the LE detection circuit periodically checked the noise level of the received signals,
through a constant false alarm rate (CFAR) loop, and set a threshold above which the
detection circuit was triggered ON thus reducing false leading edge detection [66].
The unknown node broadcasted a sequence of UWB pulses that included a short message
containing its ID to the anchors. Anchors that have received a signal from the unknown node
transmitted a message back to it but at different time intervals (labelled as time offsets Δi),
pre-assigned to them, in order to avoid signal collisions [66]. In turn, the unknown node
performed subtraction between the time offset Δi of the anchor to the time Δt taken for signal
.
71
to travel from the unknown to the anchor and back. The outcome of the subtraction in essence
represented the TDoA that when multiplied by speed of light gave distance differences that
used within a multilateration algorithm to derive the positions of the unknown nodes [66].
Tests with the above system were carried out in both indoor and outdoor environments. In the
experimental procedure, the unknown node was placed at eleven different positions around
the test area and data were recorded for about 10 seconds and each position. The positioning
error in the indoor tests was between 15cm and 90cm whilst for outdoor tests the error
remained below 15cm [66].
A prototype UWB-based positioning system, known as the precision asset location system
(PAL), was developed in [67] to determine the position of items of cargo on-board ships.
Tests were performed over a period of two weeks on a cargo hold (24.4m width, 30.5m
length and 8.23m height). Test objectives included evaluating the capability of UWB signals
to propagate through dense multipath environment of the ship and to what degree the
accuracy of UWB-based localisation was affected in partial LOS blockage in the presence of
containers.
The experimental setup consisted of four passive UWB receiving modules, placed at the
edges of the testing area, and a number of tags, powered by batteries, that transmitted UWB
pulses at 0.25mW. The tags were situated at different locations in the testing area and their
positions were measured using a tape measure and a laser surveying system that could
achieve 2 to 4mm accuracy. The experimental architecture of the PAL system is shown in
figure 35.
The receiving modules were equipped with digital circuit that allowed detecting the ToA of
the LEs. The ToA data were then forwarded to a central PC via a wired link (figure 35) were
an algorithm running in MATLAB was used to compute TDoA. Once all the TDoA data had
been calculated, the positions of the tags were computed using an iterative minimisation
algorithm discussed in section 2.4.5.
.
72
Figure 35. PAL localisation system [67]
Measurements revealed that delays spreads, that characterise multipath, were of the order of
3µs in contrast to the typical 300ns delay spread recorded in office and industrial premises.
Large delay spreads close to the LE make the detection of LEs more difficult thus
jeopardising positioning accuracies [67]. The relative large delay spreads were caused by the
presence of metal surfaces and objects in the vicinity of the measurement equipment [67].
The positioning accuracy in the absence of containers was between 0.9m to 1.5m whilst
positioning accuracy in NLOS tests, in the presence of containers, was between 3.35m and
3.66m [67].
Work in [76] presented positioning tests inside a typical electronics laboratory. This work
attempted to investigate the accuracy of positioning using a low-cost transmitter. The
experimental setup consisted of a PC, a triggering unit and a mobile unit representing the
unknown node whose position was to be found. The mobile unit transmitted UWB pulses to
four receiving antenna attached on a DSO as shown in figure 36.
Receiving
antenna
Transmit
antenna
Receiving antennas
Transmitting
antenna
Figure 36. Low cost UWB positioning experimental setup [76]
.
73
The PC initiated the tests by sending a signal to the triggering unit via a wired link (figure
36). In response, the triggering unit transmitted an activation code to the mobile unit. This
unit was equipped with a pulse generator which broadcasted a UWB signal in response to the
activation code. To ensure an effective synchronisation between the transmitter and the
receiver, the triggering unit sent a signal to the DSO (see figure 36) that initiated capturing
receive signals from the mobile unit over a fixed interval (sampling window).
Measurements were taken at a number of points by placing the mobile unit at different
distances (at intervals of 5cm over a 3m) from the receiving antenna, whilst always
maintaining line of sight (LOS) between the receiving antennas and the mobile unit.
Digitised received signals were transferred from the DSO to a PC where a MATLAB based
algorithm performed cross-correlation amongst all received signals from each antenna to
derive TDoA. An iterative minimisation algorithm, discussed in section 2.4.5, used the TDoA
data to compute the position of the mobile unit. The positioning error for all measurements
was around 58cm with standard deviation reaching 39cm [76].
The authors of [76] characterised their work as a low cost approach, however, they
recognised that the use of the DSO can easily label their title as misleading; thus justifying
the low cost on the basis that UWB pulse generators could be assembled using cheap
components [76]. Regardless whether the experimental setup was cheap or not, tests both
inside and outside the anechoic chamber offered a very good understanding of the efficiency
of the whole positioning system. More specifically it was observed that antennas acted as
differentiators to the propagated signals; a fact that needs to be carefully considered when
processing received pulses [76]. Furthermore, experiments carried out in an electronics
laboratory showed that the propagated pulses were distorted by both multipath and by
inconsistencies in the generation of the pulses due to a clock drift. The clock of the
microcontroller controlled the repetition frequency of the transmitted pulses thus any drifting
consequently resulted in jitter in generated pulses (~ns) thus causing inaccuracies in TDoA
estimates [76].
Work presented in [101] also investigated the impact of LOS and NLOS conditions on UWBbased ranging in real outdoor and indoor scenarios. Experiments were carried out using two
prototype UWB radio modules with 500MHz BW and 4GHz centre frequency that were
.
74
placed on the floor of the test areas. In all cases, one UWB transceiver was kept at a fixed
origin whilst the second transceiver was placed at various distances from the origin. The
distances between the two transceivers were measured with a tape.
Each radio module was connected, via a cable, to a PC that allowed processing digitised
received signals. An algorithm, running in MATLAB, detected the ToA of the LEs by
recording the time at which the sample with the highest amplitude occurred. This was done
for both the transmitting and the receiving sides thus providing estimates of the transmitting
and receiving times of the pulses. These time estimates were then introduced within an RTT
method (see section 2.2.1) to compute the distances amongst the two modules.
LOS tests were conducted inside a library and in an indoors hallway with doors and walls on
the both sides of the radio modules as well as with WiFi transmit points. In addition LOS
tests took place in an open field far away from buildings [101]. Further, to investigate office
like environments, ―LOS‖ tests were conducted with items having low attenuation such as
glass, chairs and doors‘ obstructing a 3m LOS path in what was called soft-LOS condition
[101]. Tests were performed both indoors and outdoors with the two radios placed at 1m up
to 45m distances from each other. In all cases distances were measured using a tape. Figure
37 shows the different test scenarios examined in [101].
Figure 37. Ranging tests at various scenarios [101]
The average ranging error for indoors LOS tests was less than ±20cm whilst the errors for
LOS tests in an open outdoors area was less than 10cm. According to the authors of [101],
there were fewer obstacles around outdoors tests thus fewer reflections in the propagated
signals and as a result accuracies for outdoors tests were better than those for indoors tests.
.
75
Note that ―Soft‖ LOS experiments using glass obstacles resulted in an average ranging error
of 6cm with a single door obstructing the LOS giving the highest error of 29cm [101].
Tests carried out with concrete walls in between the two radio modules were characterised as
hard-NLOS. Average ranging error, in hard-NLOS conditions, fluctuated greatly based on the
number of interleave walls [101]. The average error with only one wall was obstructing a 4m
LOS path between the two radio modules was 26cm with the average error increase to 87cm
when four walls obstructed a 14m LOS path. Of interest was a test case with eleven walls
obstructing a 38m LOS between the radio modules where ranging failed to work since no
signals could be received [101].
The authors of [102, 103] developed a prototype 3D UWB localisation system, based on
TDoA, in anticipation that this could be used to track lunar/Mars rovers and astronauts when
satellite navigation systems won‘t be available.
The experimental setup consisted of two commercially available UWB radio modules with
3.2GHz BW and 4.7GHz centre frequency. One of the radio modules was used as a receiver
to which five antennas were attached to using a power combiner. The so called ―One
Receiver-Five-Antenna‖ configuration aimed at eliminating synchronisation issues that might
occur when using numerous receivers together in a TDoA scheme. The receiving antennas
were fixed at different heights in an outside yard in 6.1m radius circle formation centred on a
reference antenna. This formation allowed 3D measurements along X, Y and Z axis. The
omni-directional antennas were connected to the power combiner using long low-loss cables.
Readings from the receiver module were recorded to a laptop using a MATLAB based GUI
[38, 39]. The other UWB radio module was used as a transmitted and was placed on a remote
controlled vehicle that allowed taking measurements at different locations [102].
Initial tests aimed at deriving TDoA data. To this extent, the LEs of signals at all received
antenna were detected using an algorithm that allowed processing all the digitised signals and
indentify the absolute peaks that denote the presence of the LE [102]. The LE of the reference
received signal (recorded at a reference distance between the transmitter and the receiver)
was then cross-correlated to each of the LEs of the signals from each of the five receiving
antenna to derive TDoA data as discussed in section 2.2.1.
.
76
At the end, TDoA data were used in a least square minimisation algorithm, running on a
laptop, to derive the X, Y and Z coordinates of the transmitter [38]. Experimental
investigation of the localisation method in [38], shown that it was possible, to track a moving
vehicle, with an UWB transmitter on it, in real time with an accuracy of less than 5cm [102].
A similar work to [102] is discussed in [104-106] where once more a pair of PulsON 210
modules was used to investigate a 2D TDoA based localisation system. Experiments were
conducted in a sport stadium (10m x 10m) using four static receiving antennas, situated at the
edges of the monitoring area. The antennas were connected, via a power combiner, to one of
the PulsON 210 modules that in turn was connected to a laptop. The other PulsON 210
module was used as a transmitter whose position was to be found and was placed at different
positions around the centre of the four receiving antenna. Both the receiving antenna and the
transmitting module were around 1.5 high from the ground.
TDoA data were acquired by subtracting the ToA for all received signals from the ToA of a
signal recorded at a reference transmitter-receiver distance [102]. Work in [102] proposed
recording the ToA of LE to derive TDoA. As such, two LE detection algorithms have been
proposed in [104], the maximum energy selection with search back (MESSB) and the cell
(sample) averaging constant false alarm rate (CA-CFAR) algorithms. These algorithms
allowed processing all the samples in the received signals, first to derive a threshold and then
to use this threshold to detect the LEs. Note that each algorithm employed different
approaches in setting an adaptive threshold based on specific empirical observations. A
diagram illustrating the basic principle on which these algorithms are based is shown in
figure 38.
Max
Sample
Min
Sample
Figure 38. LE detection algorithms [104]
.
77
The CA-CFAR algorithm sets the threshold based on the average of the samples, occurring
within a specific time frame, representing the portion of the signal within the LEs most
probably lies. This time frame was decided empirically through a thorough analysis of
numerous received signals. The time at which the first sample, in the received signal,
exceeded the threshold was recorded as the ToA of the LE.
On the other hand, the MESSB algorithm employed search-back window to look for the LE.
This algorithm implements the search-back window within the same time frame used for the
CA-CFAR algorithm. The boundaries of the search-back window are defined by the time at
which the sample with the maximum amplitude occurs and by the time at which the sample
with the minimum amplitude occurs. The threshold is then set empirically based on the
maximum and minimum value of the samples within the search-back window. Once again the
ToA of the first sample exceeded this threshold was considered as the LE of the received
signals and used to compute TDoA.
Localisation accuracies, using LOS-based experimental data, revealed that both the CACFAR and the MESSB algorithms resulted in almost identical positioning errors varying
between 5cm and 19cm. On the other hand, processing of NLOS data, using the CA-CFAR
algorithm resulted in better accuracies compared to the MESBB algorithm. According to the
authors of [104] the CA-CFAR algorithm contributed in more accurate results because it
employed a threshold based on the average of a specific part of the received signal. On the
other hand, the MESBB algorithm used a threshold that was derived based on the maximum
amplitude of a sample considering that this represents a strong noise-free signal. However, in
reality the strongest signal doesn‘t always represent a noise-free signal. On the contrary, it
was observed that a peak in the received signal especially in NLOS conditions can be a noise
spike as a result of multipath [104].
Research in [107] carried out an extensive experimental work to examine whether received
signal‘s noise floor could be used as a threshold above which to detect the presence of a
target obstructing LOS. LOS measurements were taken at a distance of 5.5m (measured using
a laser range finder) between one transmit and one receive UWB transceivers device from
Time Domain Corporation‘s PulsON Application Demonstrator (PAD) kit [109]. Further,
NLOS readings were taken using an 0.3m cross-section and 0.72m high steel trash that was
placed at different distances from the transmit and receive devices.
.
78
Figure 39, in [107], showed a typical case of a received UWB pulse with a very low
amplitude noise (background noise) appearing prior to the actual ToA of the pulse.
Figure 39. Noise floor in a received UWB pulse [107]
Initial measurements in [107] where taken in LOS conditions in order to acquire a ―noisefree‖ signal pattern. Thereafter, NLOS received signals were subtracted from the LOS signals
to determine a noise floor reference. The noise floor reference was then used as a threshold
whereas any received signals whose amplitude exceeded the noise floor indicated the actual
ToA of signal from the obstacle (target) in the NLOS condition. That in turn gave the
distance of the target between the receiving and transmitting UWB devices [107].
Nevertheless, it was acknowledged that noise fluctuations can occasionally exceed the noise
floor threshold thus resulting in erroneous ToA data [107].
The authors of [107] carried out relevant work to study the impact of multipath on UWB
signal propagation. Work in [108, 109] consider possible propagation of UWB signals in
indoors and urban environment and carried out tests inside a university office 4.04m by 4.2m.
Two PulsON UWB transceivers were used to transmit and receive signals centred at 1.9GHz
with a BW of 2GHz. Results shown that signal to noise ratio (SNR) decreased as the distance
between the transmitter and the receiver increased. To be more specific for every six meter
increase in distance, from 0m to 30.56 meters, the SNR reduced by 20dB reaching a
minimum of -60dB at 30.56m.
.
79
3.5 Summary
Work presented in this chapter introduced general principles governing UWB technology and
expanded in detail in examining UWB technology within the context of ranging and
localisation.
Starting from the hardware level, literature review on the antennas, used in UWB systems,
revealed that these have a significant role in determining the shape of the propagated pulses, a
factor that needs to be closely considered during signal processing leading to ranging and
localisation. To be more specific, other than ringing effects and the impact of the relative
orientation between the transmitting and the receiving antenna the electromagnetic profile of
the employed antennas alters the shape of the propagated pulses.
Overview of practical UWB-based ranging and localisation applications, that employ ToA
and TDoA techniques, revealed that their accuracy depends greatly to the signal processing
techniques used to derive the LEs of the received signals that in turn are used to compute
ToA and TDoA. To this extent, a number of techniques have been proposed centred around
capturing the LEs. Analysis of the received signals for each application allowed deriving
empirical models usually based on a threshold that was used to detect the LEs. A critical
factor in detecting the LEs was the environment within nodes were placed with indoors
placement presenting a more challenging situation since noise was significantly higher.
Nevertheless, both indoors and outdoors experiments shown that both ranging and
localisation can achieve accuracies from 5cm in LOS cases to 90cm in heavily obstructed
LOS scenarios.
Information acquired in this chapter provided a good understanding of the basic principles
governing UWB-based ranging and localisation methods and laid out the path for the
introduction of the localisation methods employed in this research presented in the following
three chapters.
.
80
CHAPTER 4
PROTOTYPE UWB-BASED LOCALISATION METHOD
4.1 Introduction
The localisation approach discussed in this chapter was developed as part of the WSN4IP
project (introduced in section 1.5) and forms the basis of the research in this thesis. The initial
version of the system was developed by other members of the WSN4IP team, with the aim of
providing high positions accuracies for sensors immersed inside industrial processes as
discussed in 1.5. The requirement for accurate localisation inside hostile (increased
multipath) radio environment, such as grain silos, led to a localisation system being based on
UWB technology and TDoA [26-28]. Indeed, the literature review on UWB-based ranging
and localisation techniques, presented in the previous chapter, revealed that UWB systems
using ToA or TDoA can provide significantly higher accuracy compared with NB-systems.
The aim of this chapter is to provide an overview of the operation of the prototype system
and to report the results of an extensive programme of experiments undertaken by the author.
Comparatively few measurements were taken with the prototype system by its developers and
so an evaluation of its performance was a necessary first step in the development of
improvements that can enhance localisation accuracy.
This chapter begins with an overview of the basic localisation system and continues with
discussion of the employed localisation algorithm. The localisation hardware, including the
electronic and the mechanical parts, is discussed in section 4.4 with experimental localisation
results presented in section 4.5.
4.2 Basic localisation system
The basic localisation system developed in the WSN4IP project was intended to be a research
vehicle for investigating the localisation of wireless sensor nodes immersed in a process
vessel. Measurements to determine the position of an immersed node are taken in by sensors
(in this case antennas) located ‗outside‘ the process. Although this scenario is applicable to
various industrial processed within containers, the actual system was developed and used for
.
81
experiments that concerned the grain storage problem discussed in detail in sections 1.5 and
1.6. However, it would also seem to be appropriate for deriving a node‘s position in the
biomass system and in nuclear ponds, all discussed in section 1.3.3.
An overview of the grain silo localisation system is shown in figure 40 where it is assumed
that the wireless sensor nodes have been distributed throughout the silo. The measurements
positions (receiving anchor antenna) were located on a single plane above the silo. The
reason for this is that it may be expensive, inconvenient or indeed impossible to place
receiving anchors in other locations. Consider for example the grain silo problem. It is
certainly difficult and inconvenient to locate receiving antennas on the walls of the silo and
suspending them within the main body of the silo thus exposing them to the forces imposed
by flowing grain. This is undesirable from the perspective of both the grain flow and the
antennas.
Receive antenna
DSO
Reference
Z axis
0
Transmit nodes
X axis
Y axis
Figure 40. UWB-based localisation layout
The localisation problem in this research was formulated as follows. A node requiring
localisation transmits UWB pulses to an antenna array located at the top of the silo as seen in
figure 41. The receiving antenna array comprises two orthogonal sets of N evenly
horizontally spaced antennas, which define the X and Z measuring coordinate system. Here,
for the purposes of TDoA measurements, one of the antennas in the array was treated as a
reference (see below). The Y coordinates were measured vertically downwards from the
.
82
receiving array. Hence the receiving antenna array was located in the plane Y=0 which was
assumed to coincide with the top of the silo.
The multiple receiving antennas have two roles. The first was to allow signals, from a
transmit node, to arrive at different times at each antenna thus setting the basis for TDoA
measurements. In addition, multiple receiving antennas offered spatial diversity, which
increases the probability that at least some of signals will arrive at the antennas with reduced
noise elements thus improving localisation accuracy [26, 27]. TDoA data were used to
determine estimates of the difference in the distances between node and receiving antenna i,
and the node and the reference receiving antenna.
Distance estimation, using TDoA data, and the prototype localisation algorithm are discussed
in more detail in the following sections.
4.3 Distance difference estimation
The first stage of the prototype localisation method involved deriving TDoA data via a crosscorrelation of the LE waveforms of the received signals since LE represent the portion of the
signals that were considered to be less affected by noise, thus enhancing ranging and
localisation accuracies, as seen in a number of similar applications discussed in section 3.4.2.
As mentioned above, the WSN4IP UWB transmitter produces a train of Gaussian pulses that
are based on the function given in equation (30) in Chapter 3. Figure 41 shows a plot of the
second derivatives of such pulse train, generated by a MATLAB program. Based upon the
discussion of section 3.3, it is clear that figure 41 represents an idealisation of the received
UWB signal.
An example of a real received signal, obtained from the WSN4IP system is shown in figure
42. Comparing the LE in figure 42 to the signal in figure 41, it can be seen that the LE of this
signal is a good approximation to the theoretical signal. However, subsequent parts of the
signal are contaminated by multipath components. Note that in figure 42, and in all following
figures depicting experimental readings, a negative time frame is shown since the reference at
the DSO was set on the negative scale instead of starting from zero; without this having any
impact on the quality of the measurements.
.
83
LE
Figure 41. Ideal received pulse train
Figure 42. Real received UWB pulse
The prototype localisation method employed a range gate to isolate the LE of the received
signals. This LE detection method is called range gate LE detection (RGLED). The range
gate is said in [26-28] to be a ‗temporal band-pass filter‘ that in essence is a square window
used to restrict the part of the signal under consideration to that which could plausibly have
arrived over the shortest path from the transmitter, whilst rejecting the rest of the received
signal. Range gate parameters, specifically the start time of the window and its duration,
determined the portion of the signal to be searched for the LE, which was identified by eye
i.e. the LE detection procedure was not automated.
An example of the RGLED implementation for the signal seen in figure 42 is shown below in
figrure 43.
Figure 43. Gated LE at reference antenna
In the prototype method, range gate parameters (start and duration ) were defined only once
by observation of the signal at referene antenna. Thereafter, these same gate parameters were
used to find the LEs of receieved signals at all the other antennas in the receiving array.
.
84
Once the LEs were isolated, the next step was to obtain TDoA information. TDoA data were
derived using cross-correlation of all signals at receiving antennas to signals at the reference
receiving antenna as discussed in section 2.2.1.
TDoA data were then multiplied by the speed of signal propagation inside grain to compute
distance difference information. For non-reference antenna the distance difference D is given
by:
D  Δ t  c0
(36)
Distance difference D represents the distance from the transmitter to antenna n (Rn) minus the
distance from the transmitter to the reference antenna (R0) as illustrated in figure 44. Note
that, in accordance to the experimental arrangement, receiving antennas lie on a plane where
Y=0 whilst a reference antenna (X0,Y0) lie at origin of axes at (0,0).
100cm
Figure 44. Distance difference diagram
Hence equation (36) can be rewritten as follows:
R n  R 0  Δ t  c0
(37)
If the distance of the transmitting node from the reference antenna R0 is known, then it is
possible to compute the distances of each of the other receiving antenna to the transmitting
node, i.e.
.
85
R n  Δ t  c0  R 0
(38)
However, none of the distances in equation (38) are known, thus an iterative procedure was
adapted that allows estimating distances between the transmitter and the receiving antennas.
Distances are then used to derive the position of the transmitter. This procedure is discussed
in more detail in the following section.
4.4 Localisation algorithm
The prototype localisation algorithm exploits the mathematical relation connecting the
distance differences to the coordinates of both of the transmitter and of the receiving antenna
to compute the position of the transmitter.
Each distance Rn, in equation (38), effectively defines a circle5 centred on a corresponding
receiving antenna, upon which the transmitting node must lie. In principle, the circles centred
on each antenna, on one axis (X or Z) of the receiver array, all intersect at unique point,
which defines the X or Z coordinate of the transmitter accordingly. This is illustrated in
figure 45. The radii of the circles along X axis (same for Z axis), plotted in figure 45, are
given as follows:
Rn 
X t  X n 2  Yt  Yn 2
(39)
where Xt,Yt are the coordinates of the transmitter and Xn,Yn are the coordinates of the
receiving antenna as seen in the following figure.
Figure 45. Localisation using circle intersection
5
.
Actually, it defines a sphere. However, the prototype localisation method simply uses circles
86
The coordinates of the transmitter can then be derived when solving equation (39) for X,Y
and Z using multilateration.
Mulilateration involves subtracting the nth from mth equation in equation (39), thus
eliminating Y parameters, and solving for Xt:
Xt 

R 2x m  R 2x n  X 2n  X 2m  2  X n - 2  X m

2  (X n - X m )
(40)
Likewise the unknown Zt position is derived by subtracting the nth equation from mth, thus
resulting in:
Zt 

R 2z m  R 2z n  Z 2n  Z 2m  2  Z n - 2  Z m

2  (Z n - Z m )
(41)
Finally, the Y coordinates for X and Z axis are derived by substituting Xt and Zt
independently into (38) and solving for Y:
Yt  
n
R 
n
 c 0  Δt n
2   t   n 2
(42)
where ζ is either Xt or Zt accordingly
A key assumption in the above equations is that the distance R0 from the transmitter to the
reference antenna in equation (38) is known. This is not the case, and so a set of assumed
values for R0 were introduced in equation (38) to allow estimating distances used in equations
(40) to (42). The assumed R0 values span the physically plausible range of R0 based on the
size of the vessel used in the experiments. In other words, the localisation calculation is
repeated a large number of times for different assumed values of R0. In cases where the
assumed value of R0 deviate from the true value, the points of intersection between arbitrary
pairs of circles centred on antennas will not coincide. This is illustrated by the inset plot in
figure 46.
.
87
Figure 46. Circular radii intersection [27]
From the above, it can be seen that for each assumed value of R0, a set of intersection points
are generated and the average X,Y and Z coordinates of these points are calculated, along
with the variance of the distribution of points of intersection. The prototype localisation
algorithm chooses as the true R0 the assumed value of R0 that minimises the variance of the
points of intersection. In other words, the value of R0 resulting in a set of intersection points
with minimum standard deviation is taken to be the true value.
On completion of all iterations, at each assumed distance R0, a single (X,Y,Z) position was
derived by taking the average of all of the individual generated positions and the standard
deviation (STD) was computed:
f R0  =
2
2
1 N-1 
X n -X  +  Yn -Y  





N n=0
(43)
The assumed range estimate Rn minimising equation (43) was then substituted in equations
(40) to (42) to derive the final X, Yand Z positions of the transmitting node.
4.5 Algorithm implementation
The prototype localisation algorithm discussed in the previous sections was implemented in a
MATLAB program that runs on a PC. Input data required by the program include digitised
versions of the signals received by each of the antennas in the receiving array and range gate
parameters. The digitised signals are stored in Excel files, and the range gate parameters are
derived from a visual inspection of the signal from the reference antenna.
.
88
The program is executed in a number of stages. In the first stage, the sampled signals are read
from file and their LEs are isolated using the range gate parameters derived from an
inspection of the signal at the reference antenna. The LEs are then cross-correlated to derive
TDoA data, Δt =(t1-tn) that in turn were used to estimate distances between the transmitter
and the receiving antenna array.
Distance differences are obtained from the TDoA values by multiplication by the speed of
propagation in grain. It is important to note this is different from the signal propagation speed
in air, due to differences in the dielectric constants. As indicated in section 1.6.2, the
literature suggests that dielectric constant of grain is between 2.5 and 2 an observation that
was confirmed by WSN4IP researchers in their measurements over frequencies between
1GHz to 8GHz [28]. The speed of propagation is equal to the inverse of the square root of the
medium‘s dielectric constant [111], and as such the propagation speed c used in the algorithm
was 1.875×108m/s.
In the second major stage of the program‘s execution, the iterative procedure discussed in the
previous section is applied to the data for a number (n) of assumed values for R 0. In the code,
these, assumed values range from 0.1m to 1.5m, in steps of 0.01m. Distance difference data
are combined with the assumed value of R0, defining the radii of the circle on which the
transmitter is thought to lie. This enables all points of intersection between circles to be
computed and the mean and STD of thee distribution of the intersection points to be found.
This process is repeated for all n assumed values of R0 and the value that minimised the STD
is selected as the true value. This minimisation is illustrated in figure 47, which shows a plot
of the STD cost function in equation (43).
Figure 47. Variance minimisation Vs assumed distances [27]
.
89
Note that there are two values for the Y positions– one for the X coordinate and one for Z
coordinate given equation (42) gives Y estimates using the equations of circles (2D).
4.6 The localisation system
A basic experimental UWB-based positioning system was developed by WSN4IP researchers
as shown in figure 48. This consisted of a cylindrical plastic container, 1.5m high and 1m in
diameter, was filled with grain.
Figure 48. UWB positioning in a grain silo [27]
The following sections discuss in detail both the electronic and mechanical systems employed
in the localisation procedure.
4.6.1 Electronic systems
The electronic system consisted of two parts: the circuitry inside the node and the external
system. In terms of the former, it should be noted that in the work reported in this chapter,
and in chapters 4 and 5, only the hardware associated with localisation were included within
the nodes (see section 1.5). The external system included the receiving antenna array, a
Digital Sampling Oscilloscope (DSO), a PC and a number of other elements.
Transmitting nodes were equipped with a pulse generator that generated UWB pulses with a
width of 0.5ns. The pulse generator circuit (figure 49) comprised an avalanche transistor
driven by a 1MHz crystal oscillator, through a frequency divider, thus setting the output pulse
repetition frequency (PRF) at 500kHz. The avalanche transistor requires 70V to operate
.
90
which is provided by a DC to HV DC converter that takes 2.4V from a battery embedded
within the transmitting node.
Figure 49. Pulse generator circuit [27]
Initially the pulse generator was powered by two AA rechargeable batteries but that was not
efficient as the batteries needed long time to recharge thus causing delays in the experiments.
In all measurements presented in this thesis, the pulse generator was powered via cables
connected to an external power supply.
Both the transmitting and the receiving antenna were ring monopole broadband antenna that
can handle UWB pulses. Receiving antenna were fabricated exclusively from FR4 substrate
whilst a number of receiving antenna used a combination of FR4 substrate and flexible
ground plane with another set of receiving antenna being made exclusively from the flexible
material. The ring shape of the antenna and the flexible ground plane allowed mounting the
antennas inside the protective spheres. The transmitting antennas are seen in figures 50a and
50b whilst the receiving antenna is shown in figure 50c.
(a)Flexible antenna
(b) Antenna inside mortar shell
(c) FR4 antenna
Figure 50. Ring monopole transmit and receiving antennas
.
91
As stated in section 1.5, experiments were performed in small scale and large scale vessels.
Tests in the small vessel employed single antennas as the one seen in figure 50b, on the other
hand in the large vessel tests were carried out using pairs of transmit antennas placed inside
mortar shells as seen in figure 51.
Batteries
Pulse generator
Ground
plane
Antenna
Signal combiner
Figure 51. Transmitting antenna for tests in the large vessel
Figure 51 shows two ring monopole antennas fabricated from FR4 substrate with a flexible
ground plane fixed on each of the hemispheres of the protective cover. The antennas were
placed with orthogonal polarisation so they could propagate simultaneously but minimise
signals from one antenna interfering with signals from the other antenna. The two antennas
were connected together via a broadband signal combiner that in turn was connected to the
pulse generator (figure 49) to enable the antennas to transmit simultaneously. The two
antenna formation aimed to alleviate problems which would arise from random orientation
(see section 3.3.1) of nodes inside grain silos and hence approach the condition where the
antenna array could be regarded as omni-directional.
Received signals were amplified using a LNA (to boost the weak received signal (~µV) and
then digitized using an Agilent 54854A DSO. It is important to note that received signals
enter the DSO through two different antennas at the same time. Signals from the array of the
receiving antenna are used within correlation procedure to provide TDoA information.
Signals from a second receiving antenna (not part of the receiving antenna array), situated at
a fixed position inside the grain, are used by the triggering circuit to start the sampling
procedure. The objective of the triggering circuit was to detect received pulses and in turn
trigger the DSO to start sampling [26-28]. The DSO was configured to sample signals from
.
92
the antenna array at five giga samples per second (GSa/s). A received UWB pulse trail
captured using the DSO can be seen in figure 52.
Figure 52. DSO UWB pulse trail
The averaging function of the DSO was enabled in order to remove random noise thus
guaranteeing the reliability of the detected signals and consequently the reliability of the
whole localisation procedure.
4.6.2 Mechanical system
The mechanical parts of the system included two plastic vessels to hold the grain, a system
based on an industrial vacuum cleaner, for filling and emptying the testing vessel, a set of
plastic mortar shells to hold the node electronics and supports for the antenna array.
The small vessels used in tests were located in a university electronics laboratory whilst the
large vessels were located at a wide space of a University Chemical engineering laboratory.
Strict health and safety requirements imposed specific working regulations to protect users
from damaging effects of grain dust as well as to prevent users from falling inside the vessels
during experiments (handmade safety net acquired by the author, seen in figure 53c,
required). Nevertheless, these health and safety regulations caused significant delays in large
scale tests resulting in a limited number of experimental data compared to the data from small
scale tests.
In all cases a second plastic vessel was used for grain storage to enable the primary vessel, in
which the experiments were contacted, to be partially or fully emptied for the purposed of the
nodes. Figure 53a shows a schematic diagram of the system used to convey grain from one
.
93
vessel to another. The conveyor system was based on a powerful industrial vacuum cleaner
and a hopper, seen in figures 53b (small scale tests) and 53c (large scale tests). The hopper
was filled with grain from one vessel (filled with grain) and then transferred to the top of the
empty vessel to release the grain.
(a) Grain conveyor
(b) Small vessel test rig
(c) Large vessel test rig
Figure 53. Experimental rig
One of the main issues with grain transfer was the time (~1hour) required to fill or empty the
vessels to place or remove nodes accordingly. As therefore, a faster method of node insertion
and removal, in small scale tests, was developed by the author. This used a hollow metal tube
that can be seen in figure 54. The tube was pushed inside grain at the desired position and the
vacuum cleaner (figure 54) used to remove the grain from the tube. Here it is important to
note that the tube was always placed 15cm deeper than the required position to compensate
for measuring node‘s positions from the upward side of the sphere. The penetration of the
tube into the grain was eased by helical extensions on its outside periphery which allowed the
tube to be ‗screwed‘ into and out of the grain.
Once grain had been removed from the tube the node and the required depth was reached, the
sphere was placed inside and measurements taken of its position with a 1.5m rule and a tape.
The rule was placed on the middle, upward side (facing top of the silo), of the spheres to
measure the depth, Y coordinates, whilst the tape was used to measure X and Z coordinates
from the ruler. It is estimated that the accuracy of these measurements was ±5cm in all cases.
.
94
(a) Metal tube
(b) Industrial vacuum cleaner
Figure 54. Small scale experimental setup
Once nodes were placed at the required position, the tube was removed from grain by pulling
it upwards, in clockwise and anticlockwise rotations, whilst grain at the immediate external
side of the tube was pulled away by hand. Throughout the tube extraction process, the rule
was hold firmly on the top of the sphere minimise possible displacements from the required
position. The exact point of the tube extraction was marked to allow using the tube once
measurements were completed to extract the spheres. It is estimated that during the extraction
of the tube there was s slide movement of the node but that wasn‘t significant and certainly
less than 5cm.
Overall, since measurements of the sphere positions was taken from their centre and given
their diameter was 15cm deviations in position estimates, by the localisation method, within
±10cm is an acceptable degree of error. Even though the tube-based procedure for inserting
and removing nodes was quicker that partially emptying/refilling the vessel, still required a
lot of effort and could have taken from one hour up to four hours to be completed, depending
on the depth to which a node had to be delivered.
The whole procedure of placing a node inside grain, taking of measurements and the
extraction of the nodes from grain required a lot of effort and could have taken from one hour
up to four hours depending on the depth to which a node had to be delivered.
Preliminary small scale tests were performed using a single receiving ring monopole antenna
that was placed on a rotating plastic rig situated on the top of a vessel as seen in figure 55.
This antenna was continuously connected to the DSO, via a cable, and was moved in steps of
.
95
10cm a time, across the plastic rig, to allow for sixteen measurements to take place eight
along the X axis and another eight along Z axis. Note that another antenna on the side of the
tank (seen in figure 55) was constantly connected on the DSO and, acted as a clock, to trigger
the DSO to start sampling in the presence of received signals [26-28].
Figure 55. Single receiving antenna on a rotating rig
In a later stage of the research, a different arrangement was employed were nine receiving
antenna were fixed on cross formation (figure 56) with respective cables connecting one
antenna at a time to the DSO. This cross antenna arrangement was developed by the author
along with another researcher working on WSN4IP project. Notice that all antenna faced in
the same direction, in another measure, to combat orientation issues. To this extent,
transmitting nodes were placed inside the small vessel in a way so that their antenna faced the
receiving antenna in the same orientation.
The plan view of the silo with the orthogonal antenna array made of ring monopole antennas
is shown in figure 56.
Z axis
X axis
Figure 56. Small vessel receiving antenna on a cross formation
.
96
The large scale vessel is shown in figure 57a with the same receiving antenna formation used
in small scale vessel also used on the top of the large vessel as seen in figure 57b.
Z axis
X axis
(a) Large testing vessel
(b) Received antenna on top of large vessel
Figure 57. Large silo and receiving antenna
In all cases receiving antenna were held at their position on the plastic rig using plastic
screws to prevent interferences to the received signals. Further, during tests, only one
receiving antenna was connected to the DSO thus only one measurement was taken a time.
The procedure of connecting the antenna cables to the LNA and to then to the DSO needed
special attention since cables had to be firmly connected to the LNA to prevent erroneous
measurements. As therefore, every time a new cable had to be connected to the LNA a
wrench was used to ensure the cable was firmly connected to it.
4.7 Experimental investigation
Extensive evaluation of the prototype localisation method was carried out by the author
through a study of thirty test cases; fifteen along X axis and of another fifteen along Z axis
resulting in 202 measurements for tests in the small vessel and in 30 measurements in the
large vessel. The MATLAB algorithm gave TDoA estimates and the positions of the transmit
nodes and as such those two were used to evaluate the accuracy of the localisation procedure.
Figure 58 shows the measured positions (measured manually using a ruler) of the
transmitting nodes inside the grain silo as opposed the estimated positions for each test case
via the prototype method for both small scale tests and for large scale tests. Note that the
numbers inside every marker depict a different transmitter position representing a different
.
97
experiment. In all cases the measured positions are depicted by yellow squares whilst the
estimated positions are depicted by red circles. The dimensions of the testing vessels are
denoted by the red square enclosing the results.
(a) Small scale tests
(b) Large scale tests
Figure 58. X axis position estimates from prototype method
The absolute error (AE) (AE=|Experimental position – Measured position|) for the X
coordinates, from each test, is presented in histograms in figure 58 and 59. Note that in all
histograms the results for tests in the small vessel are presented with the first 16 experiment
numbers whilst results from the large vessel are presented in experiment numbers 17 to 19. In
all cases the experiment numbers in the histograms in figure 59 and in all histograms
presented thereafter represent different experiments conducted at different positions inside
the test vessels.
(a) AE for X coordinates
(b) AE for Y coordinates
Figure 59. AE for experiments across X axis
The estimated and the measured Z positions, for each experiment, are shown in figure 60.
.
98
(a) Small scale tests
(b) Large scale tests
Figure 60. Z axis position estimates from prototype method
The AEs for the estimated Z coordinates are shown in figure 61.
(a) AE for Z coordinates
(b) AE for Y coordinates
Figure 61. AE for experiments across Z axis
Table 1 gives the mean AE (MAE) for all X and Z positions. Note that a single, common, Y
estimate for X and Z results is derived by taking the average of the two Y estimates recorded
for X and Z readings.
Small scale tests
MAE(cm)
STD(cm)
Large scale tests
X
Yx
Y
Yz
Z
X
Yx
Y
Yz
Z
10.84
12.12
21.5
22.53
20.7
17.3251
19.9
18.73
12.94
11.43
33.75
24.3
69.44
55.2
78.32
48.6
87.2
30.84
14.3
8.5
Table 1: Prototype localisation MAE and STD
.
99
4.7.1 Analysis of experimental results
A close examination of the localisation results, out of the prototype method, showed that
errors are directly related to the TDoA data. As such, understanding errors in TDoA will help
improve understanding of the localisation results.
This section presents an analysis of the experimental TDoA used to derive the position
estimates. For practicality reasons, TDoA data were multiplied by the speed of signal
propagation inside grain thus resulting in an experimental distance difference Dexp. Given the
coordinates of the transmitting and receiving antennas, distance differences could be
estimated in theory thus providing a basis to which experimental distance difference can be
compared to. The mean absolute error (MAE) between the experimental and theoretical
distance difference profiles was used to evaluate the accuracy of the experimental data:
MAE 
1
n
n
D
i
theor
 D iexp
(44)
i 1
Table 2 shows an example of distance difference profile values, obtained using the RGLED
method, at transmitter position of X=42cm, Z=42cm and Y=72cm.
Receiving antenna
positions
Theoretical
profile(cm)
Experimental
profile(cm)
Absolute
Error(cm)
MAE(cm)
0cm
10cm
20cm
30cm
40cm
50cm
60cm
70cm
0
-4.6
-8.1
-10.4
-11.3
-10.9
-9.14
-6.10
0
14.5
10.2
6.4
4.3
3.4
3.11
6.7
0
19.1
18.3
16.8
15.6
14.3
12.25
12.8
13.64
Table 2: RGLED distance difference profiles along X axis
What can be seen in table 2 is that experimental distance difference values deviate
substantially from the theoretical distance difference; an observation seen throughout results
acquired using the RGLED method.
Figures 62 and 63 show the MAE for distance differences computed via the RGLED method,
for each experiment, averaged over the number of the receiving antenna in the X and Z axis
respectively. The MAE recorded in table 2 is given at experiment number 2 in figure 63.
.
100
Figure 62. Distance difference X axis
Figure 63. Distance difference Z axis
Table 3 gives the overall MAE and STD values for distance difference profiles for both small
scale and large scale tests.
Small vessel tests
MAE (cm)
STD (cm)
Large vessel tests
X
Z
X
Z
9.21
11.65
8.76
9.57
14.4
7.46
10.65
3.9
Table 3: RGLED distance difference profile errors
The following results, for transmitting node at X=42cm, Y=72cm, representing experiment
number 2 in figures for X axis results in section 4.7, reveal the drawbacks in the RGLED
localisation method. Figures 64 and 65 show both the raw received signals and the gated
signals at the reference antenna.
Figure 64. Raw data at reference (0cm)
Figure 65. Gated LE at reference (0cm)
What can be seen in figure 65 is that the gated signal at the reference is a good
representation of 2nd derivative of Gaussian pulse as discussed in chapter 3.
.
101
Figure 67 shows a gated received LE, for transmitting node at X=42, Y=72, at an antenna
10cm from the reference of which the raw received signal is shown in figure 66. Here it can
be seen that the gated LE starts at the exact time as the signal in figure 65 something that
shouldn‘t happen given its distance from the reference antenna. Further, the waveform in
figure 67 deviates from the expected second derivative of a Gaussian pulse that was
expected as discussed in chapter 3.
Figure 66.Raw signal at reference at 10cm Figure 67.LE at 10cm from the reference
The signal in figure 67 gave rise to an error of 19.1cm in the computed distance difference.
An increased error in distance differences is a characteristic of the effect of the RGLED
method and its failure to detect the appropriate LEs.
4.8 Summary
This chapter presented a prototype UWB-based localisation method where LEs have
significant role in determining the accuracy of localisation. Nevertheless, the process
followed in selecting the LEs was not automated and lacked robustness.
The dependency of position estimates on the LEs is reflected on the errors recorded for
individual experimental distance difference profiles and their respective localisation errors;
whereas the latter increases or decreases depending on whether distance difference profile
error increases or decreases respectively.
The MAE and the STD for positions estimates in the small vessel exceeded 10cm whilst
exceeded 70cm for positions in the large vessel. Reflecting the errors in position estimates the
MAE and STD for the distance differences exceeded 8cm in most cases. A total of 68.75% of
.
102
all position estimates resulted in errors exceeding 10cm with a 46.87% of all distance
differences MAE being above 10cm for all measurements.
To deal with problems rising in the prototype localisation method and specificly in the LEs
detection method, a new approach was examined were LEs are adaptively selected for each
receiceived signal in automated method discussed in the following chapter.
.
103
CHAPTER 5
ADAPTIVE LEADING EDGE DETECTION
5.1 Introduction
Chapter 4 introduced a UWB-based localisation method for wireless sensor nodes immersed
in a process vessel, which in this work was a grain silo. The results of an extensive
programme of testing showed that results obtained with this method were highly prone to
errors. Further analysis of the results indicated that an important source of error was the
accuracy with which the LEs of the signals, received by the antenna array were detected. In
many cases the LEs do not appear to be correctly detected leading to errors in computed
cross-correlations between signals. Information out of cross-correlations was used to derive
TDoA values which were consequently in error, a fact that is confirmed by comparison with
theoretical values. TDoA data forms the input to the localisation algorithm and so the
transmitter locations computed by the method were incorrect.
The LE detection method used by the localisation system described in Chapter 4 was based
on parameters derived by observation of a single signal by a human user of the system. These
parameters were then applied to all other signals emerging from the receiver array. There are
numerous potential problems with this procedure including the fact that LE identification was
based on human observation and the results are likely to vary over time in an unpredictable
manner, especially when determined by different users. Moreover, a technique requiring
human intervention is not viable for practical systems. Another important problem is that the
LE detection criteria for a signal received on channel i (i.e from the ith antenna in the array)
are independent of this channel. Given that a key aspect of LE detection is to distinguish a LE
from the noise floor of a particular channel, it seems reasonable that an algorithm seeking to
detect a signal LE on channel i should consider the noise floor on that channel.
This chapter presents two new LE detection methods that are automated, in the sense that
neither required input from a human user. Moreover, both determine detection for a signal on
channel i, by measuring parameters associated with channel i. The first method developed is
called the Basic Advanced LE Detection (BALED) method, which employs a measured
.
104
threshold in LE detection. This is discussed in section 5.2. The second method, presented in
section 5.3, is an improved version of BALED so called IALED. Results obtained from
applying the two approaches are presented in section 5.4 and these are compared with the
results acquired from applying the ‗manual‘ technique used in Chapter 4.
5.2 Basic advanced LE detection method
The BALED uses a single threshold to identify the LE of an incoming UWB pulse. The
BALED is similar to the methods in [104, 107] discussed in section 3.4.2 although those
methods considered using a threshold to only derive the ToA of the LEs. The novelty of the
BALED method lies on the fact that it uses a threshold to detect the start of the LE (can be
considered as ToA) but it goes one step further and detects a complete waveform that is used
within correlation to derive TDoA.
The threshold applied to the signal received from antenna i in the receiver array (i.e. the
signal received on channel i) is computed from measurements of the noise floor on that same
channel. The noise floor is defined as that part of the received signal that precedes the arrival
of the actual received pulse as in [107]. Figure 68 shows a typical received signal, for a
transmitting node at X=72cm, Y=42cm, Z=42cm, were the noise floor is easily identifiable.
Noise
floor
Noise floor
(a) Received signal 10cm
(b) Closer view of signal at 1ns/div
Figure 68. Noise floor identification
The noise floor on channel i is used in the setting of a threshold for the LE detection on the
same channel. Specifically, it was observed that, in all cases, the noise on channel i
approximates a Gaussian distribution as shown in the histogram in figure 69; that depicts the
.
105
distribution of the noise floor for eight signals at eight different antenna (different colour
depicts different signal).
Figure 69. Histogram of received signal noise floor
The threshold for LE detection is then set as an integer multiple (α) of the STD to avoid noise
spikes being erroneously identified as the LEs of the received pulses. For example, if α = 2,
then 95% of the noise distribution lies below the detection threshold, and if α = 3 then almost
98% of the noise distribution lies below the detection threshold [113].
Relative to the RGLED method discussed in the previous chapter, the BALED method has
the advantage that is automated and hence produces a greater degree of consistency than the
method associated with the RGLED approach. It also takes into account the channel noise,
which forms the background against which the LE must be detected.
The BALED method follows an adaptive approach to detect the LE. At the beginning the
algorithm records the STD of the noise floor, for each individual received signal, and sets a
threshold above which the LE is to be detected. It is important to note that the BALED
method can adjust the threshold accordingly to accommodate for increased or decreased
STD. Nevertheless, in all cases the threshold remains above a specific level (three times the
STD) in order to avoid picking up noise spikes instead of the LE. The time at which the
amplitude of a sample, within the received signal, exceeds the computed threshold,
determines the start of the LE.
The following pseudo-code describes the procedure with which the BALED method detects
the beginning of the LE:
.
106
1. FOR i=1 to all received signals, N
2.
Compute noise floor standard deviation, STD(i)
3.
Compute threshold, THRESHOLD(i)= STD(i)×α
4.
IF THRESHOLD(i) less than minimum value β then set
THRESHOLD(i)=β+γ
5.
END IF
6.
FOR j=N to P samples
7.
IF sample(j,i)> Threshold(i)
8.
Start(i)=j
9.
END IF
10.
END FOR
11. END FOR
Threshold values α (α >1), β and γ, in the pseudo-code, were empirically selected based on
thorough analysis of data available.
The time at which the LE begins, is used as a reference from which three successive
crossings with the zero axis are recorded in order to capture a complete LE that is used within
correlation to derive TDoA.
An example of a LE taken out of the experimental procedure is shown in figure 70 (for
transmitter at X=42cm, Y=72cm and received signal at 10cm from reference antenna) where
the start of the LE as well as three successive crossings with the zero axis are clearly marked.
The pulse in figure 70 approximates the 2nd derivative of the Gaussian pulse that was
expected to be seen at the receiving antenna.
Figure 70. LE at X=10cm
At this stage, it is important to mention that both the BALED and the IALED methods record
LEs that approximate the second derivative of the Gaussian pulses as discussed in section 4.2.
.
107
The following flow chart exemplifies all the steps involved in the BALED method and how
these are linked to the RGLED localisation algorithm.
Read received signal from
channel i
Compute noise floor in channel i
Detect start of LE(i)
Detect three crossings with zero
axis to get a complete LE(i)
Correlate complete LEs to derive
TDoA
Use prototype algorithm to
compute transmitter coordinates
Figure 71. BALED-based localisation method
Detailed experimental evaluation of the BALED method will be discussed in detail in section
5.4. Further analysis of the results, out of the BALED method, revealed that these are prone
to errors due to the possibility that noise spikes could mistakenly detected as the LEs.
An example of the deficiencies in the method presented here is exemplified in figures 72 and
73 that show a raw received signal at 60cm from the reference receiving antenna for
transmitter at X=75cm, Y=30cm and its detected LE respectively.
The LE pulse, seen in figure 73, resulted in a distance difference value of -94.4cm instead of
-44.5 that was expected at the specific transmitter position. On the other hand, if the correct
LE pulse (sown inside red circle in figure 72) was selected then the distance difference value
resulted in -45.74cm that is very close to the theoretical value
.
108
LE
Figure 72.Raw received signal at 60cm
Figure 73. Detected LE at 60cm
Additional study of distance difference and localisation results revealed that in some cases
the error was high when pulse duration extends beyond 1ns. One such case can be seen in
figures 74 and 75 that shows a raw received signal and the gated signal respectively at 40cm
from the reference antenna at transmitter of X=70cm, Y=30cm.
Figure 74.Raw received signal at 40cm
Figure 75.Detected LE at 40cm
The waveform in figure 75 is significantly distorted by multipath, thus giving a distance
difference of -6.98cm instead of -33.7cm that was the expected value.
To deal with problems in the BALED method an improved advanced LE detection method
(IALED) was developed.
5.3 Improved advanced LE detection method
IALED attempts to combat the deficiencies observed in the basic method by taking advantage
of the restricted area of the vessel in which measurements took place. In order to deal with
problems observed in BALED method two approaches have been developed.
.
109
In order to deal with issues arose from prolonged signals (see figure 75) it was examined
whether it is possible to process the earliest part, in time, of the LEs since the further in time
the signal the further multipath increases. This reduced portion of the LE is considered only
for the cases when the time difference between the start of the LE, and the time at which the
maximum or the minumum of the LEs occurs, exceeds 0.5ns.
Figure 76 shows a typical LE when used in the localisation process results in accurate results.
What is seen here, is that both the maximum and minimum points after the start of the LE are
within 0.5ns.
maximum
0.47ns
LE start
0.17ns
minimum
Figure 76. LE minima and maxima points
Figure 78 shows a gated signal for transmitter at Xt=70cm, Yt=30cm where only two
crossings with the zero axis are considered thus giving a reduced waveform than in figure 78.
The raw received signal for the gated pulse is shown in figure 77.
Figure 77. Raw received signal at 40cm
Figure 78. Detected LE at 40cm
The LE, seen in figure 78, resulted in a distance difference of -24.92cm deviating by only
7.61 cm from the expected theoretical distance difference value.
.
110
In order to deal with cases where the LE is missed, when is taken to be far away from the
expected time it should have occurred, a boundary was set within which to look for the LE.
Given the dimensions of the container in which the tests were carried out, and the positions of
the receiving antennas, it is possible to restrict the search for the LE. Specifically, since
adjacent antennas are separated by 10cm up to 80cm, then it is anticipated that the LEs will
occur within short time period from each other. In fact, analysis of all available data revealed
that LEs occur within about 3ns from the start of the LE recorded at the reference antenna.
Here the 2ns limit between pulses can be confirmed mathematically based on the example
seen in figure 79.
B(80,0)
A(0,0)
80cm
X
T(18,10)
Y
Transmitt
er
Figure 79. 3ns Pulse limit example
In the example depicted in figure 79, given a transmitter position T (X=18cm, Y=10cm) and
two receiving antenna A(X=0m, Y=0m) and B( X=80cm, Y=0cm) then Pythagoras theorem
will give the distance of the transmitter to receiving antenna B at 80.62cm. The optimum
propagation time, (time=distance/speed), of a signal from the transmitter to receiving antenna
A will be 0.33ns whilst the propagation time to receiving antenna B will be 2.7ns, given
speed of propagation is 1.9E11m/s.
The flow chart in figure 80 shows all the steps involved in the IALED method where the first
three steps are the same as in the RGLED method. The time period from the start of the LE at
the reference antenna within to look for the ToA of signals at the rest of the antennas is
denoted by the letter ε, also the optimum duration of the LE is denoted by letter η.
.
111
Read received signal from
channel i
Compute noise floor in channel i
Detect start of LE(i)
Set LE detection boundaries:
Detect start of LE(i) within
[start LE(i) to ±ε× start LE(1)]
No
Yes
Duration of LE(i)>η?
Detect three crossings with zero
axis to get a complete LE(i)
Detect two crossings with zero
axis to get a complete LE(i)
Correlate complete LEs to
derive TDoA
Use prototype algorithm to
compute transmitter coordinates
Figure 80. IALED localisation method flow chart
5.4 Evaluation of the BALED method
The need for the BALED method was thoroughly discussed in the previous sections, yet a
comparison between experimental distance difference profiles, obtained both by the RGLED
LE detection method and the BALED method, reinforces the need to implement an adaptive
LE detection approach.
Figure 81 shows the measured positions (obtained manually using a ruler) of the transmitting
nodes as opposed the estimated positions via the BALED localisation method presented in
this chapter.
.
112
(a) Small vessel tests
(b) Large vessel tests
Figure 81.Measured and estimated X axis positions
Figure 82 gives the AE for the estimated positions for all data recorded along X axis.
(a) AE for X coordinates
(b) AE for Y coordinates
Figure 82. AE for X axis positions
Figure 83 shows the measured as opposed to the estimated positions for data recorded along
Z axis.
(a) Small vessel tests
(b) Large vessel tests
Figure 83.Measured and estimated Z axis positions
.
113
Figure 84 gives the overall AE for measurements along Z axis and the respective overall AE
positioning error along Y axis.
(a) AE for Z coordinates
(b) AE for Y coordinates
Figure 84. AE for Z position estimates
The MAE and STD for X, Y and Z position estimates when using the BALED localisation
method are shown in table 4.
Small vessel tests
MAE(cm)
STD(cm)
Large vessel tests
X
Yx
Y
Yz
Z
X
Yx
Y
Yz
Z
5.93
5.934
10.7
7.54
11.95
11.16
13.02
19.11
6.3
7.82
25.4
26.74
42.4
30.1
42.5
33.8
42.6
36.4
7.1
9.75
Table 4: BALED MAE and STD
5.4.1 Analysis of BALED results
A more detail analysis of the BALED localisation method is carried out based on an
examination of the acquired distance difference profiles. The MAE for all distance
differences are shown in figures 85 and 86.
Figure 85. Distance difference X axis
.
Figure 86. Distance difference Z
114
Table 5 gives the MAE and STD for the distance difference profiles acquired using the
BALED method.
Small vessel
tests
X
Z
MAE(cm)
STD (cm)
3.3
4.5
4.23
4.1
Large vessel
tests
X
Z
17.82
4.4
3.4
3.24
Table 5: BALED distance difference profile errors
The following four figures show characteristic examples of the improvements achieved in
distance difference computation when using the BALED method as opposed to the RGLED
method. All figures give plots of both theoretical and experimental distance differences
against the receiving antenna positions for a transmitter position inside the small vessel. Note
that theoretical distance difference profiles were acquired for the transmitter position used in
the experiments and for a reference receiving antenna position at X=0cm, Y=0cm,Z=0cm.
Figure 87a shows the experimental distance difference profile, for a transmitter at X=53cm,
Y=89cm, derived using the LE detection method used in the RGLED method presented in the
previous Chapter. In addition figure 87b shows the experimental distance difference profile,
at the same transmitter position, computed based on the BALED method. Both figures show
the theoretical distance difference profiles at X=53cm, Y=89cm for comparison.
(a) RGLED method
(b) BALED method
Figure 87. Distance difference profiles at X=53cm, Y=89cm (experiment number 3)
.
115
Similarly figure 88 shows the theoretical against the experimental distance difference profiles
computed based on the RGLED method on BALED for a transmitter position of X=42cm,
Y=72cm.
(a) RGLED method
(b) BALED method
Figure 88. Profiles at X=42cm, Y=72cm (experiment number 3)
What can be seen in the above figures is that the BALED method outperforms the RGLED
method resulting in more accurate distance differences and consequently giving more
accurate position estimates. Nevertheless it was observed that there are cases where BALED
fails to produce good representations of distance differences thus introducing errors in
localisation. Two characteristic examples of distance differences pointing out the weaknesses
of BALED method are shown in figure 89.
(a) RGLED method
(b) BALED method
Figure 89. Profiles at Z=44cm, Y=45cm (experiment number 16)
The MAE for the profile seen in figure 89b, derived when using BALED, was 25.3cm
resulting in 9.1cm AE in the position estimate.
.
116
5.5 Evaluation of the IALED method
Figure 90 shows the measured as opposed the estimated positions along X axis.
(a) Small vessel tests
(b) Large vessel tests
Figure 90.Measured and estimated X axis positions
The AE for X position estimates are graphically shown using histograms in figure 91.
(a)
AE for X coordinates
(b) AE for Y coordinates
Figure 91. AE for X axis position estimates
Figure 92 shows the measured as opposed the estimated positions across Z axis.
(a) Small vessel tests
(b) Large vessel tests
Figure 92. Measured and estimated Z axis positions
.
117
The AE for estimated Z coordinates are depicted in a histogram in figure 93.
(a) AE for Z coordinates
(b)
AE for Y coordinates
Figure 93. AE for measurements across Z axis
The MAE and STD for all position estimates using the IALED localisation method are given
in table 6.
Small vessel tests
MAE(cm)
STD(cm)
Large vessel tests
X
Yx
Y
Yz
Z
X
Yx
Y
Yz
Z
5.1
4.4
7.51
6.4
7.81
5.94
8.1
7.22
6.33
5.02
26.45
25.01
13.65
18.25
33.425
25.8
53.2
39.4
13.2
20.2
Table 6: IALED localisation method MAE and STD
5.5.1 Analysis of IALED results
Figures 94 and 95 next depict the distance difference MAE for the data points along X and Z
axis at which readings were observed.
Figure 94. Distance difference X axis
.
Figure 95. Distance difference Z axis
118
The MAE and STD for distance differences along X axis as well as the MAE and STD for
distance differences along Z axis are given in table 7.
Small
vessel tests
X
Z
MAE(cm)
STD(cm)
2.9
2.55
3
2.5
Large vessel
tests
X
Z
7.131
5.92
6.431
6.12
Table 7: MAE for IAELED distance difference profile errors
Figures 96 and 97 exemplify the enhance accuracies achieved with the IALED method
compared to profiles computed using the RGLED LE detection method, the BALED and
IALED methods. Figure 96 depicts profiles for transmitter position at Z=44cm and Y=45cm.
(a) RGLED & BALED
(b) RGLED & IALED
Figure 96. Profiles at Z=44cm, Y=45cm (experiment number 16)
Likewise figure 97 shows profiles for transmitter position at Z=44cm and Y=45cm for all
employed LE detection methods where IALED is shown to be the most accurate.
(a) RGLED & BALED
(b) RGLED & IALED
Figure 97. Profiles at X=70cm, Y=30cm (experiment number 11)
.
119
Figures 98 and 99 show the raw received signal and LE respectively for a signal at the
reference receiving antenna, for a transmitter at Xt=70cm, Yt=30cm.
Figure 98. Reference received signal
Figure 99. LE at reference antenna
Figure 101 presents the gated signal for the same received signals as in figure 73, in page
108, but this time using IALED. Here, it can be seen that the gated signal is different from the
one shown in figure 73 that was detected solely based on the noise floor threshold.
Figure 100.Raw received signal at 60cm
Figure 101.Detected LE at 60cm
The LE, in figure 101, (Xt=70cm, Yt=30cm) gave a distance difference of -45.7cm compared
to the theoretical distance difference of -44.5 The example seen in figure 101 and indeed
many other cases revealed that IALED can deal effectively with noise spikes and can give
more accurate results compared to the original noise floor based threshold detection.
Overall, results from experimental distance differences and localisation estimates using all
the LE detection methods showed that although there are improvements for the results out of
the small vessel tests, results using data out of the large vessel tests do not provide firm
conclusions. Nevertheless, analysis of raw received signals revealed that signals in the small
test vessel were least affected by noise as opposed to signals out of the large vessel as seen in
figures 102 and 103 respectively (received signals at an antenna 20cm from the reference).
.
120
Small vessel
Figure 102. Raw signal at X=28, Y=89cm
Large vessel
Figure 103. Raw signal at X=17, Y=150cm
Figure 103 shows a representative case of a received signal in the large vessel where
increased noise caused problems in the detection of the LEs and subsequently contributed in
increase errors in distance and position estimates.
5.6 Summary
This chapter presented two adaptive LE detection methods that significantly improve
positioning accuracy as opposed to the RGLED method discussed in the previous chapter.
Table 8 gives the MAE for positions estimates using all LE detection methods in all tests
inside the small and in the large vessel.
Algorithm
Prototype (RGLED)
Prototype (BALED)
Prototype (IALED)
Small vessel MAE (cm)
X
Y
Z
10.84
20.7 12.94
5.93
11.95
6.3
5.1
7.81
6.33
Large vessel MAE (cm)
X
Y
Z
33.75
78.32
14.3
25.4
42.5
7.1
26.45
33.425
13.2
Table 8: MAE for all position estimates
Table 8 shows that the BALED and IAELED methods outperform the RGLED-based
localisation method in terms of localisation accuracy. To be more specific, in a total of 64
position estimates for X, Z and their respective Y coordinates, 48.44% of positions had errors
above 10cm for the RGLED method, 39.54% of position estimates had errors above 10cm for
the BALED method whilst only 18.75% of position estimates had errors above 10cm when
employ the IALED-based localisation method.
In addition the advanced LE detection
methods resulted in position estimates with reduced STD thus confirming the advanced LE
detection methods are more accurate that the RGLED method.
.
121
The MAE for distance differences derived using all mentioned LE detection methods are
summarised in table 9.
Algorithm
RGLED
BALED
IALED
Small vessel
MAE(cm)
X
Z
9.21
3.3
2.9
8.76
4.23
3
Large vessel
MAE(cm)
X
Z
11.65
4.5
7.13
9.57
4.1
6.43
Table 9: Distance difference MAE
Table 9 supports the trends observed in the position estimate results, whereas once more, the
IALED method gives more accurate results. In fact, 46.88% of distance profiles derived from
the RGLED method had errors above 5cm. 31.25% of profiles from the BALED method had
errors above 5cm and only 18.75% of profiles from the IALED method had errors above
5cm.
On the whole, this chapter shown that the accuracy with which LEs are detected and the
resulting distance difference values determine to great extent the localisation accuracy. A
close examination of all distance difference values out of the IALED method showed that
these don‘t follow any specific trends with respect to the size of the observed errors. Distance
difference values for transmitters deep inside the small vessel, up to 90cm, shown to be as
accurate as distance difference values observed at small depths of 30cm. As far as tests in the
large vessel, the limited amount of data (only three test cases) can only serve as reference for
future work and cannot be used to provide solid conclusions; although it was observed that an
increased noise made LE detection and computation of distance difference values less
accurate.
Position estimates and distance differences achieved higher accuracies when using the
IALED method. As such IALED-based data are used to evaluate the novel localisation
methods presented in the following chapter.
.
122
CHAPTER 6
NOVEL LOCALISATION METHODS
6.1 Introduction
The work discussed in chapters 4 and 5 considered a positioning method that was based
solely on the geometry of circles. This chapter presents alternative localisation methods that
are based on analysis of experimental distance difference profiles and on mathematical
expressions used to derive the distance differences.
Methodical analysis of numerous theoretical and experimental distance difference profiles
revealed that these follow specific trends that are repeated throughout all test results and can
be used to derive the positions of the transmitting nodes.
This chapter presents a thorough analysis of 2D and 3D theoretical distance difference
profiles encapsulating the trends leading into the identification of the transmitting node‘s
position. All the observations are mathematically proven and incorporated into novel
localisation algorithms. Experimental distance difference profiles were introduced within the
new localisation algorithms to validate their accuracy.
6.2 Theoretical distance difference profile analysis
The prototype localisation method discussed in Chapter 4 essentially decomposed the 3D
localisation problem into 2D localisation problems in orthogonal planes. Hence, the
discussion of distance difference profiles starts with a consideration of the 2D case, which
provides a context for the analysis which applied to the 3D case.
6.2.1 Analysis of 2D theoretical distance difference profiles
Trends observed in the experimental distance difference profiles obtained using the
localisation approach discussed in Chapter 4, led to a simple theoretical investigation. Based
on observed profiles, a simple MATLAB model was developed to predict the distance
difference profiles that would be observed by a number of receivers situated at the top of a
2D vessel with a transmitting node placed at different locations beneath the receivers. This is
.
123
a subtraction of the measurements taken in the X (or Z) planes in the experiments reported in
Chapter 4.
In the results presented in the next section in all cases the reference receiving point was
situated at (X,Y, Z)=(0, 0,0) as illustrated in figure 104.
X0,Y0,Z0=(0,0,0)
(Xi,0i,Z)
(Xn,0n,Zn)
Reference
Rref
Rref
130cm
X, Z
Rn
Rn
Y
Tx node
(Xt,Yt,Zt)
100cm
Figure 104. 2D distance difference diagram
The following figures depict theoretical distance difference profiles taken at a number of
points along X axis. It is important to note that the same plots would have been resulted if
receiving points were situated along Z axis.
Figures 105 and 106 depict five distance difference profiles, taken at five different Y
positions at two fixed X transmitting positions of 28cm and 72cm respectively.
Figure 105. Profiles at Xt=28cm
Figure 106. Profiles at Xt=72cm
Figure 107 shows five distance difference profiles based on eleven receiving points with the
transmitter at X=50cm.
.
124
Figure 107. Profile at Xt=50cm
An examination of the distance difference profiles appearing in figures 105 to 107 reveals
that, in all cases, the minimum of all profiles approximates closely the position of the
transmitter along X (or Z) axis. In addition, it appears that, in the situations where the
distance difference profile becomes negative, the negative part of the curve is symmetrical
about the transmitter position. Finally, distance difference profiles seem to become flatter as
the depth of transmitter increases, for a fixed X transmitter coordinate.
6.2.2 Analysis of 3D theoretical distance difference profiles
The analysis presented here makes the transition from 2D distance difference profiles to 3D
profiles using a scenario where a process is enclosed in a cuboid shaped vessel. A sensor
whose position is to be computed is immersed within the process and is able to transmit data
to an array of receivers located in a measurement plane above the vessel. The 3D localisation
process within a cuboid volume is illustrated in figure 108.
Theoretical 3D distance difference values, D, are obtained from the familiar equation (44).
The position of the transmitter is (Xt,Yt,Zt) and the positions of the receiving points scattered
in a plane are (X,Y,Z) as seen in figure 108.
D  R  R ref
(44)
where
R 2ref  X 0  X t 2  Y0  Yt 2  Z 0  Z t 2

R 2  X  Xt
.
2  Y
 Yt
2  Z
 Zt
2
(45)
(46)
125
Without loss of generality, Y coordinates across the measurement plane are considered to be
at zero and the reference point of TDoA calculations was X=0cm,Y=0cm,Z=0cm.
Measurement
plane
(0,0,0)
Figure 108. 3D distance difference profile on a plane
A 3D MATLAB model, based on the scenario discussed above was developed and a number
of 3D plots of distance difference profiles over received antenna positions across X and Z
axis were produced (figure 109). In all cases it was observed that the 3D plots form surfaces
whose minimum coincides with the X and Z position of the transmitters. One such example is
illustrated in figure 109 that shows a 3D profile for a transmitter at (Xt,Yt,Zt)=(50,50,50).
Figure 109. Distance difference profile on a plane (Xt,Yt,Zt)=(50,50,50)
Likewise different transmitting positions result in 3D profile surfaces with the minimum
giving the X and Z positions as seen in figures 110 and 111.
.
126
Figure 110. Profile at Xt=20,Yt=50,Zt=80
Figure 111. Profile at Xt=80,Yt=50,Zt=20
The observations made with respect to the 2D model, about the symmetry of the negative part
of the distance difference curve and about the flattening of the profile curves, are repeated in
the 3D profiles as seen in figures 112 and 113.
Figure 112. Profile at Xt=80,Yt=50,Zt=20
Figure 113. Profile at Xt=50,Yt=10,Zt=50
Further analysis of several theoretical distance difference profiles, revealed that the position
of the transmitting node defines an axis of symmetry whereas any two negative values
resulting from equidistance, (Δχ), receiving points, from a transmitter position, Xt, result in
distance difference values that are equal between them.
A clear view of the distance difference symmetry about the axis, defined by the position of
the transmitting node, is seen in figure 114 where it is obvious that any two profile points that
are equidistant from the transmitter are equal between them.
.
127
Figure 114. 2D profile for transmitter at (Xt,Yt)=(50,50)
Theoretical proofs for the properties observed in the analysis of both 2D and 3D distance
difference plots are presented in the next section whilst a number of simple approaches to
localisation, based on these properties, are discussed in section 6.3 onwards. Note that the
properties discussed above were also observed in experimental distance difference profiles
seen in Chapter 5 and as such the localisation methods presented in this chapter were
evaluated using experimental data.
6.2.3 Properties of distance difference profiles
In this section, a number of properties of distance difference profiles are proved. Before the
analysis is presented, a recap of the physical situation that is modelled is now given. As
discussed in section 6.2.1, it is assumed that all the receiving antennas are located in a plane
that lies immediately above the vessel. This plane is called the measurement plane and it
contains the origin of coordinates and the X and Z coordinates as seen in figure 108. The Y
axis is measured perpendicular to this plane, positive into the vessel. See figure108.
What follows are proves of a number of theorems derived based on trends in TDoA profiles
discussed in the previous two sections.
Theorem 1: The minimum of the distance difference surface recorded in the measurement
plane is located at the X, Z coordinates of the transmitter.
Proof: The above theorem may be proved by taking the first and second derivatives of
equation (44) as follows:
.
128
D 
D D D


X Y Z
2D 
 2D
X 2

 2D
Y 2

(47)
 2D
(48)
Z 2
In order to determine the stationary points in the 3D surface the first order gradient in
equation (47) is set to zero and solved for X and Z as follows:
2  X  Xt   0
(49)
2  Z  Zt   0
(50)
Equations (49) and (50) indicate that stationary points exist when X=Xt and Z=Zt. In order to
determine if the stationary points represent a minimum or a maximum or a saddle point, the
second derivatives of equation (44) must be considered as in equations (51) and (52):
  D 

Xn  X t 
D

X  X  X
D
(51)
2
 2D
 D 

 1
X 2  X 
(52)
At Xt=Xn
D
0
X
(53)
And so equation (52) becomes
 2D
X 2

1
D
(54)
A similar equation to (54) exists for the second derivative with respect to Z.
A stationary point is a maximum if the inequality in equation (55) is true [112]:
  0 and
2D
X 2
 0 or
2D
Z 2
0
(55)
Similarly, equation (56) states that a stationary point is minimum if and only if the following
is true [112]:
  0 and
.
2D
X 2
 0 or
2D
Z 2
0
(56)
129
where

 2 D(X, Z)
(XZ) 2

 2 DX, Z  2 DX, Z

X 2
Z 2
(57)
In equation (57) D is always positive thus stationary points across both X and Z axes are
always positive. Further, Δ in equation (57) is always negative given that
 2 D(X, Z)
 (XZ) 2
0.
Therefore, given the two necessary conditions for a stationary point to be minimum are
satisfied in equation (56), it can be concluded that distance difference distribution over the
measurement plane has a minimum at the position of the transmitter i.e. at (Xt, 0, Zt).
Theorem 2: The distance difference surface observed in the measurement plane for which
distance differences are negative, is symmetrical about the line Xt =0, Zt=0.
Proof: Consider the distance difference Dt+Δ at a point (Xt+Δx, 0, Zt + Δz):
D t  
X t   x  X t 2  Yt 2  Z t   z  Z t 2

X 0  X t 2  Yt 2  Z 0  Z t 2
(58)
Similarly, the distance difference Dt-Δ at a point (Xt-Δx, 0, Zt - Δz) is:
D t  
X t   x  X t 2  Yt 2  Z t   z  Z t 2

X 0  X t 2  Yt 2  Z 0  Z t 2
(59)
Hence:
Dt+Δ= Dt-Δ
(60)
for any values of Δx and Δz.
Note, however, that distance differences will vary along circles centred on the transmitter
position as discussed in the following theorem.
Theorem 3: Curves of constant distances difference are circles in the measurement plane,
centred on (Xt, 0, Zt).
Proof: Consider the distance difference, D, at a position (Xt+Rcosθ, 0, Zt+Rsinθ) where angle
θ and distance R are seen in figure 115. Figure 115 shows the contour plot of distance
.
130
difference profiles where all distance differences are circles centred on the position of the
transmitter in this case (Xt,Zt)=(50,50).
R
θ
Figure 115. Contour plot for (Xt,Zt)=(50cm,50cm)
The distance difference equation becomes as follows:
D
X t  Rcosθ  X t 2  Yt2  Z t  Rsinθ  Z t 2

X 0  X t 2  Yt2  Z 0  Z t 2
(61)
Simplifying equation (61) gives:
D  R 2  Yt2 
X 0  X t 2  Yt2  Z 0  Z t 2
(62)
In equation (62) distance difference is independent of θ, and constant for constant R, and
hence curves of constant distance difference in the measurement plane are circles centred on
(Xt, 0, Zt) for a fixed transmitter and reference antenna position.
Corollary 2.1: In the measurement plane, distance difference increases with distance R from
(Xt, 0, Zt).
Proof: Consider two points at distance R1 and distance R2 from point (Xt, 0, Zt) in the
measurement plane. Assume R2> R1. Then by equation (63):
DR1 -DR 2 = R 22 +Yt - R12 +Yt
(63)
This is positive and so distance difference increase with distance R from (Xt, 0, Zt).
Theorem 4: The points of zero distance difference in the measurement plane lie on a circle of
radius K, where
.
131
K=
 X0 -X t  +  Z0 -Zt 
2
2
(64)
For fixed transmitter and reference antenna positions:
 X-Xt  +  Z-Zt 
2
2
=K 2
(65)
where K 2   X0 -X t  +  Z0 -Zt 
2
2
Corollary 4.1: The points of zero distance difference on the X and Z axes are:
XD=0 =Xt ±K and Z=Zt ±K
Corollary 4.2: K is the projection on the measurement plane from the reference node to the
transmitter.
Theorem 5: Distance difference profiles flatten (i.e. variations amongst distance difference
values reduces) as the depth of the transmitter inside a vessel increases.
Proof: As has already been demonstrated, distance difference is given by:
D=
 X-X t 
2
+(Yt )2 +  Z-Zt  2
 X0 -X t 
2
+(Yt )2 +  Z0 -Zt 
2
(66)
Consider a fixed, finite values Xt, Zt, X0, Z0 and X, Z positions within the measurement plane
directly above the vessel.
As Yt increases, the difference between the two square root terms will be reduce and
therefore the distance difference curves become flatter (reduce in height) as transmitter depth
Yt increases. As Yt  ,D  0 .
Corollary 5.1: The value of the maximum distance difference at X=Xt , Z=Zt in equation (66)
given by:
D  Yt  K 2  Yt2
(67)
where K is defined in equation (64)
The effect of flattening of the distance difference when Yt increases is illustrated in figure
116 for Xt=50cm Zt=0cm.
.
132
Figure 116. Maximum distance difference Vs depth
Figure 116 implies a problem with establishing the Y coordinate of sensor nodes buried deep
inside a vessel. From the above graph it can be seen that the peak distance difference varies
between 50cm and 15cm over a depth of the first 50cm, but subsequently distance difference
varies between 15cm and about 10cm. This suggests that performing depth estimation using a
single TDoA value is likely to result in increased errors since it will be difficult to distinguish
a unique Yt especially in the presence of errors.
Theorem 6: Each transmitter position within the vessel gives rise to a unique distance
difference profile in the measurement plane.
Proof: Consider the distance difference equation
D=
 X-X t 
2
+(Yt )2 +  Z-Zt  2
 X0 -X t 
2
+(Yt )2 +  Z0 -Zt 
2
(68)
Consider the 2D case localisation with a transmitter and received positions as illustrated in
figure 117.
(Xt,-Yt)
Reference
position
Receiver position
Above vessel
X
Inside vessel
R
Rref
(Xt,Yt)
Transmitter
position
Y
Figure 117. Illustration of distance profile uniqueness
.
133
When the transmitter occupies the position shown, in figure 117, a distance difference
function is produced along the X axis, i.e. a distance difference value exists for each point X,
which is known to have a minimum at X=Xt (Theorem 1). The position of the zero crossings
of the curve is also defined by the transmitter position (Theorem 4). Then again consider that
the transmitter position is changed by a small amount δX to (Xt +δX, Zt). The distance
difference profile produced is very similar to the previous profile, but displaced by δX i.e the
minimum of the distance difference curve us now located at Xt +δX, the lower zero crossing
is still 0, but the upper zero crossing is now 2(Xt +δX). Hence the second distance difference
profile is different from the first.
From the above arguments it can be concluded that the distance difference profiles arising
from a transmitter at a fixed depth Yt , but at every possible X position are unique.
Now consider that the transmitter position is changed by a small amount Δy to (Xt , Yt +ΔY).
This results in a distance difference profile with a minimum in the same transmitter position
as the original case. However, from Theorem 3, it is clear that the height of the distance
difference curve of the profile resulting from a transmitter at (Xt , Yt +ΔY) is smaller than that
corresponding to the original case. Again the two curves are distinct.
From the argument of the previous paragraph, profiles originating from transmitters, at a
fixed Xt coordinate but different Yt coordinates, are distinct.
The above arguments generalise to the 3D case in the obvious manner. Hence, it can be
concluded that each unique transmitter position within the vessel gives rise to a unique
distance difference profile.
It should be noted that Theorem 6 applies only to transmitter positions within the vessel.
From the distance difference equation and from figure 117, it is clear that a transmitter
located at (Xt , -Yt ) i.e. above the vessel, would produce the same distance difference profile
as the transmitter located at (Xt , Yt ).
.
134
6.3 Novel localisation algorithms
The following sections present localisation methods based on the information acquired from
analysis of distance difference profiles presented in the previous three sections. All methods
were incorporated into 2D algorithms since TDoA results were derived based on an algorithm
that treated X and Z readings independently as discussed in Chapter 4. Note that some
algorithms implement a straight forward implementation of the mathematics and the theory
defining each method whilst other algorithms introduce certain variations to enhance
localisation accuracy.
All localisation methods were executed in MATLAB and evaluated using experimental
distance difference profiles derived from the IALED method discussed in chapter 5. In
addition, a number of localisation methods employ theoretical distance difference profiles
derived based on the configuration discussed in section 6.2.1 and specifically figure 104.
6.4 Basic profile positioning method
The analysis of distance difference profiles, presented in the previous section, shows these
have a minimum at the position of the transmitter. Hence it is possible to estimate the X and
Z coordinated of the transmitter position from a set of measurements taken at a variety of
locations over the measurement plane Y=0. However, this approach does not provide
information about the Y coordinate of the transmitter. Nevertheless, since the Xt and Zt
coordinated can be estimated and distance differences are measured, then an approximate
value for Yt can be obtained.
Consider equation (theorem 3). If this is evaluated at (Xt, 0, Zt) (recall that the reference
position is at (0,0,0)), then it becomes:
D  Yt  X 2t  Yt2  Z 2t
(69)
where Dt corresponds to the minimum value of the distance difference profile along X axis
Rearranging and squaring both parts of equation (69) gives the following:
D 2  Yt2  2DYt  X 2t  Yt2  Z 2t
(70)
Finally, solving equation (70) for Yt results in the following:
.
135
Yt 
D 2t  (X 2t  Z 2t )
2  Dt
,
Yt  
D 2  (X 2t  Z 2t )
2  Dt
(71)
Note that although equation (71) indicates there are two estimates for Y, one positive and one
negative, for any X or Z position, only the positive solution is physically realistic due to the
arrangement of the measurement system. To be more specific, given that the vessel is isolated
in the positive Y part of the coordinate space and the transmitter is located within the vessel,
then the positive solution for Yt is the correct one.
Note that Yt estimates using the approach discussed above are clearly sensitive to errors in
the measurement of distance differences and in the Xt and Zt estimates.
6.4.1 MATLAB algorithm
The principles of the simple profile-based localisation method were incorporated in a
MATLAB program. The program treats the X and Z axes independently and uses the minima
in the distance difference profiles to deduce the X, Y and Z coordinates of the transmitter.
The basic method STRAIGHT-FORWARD (BMSF) was examined using a number of
theoretical distance difference profiles. Here it is important to note that, in all cases, the
algorithm choose the X or Z position of the transmitter to be at the receiving antenna position
corresponding to the minimum distance difference profile value.
In the example seen in figure 118a, the BMSF algorithm chooses the position of the
transmitter to be at 50cm and used equation (71) to compute the Yt at 30cm, as expected.
(a) X=50cm, Y= 30cm
(b) X=44cm, Y= 30cm
Figure 118. Theoretical profiles
.
136
Additional plots at different transmitter positions revealed certain weaknesses in the BMSF
algorithm. Figure 118b shows a profile taken at transmitter position of X=44cm and Y=30cm.
In this case, BMSF choose the transmitter position to be at X=40cm, Y=23.31cm resulting in
an error of 6.69cm.
Another situation where the BMSF algorithm results in errors is when more than one distance
difference profiles have the same minimum value. This is exemplified in the example seen in
figure 119 where the highlighted profile values at receiving antenna at 40cm and 50cm are
the same. In such cases the BMSF algorithm simply processes the first minimum value and
the corresponding receiving antenna. The estimated positions for the profile seen in figure
119 were Xt=40cm and Yt=21.96 as opposed to the expected position of Xt=45cm, Yt= 30cm.
Figure 119. Profile for Xt=44cm, Yt= 30cm
It is clear that the accuracy of the BMSF is limited by the number of the receiving antenna. In
any case, in real applications, it is anticipated that the number of receiving antennas will be
limited and as such other methods need to be considered to improve the accuracy of the
BMSF method. To deal with problems in the BMSF algorithm, as discused above, two
modified versions of the BMSF method were developed.
The first approach considered interpolating a least squares best fit of the profiles to increase
the resolution of the distance different profiles with respect to the receiving antenna. This
method is called basic method least squares (BMLS). When the BMLS method was
implemented for the data seen in figures 120 and 121 the accuracy improved significantly as
opposed to the BMSF method. The BMLS resulted in X=44cm, Y=31.71cm for data in figure
120 and X=45cm, Y=31.81cm for data in figure 121. Absolute errors in these cases were
about 1cm as opposed to the basic method were errors reached up to 6cm.
.
137
Figure 120. Profile for X=44, Y=30
Figure 121. Profile for X=45, Y=30
A second approach considered using the first derivative of the equation giving the least
square curve best fitted into the experimental data to find the minimum of the profiles. This
method was named basic method least squares derivative (BMLSD).
Experimental evaluation of three basic localisation methods BMSF, BMLS, and BMLSD is
discussed in the following sections. A reminder that the terms Absolute Error (AE) refer to
the absolute results of the subtraction of the experimental from the measured values and
Mean Absolute Error (MAE) denotes the mean of the AE for all distance differences or
position estimates for X or Z axis accordingly. As in the previous two chapters the AE are
shown in histograms whilst the position estimates are plotted on scale against the measured
positions.
6.4.2 BMSF experimental evaluation
First of the basic localisation algorithms to be examined is the BMSF. Figure 122 shows the
measured positions of the transmitting nodes as opposed the estimated positions along X axis.
(a) Small vessel tests
(b) Large vessel tests
Figure 122.BMSF positions along X axis
.
138
The AEs from the implementation of the BMSF approach for data along X axis are shown in
figure 123.
BMSF
Absolute error (cm)
Absolute error (cm)
BMSF
(a) X axis coordinates error
(b) Yx axis coordinates error
Figure 123. X axis coordinates error
Figure 124 shows the measured positions and the estimated positions along Z axis.
(a) Small vessel tests
(b) Large vessel tests
Figure 124. BMSF positions along Z axis
Figure 125 show the AEs for Z position estimates.
BMSF
Absolute error (cm)
Absolute error (cm)
BMSF
(a) Z axis coordinates error
(b) Yz axis coordinates error
Figure 125. Z axis coordinates error
.
139
The MAE for BMSF for all positions along both X axis and Z axis as well as the respective
STDs are summarised in table 10.
Small vessel tests
MAE(cm)
STD(cm)
Large vessel tests
X
Yx
Y
Yz
Z
X
Yx
Y
Yz
Z
9.63
12.5
27.96
40.86
37.2
51.3
44.42
82.91
9.42
7.51
30
23.64
76.3
93.2
124.9
89.31
173.5
173.8
15
10
Table 10: MAE and STD for the BMSF
In general, the AEs for position estimates in the BMSF method showed a small increase
compared to the AEs in the IALED-based localisation method. The MAE for X and Z
positions when using the BMSF (table 10) was about 4cm higher than the one recorded for
the IALED-based method, in table 6 (pg118), for both small and large vessel tests. In
addition, the MAEs for Y estimates out of the BMSF were about five times larger than the
ones observed from the implementation of the prototype localisation algorithm using IALED.
6.4.3 BMLS experimental evaluation
The second basic algorithm to be examined is the BMLS. The position estimates and the
localisation accuracies are graphically displayed in the following figures.
Figure 126 shows the measured positions of the transmitting nodes and the estimated
positions along X axis.
(a) Small vessel tests
(b) Large vessel tests
Figure 126. BMLS measured and estimated X axis positions
The histograms showing the AEs for X position estimates are shown in figure 127.
.
140
BMLS
BMLS
(a) X axis coordinates error
(b) Yx axis coordinates error
Figure 127. X axis coordinates error
Figure 128 shows the measured positions and the estimated positions along Z axis.
(a) Small vessel tests
(b) Large vessel tests
Figure 128. BMLS measured and estimated Z axis positions
The AEs for position estimates along Z axis are shown in figure 129.
BMLS
BMLS
(a) Z axis coordinates error
(b) Yz axis coordinates error
Figure 129. Z axis coordinates error
.
141
The MAE for the BMLS, for all positions along both X axis and Z axis as well as the
respective STDs are shown in table 11.
Small vessel tests
MAE(cm)
STD(cm)
Large vessel tests
X
Yx
Y
Yz
Z
X
Yx
Y
Yz
Z
4.31
3.42
7.14
7.42
10.3
5.65
13.41
8.2
4.62
3.243
37.57
29.9
64.31
44.03
39.94
28.72
15.6
19.61
11.91
17.61
Table 11: MAE and STD for the BMLS method
Results from the implementation of the BMLS, using experimental data from the small
vessel, showed a reduction in the MAEs for X and Z positions of about 5cm compared to the
BMSF. A more noticeable reduction in the MAE, reaching 20cm, was observed in the
computation of the Y estimates for small vessel tests (table 11). Similar trends were observed
for data out of large vessel tests for Z positions and especially for Y positions where the
MAE was reduced by a staggering 85cm compared to the ones in table 10.
6.4.4 BMLSD experimental evaluation
The final basic localisation approach to be examined is the BMLSD. The following figures
depict the localisation results for the BMLSD method.
Figure 130 shows the measured positions of the transmitting nodes as opposed the estimated
positions, using BMLSD, along X axis.
(a) Small vessel tests
(b) Large vessel tests
Figure 130. BMLD measured and estimated X axis positions
The AEs for X position estimates are shown in the histograms depicted in figure 131.
.
142
BMLSD
BMLSD
(a) X axis coordinates error
(b) Yx axis coordinates error
Figure 131. X axis coordinates error
Figure 132 shows the measured positions as opposed the estimated positions along Z axis.
(a) Small vessel tests
(b) Large vessel tests
Figure 132. BMLD measured and estimated Z axis positions
The AEs for Z position estimates using BMLSD are shown in figure 133.
BMLSD
BMLSD
(a) Z axis coordinates error
(b) Yz axis coordinates error
Figure 133. Z axis coordinates error
.
143
The MAE for the BMLSD, for all positions along both X axis and Z axis as well as the
respective STDs are shown in table 12.
Small vessel tests
MAE(cm)
STD(cm)
Large vessel tests
X
Yx
Y
Yz
Z
X
Yx
Y
Yz
Z
5.56
6.2
13.13
11.65
12.72
8.82
12.3
9.8
4.71
3.33
37.7
28.15
90.31
78.52
87.58
64.2
84.86
68.85
15
19.1
Table 12: MAE and STD for the BMLSD
The MAEs in table 12 are similar to the ones observed in table 11 for the BMLS method.
This was expected since both the BMLS and the BMLSD methods are based on the
derivation of the minimum out of the least square best fit curve.
A characteristic example of the improvements in localisation accuracy, in the basic method,
when using the least square best fit, can be seen in figure 134.
Figure 134. Basic localisation Z=65cm, Y=30cm
Figure 134 shows the distance difference profile for the experiment number 11. The BMSF
method gives a position estimate Zt at 80cm (minimum of the experimental profile) giving an
error of 15cm (AE shown in figure 125) and Yt=50cm with an error of 20cm (AE shown in
figure 125). On the other hand a BMLSD implementation for the same data resulted in a Z
position of 73cm giving an error of just 8cm and Yz of 35.1cm with an error of only 5.1cm
the BMLSD (AEs shown in figure 133).
.
144
6.5 Centroid based localisation
The centroid based localisation method is essentially an advanced variation of the basic
method described in section 6.4 and deals effectively with the problem arose from the use of
a single receiving antenna position as the position of the transmitter.
In section 6.2.2 it was observed that distance difference values resulting from equidistantly
separated receiving antenna, from a transmitter, give the same distance difference data. This
statement can be used in reverse to allow for the position of the transmitting node to be
estimated. In fact any two distance difference values that are equal to each other reveal that
their respective receiving points are equidistant from the transmitter. As such, it is possible to
take the centroid of those particular receiving points to find the position of the transmitting
node along X and Z axis.
In an example shown in figure 135 a transmitter is located at X=50cm, Y=50cm. Distance
difference points D1, D2 and D3 are equally distant from the transmitter as points D6,D5 and
D4 respectively.
D1
D6
D5
D2
D3
D4
Figure 135. Equidistance receiving antenna positions
The mean of the distance difference points, in figure 135, along receiving antenna position
axis, D1D6(0+100/2), D2D5(20+80/2) and D3D4 (40+60/2) all point to the exact position of the
transmitter. The centroid of all mean positions gives a unique value that is the position of the
transmitting node.
Two versions of the centroid method were incorporated into two algorithms and tested using
all available experimental data.
.
145
The first version implemented the exact mathematical expressions presented in the previous
section; this was called basic centroid localisation (BCL). Nevertheless, it was observed that
the presence of distance profile points out of the axis of symmetry, centred on the transmitter,
can cause increased errors with positioning errors increase as the number of distance profile
points out of the axis of symmetry increases.
The following figures shows how the presence of distance profile points out of the axis of
symmetry can affect localisation accuracy. In these figures the term original data refers to the
raw experiment distance difference data, out of the IALED method, whilst least squares
refers to the fit of least squares into those raw data. Note that the flat green line in figure 136
indicates a function in the BCL algorithm that forces the best fit to terminate at zero axis.
This allows processing data that are below the zero axis and are part of the symmetrical curve
thus enhancing localisation accuracy.
A
B
H
C
D
E
F
G
(a) Xt=40cm,Yt=89cm
(b) Xt=30cm, Yt=89cm
Figure 136. BCL plots
To improve accuracy, the BCL approach considered taking the centroid of those receive
antenna positions corresponding to pairs of distance difference values that resulted in a
subtraction result (between them) that was less than 2cm. In figure 136a, only pairs CG, DF,
DE, DF and EF had a difference less than 2cm in between them, thus only those were
considered to be equidistant from the transmitter. As such, points A, B and H were excluded
from the localisation process.
To further deal with problems arose from the use of the BCL method; two approaches were
incorporated into another algorithm named advanced centroid localisation (ACL) method. In
the first approach, only distance profile points with negative sign are considered in the
localisation process. This effectively deals with problems such as those illustrated in figure
.
146
136b whereas an increased number of positive profiles points exceed the desired transmitter‘s
axis of symmetry profile values. The second approach implements a least squares best fit
approximation along with interpolation of numerous points into the existing distance profile
points thus forming a new curve that always ensures crossings with the zero axis. This deals
with localisation problems rose from cases where distance profile points do not cross the zero
axis such as the case seen in figure 136a.
6.5.1 MATLAB algorithm
The above statements were incorporated into a mathematical expression seen in equations
(72) and (73). Any two distance difference values, D, out of an n receiving antenna positions
minimising equation (72) indicate that these are equidistant from the transmitter.





F X in , X nj , Z in , Z nj   arg min D nXi  D nX j
  Xi ,X j
 , arg minD
2
 
Zi , Z j
n
Zi
 D nZ j
  
2

(72)
The mean centroid of all positions minimising equation (72) over an n number of distance
difference values, were used in equation (73) to find the final X and Z positions of the
transmitting node:
X, Z  1
n

 X in  X nj Z in  Z nj

,

2
2





(73)
Finally the centroid X and Z position estimates are substituted in equation (71) to compute
the Y coordinates of the transmitter. However, equation (71) requires a value for distance
difference as well. As therefore, the mean of the distance difference values corresponding to
respective equidistance receiving X or Z points from the transmitter, was used as follows:
 D nX  D nX D nZ  D nZ
j
i
j
D X , D Z    i
,
2
2






(74)
6.5.2 BCL experimental evaluation
The accuracy of the BCL method is examined using the experimental distance difference
values derived from tests in the small and in the large testing vessels.
Figure 137 shows the measured positions as opposed the estimated positions along X axis.
.
147
(a) Small vessel tests
(b) Large vessel tests
Figure 137. Measured and estimated X axis positions
The AE for data along X axis are shows in the histograms in figure 138.
BCL
BCL
(a) X axis coordinates error
(b) Yx axis coordinates error
Figure 138.X axis coordinates error
Figure 139 shows the measured positions of the transmitting nodes along Z axis.
(a) Small vessel tests
(b) Large vessel tests
Figure 139. Measured and estimated Z axis positions
.
148
The MAE for the implementation of the BCL approach for data along Z axis are graphically
shown in figure 140.
BCL
BCL
(a) Z axis coordinates error
(b) Yz axis coordinates error
Figure 140. Z axis coordinates error
The MAE for the BCL method, for all positions along both X axis and Z axis as well as the
respective STDs are shown in table 13.
Small vessel tests
MAE(cm)
STD(cm)
Large vessel tests
Xt
Yxt
Yt
Yzt
Zt
Xt
Yxt
Yt
Yzt
Zt
6.9
5.8
27.45
21.15
22.83
15.6
18.2
17.61
5.94
6.52
24.67
28.04
64.96
17.3
89.67
69.9
114.4
123.64
15
17.32
Table 13: MAE and STD for the BCL method
The BCL method appeared to be equally ineffective as the BMSF approach, especially when
concerns Y estimates, with MAE reaching 20cm and 90cm for tests in the small and in the
large vessels receptively. This degradation in the computation of Y was caused by the fact
that certain values within the distance difference profiles were excluded from the localisation
process. The exclusion of a number of distance difference values from an already limited
amount of eight or five values inevitably reduces the probabilities for an accurate localisation
result. A similar condition with increased errors in Y estimates was observed in the BMSF
method.
6.5.3 ACL experimental evaluation
Figure 141 shows the measured positions of the transmitting nodes as opposed the estimated
positions along X axis for the ACL method.
.
149
(a) Small vessel tests
(b) Large vessel tests
Figure 141. Measured and estimated X axis positions
The AE for data along X axis are graphically shown in figure 142.
ACL
ACL
(a) X axis coordinates error
(b) Yx axis coordinates error
Figure 142. ACL X axis coordinates error
Figure 143 shows the measured positions as opposed the estimated positions along Z axis.
(a) Small vessel tests
(b) Large vessel tests
Figure 143. Measured and estimated Z axis positions
The AE for Z position estimates are shown in figure 144.
.
150
ACL
ACL
(a) Z axis coordinates error
(b) Yz axis coordinates error
Figure 144. ACL Z axis coordinates error
The overall MAE for the ACL method, for all positions along both X axis and Z axis as well
as the respective STDs are shown in table 14.
Small vessel tests
MAE(cm)
STD(cm)
Large vessel tests
X
Yx
Y
Yz
Z
X
Yx
Y
Yz
Z
7.1
6.7
8.13
7.7
9.24
5.54
10.35
8.3
7.91
7.2
28.6
19.95
62.97
29.87
66.72
43.1
70.46
58.21
16.5
20.43
Table 14: MAE and STD for the ACL method
The ACL algorithm resulted in X and Z positions with a slight increase in MAE, up to 3cm,
from the results in BCL method. This was not unexpected since the two algorithms followed
different approaches in estimating the position of the transmitter. Nevertheless, the ACL
method performed much better than the BMSF approach and provided as accurate results as
the BMLS and the BMSD methods.
Of interest is the significant improvement in the MAE for the Y estimates, up to 13cm for the
small vessel and up to 20cm for the large vessel tests, which is achieved when using the ACL
method. This improvement is attributed to the ability of the ACL method to include a
significant higher number of distance difference values (using interpolation into least
squares) when estimating the centroid thus dealing with possible outliers that cause errors in
the BCL method.
The following two figures show plots of experimental distance differences over the receiving
antenna positions when using BCL (original data) and the ACL method (least square).
.
151
(a) Basic centroid localisation Z=43cm
(b) Centroid best fit basic Z=43cm
Figure 145. Basic centroid localisation Zt=43cm, Yt=92cm
Figure 145a shows a basic localisation result where the minimum of the distance difference
profile resulted in a Z position of 45cm giving an error of 2cm and Yx of 100.8cm with an
error of 8.8cm. On the other hand the least squares best fit, shown in figure 145b, resulted in
a minimum of 43cm giving an error of just 0cm and Yx of 91.8cm with an error of 0.2cm.
6.6 Comparative positioning method
The comparative localisation method is different from the methods discussed so far, in this
chapter. First it adopts an iterative approach and second it provides Y estimates
independently of X and Z estimates unlike all the methods discussed so far.
The comparative method followed an iterative approach whereas numerous theoretical
distance difference profiles, DT, were computed over the testing area and then compared to
the experimental distance difference profile, DE, in a much up process to identify the position
of transmitting node. The X, Z and Y positions resulting in a theoretical distance difference
profile minimising the least square approximation in equation (75) were taken as the final
positions of the transmitter.
X
n
n
i ,Zj
  arg min D
Xi , Z j
E in
 D T n 
j 
2
(75)
The operating principle of the comparative method was based on the fact that the
measurement arrangement implied that for different transmitter positions there will always be
a different distance different profile. This uniqueness of the profiles was used as a fingerprint
to allow for the positions of the transmitter to be derived.
.
152
6.6.1 MATLAB implementation
An initial study considered a localisation approach whereas the whole vessel is divided into a
grid of 1cm2 squares covering all of its area. Here it is assumed that unknown nodes are
placed on the grid (figure 146a) and theoretical distance difference readings are recorded
along each and all unknown nodes with each and all of the receiving antennas. The above
procedure gives a number of distance difference profiles, covering every possible
combination between a transmitter and a receiver.
(a)
(b)
Figure 146. Comparative positioning diagrams
In order to reduce the number of iterations required in the initial brute force approach a new
method was developed. This method builds on the basic profile method whereas the X and Z
coordinates of the transmitting nodes are given based on the minimum of the profiles. Based
on the approximations of X and Z coordinates the comparative methods looks in an area
covering ±10cm from initial X and Z estimates. Simultaneously the algorithm looks for the Y
positions of the transmitter.
The algorithm compares the experimental distance profile with numerous theoretical distance
profiles within an area covering 10cm from either side of X or Z estimates as well as
covering a vertical area of the container just below the estimated X or Z positions as seen in
figure 146b. The choice for a look-up area of 10cm on either side from X or Z estimates was
based on the analysis of experimental results in the previous chapter. To be more specific the
overall localisation error for positions along X and Z axis in chapter 5 was less than 10cm.
.
153
The ±10cm covering area was chosen in order to increase the probability for the comparative
method to give a good approximation of the transmitter position.
6.6.2 Experimental evaluation
Figure 147 shows the measured positions of the transmitting nodes as opposed the estimated
positions along X axis for the comparative method.
(a) Small vessel tests
(b) Large vessel tests
Figure 147. Measured and estimated X axis positions
The AEs from the implementation of the comparative approach are shown in figure 148.
(a) X axis coordinates error
(b) Yx axis coordinates error
Figure 148. X axis coordinates error
Figure 149 shows the measured positions as opposed the estimated positions along Z axis.
.
154
(a) Small vessel tests
(b) Large vessel tests
Figure 149. Measured and estimated Z axis positions
The AE for data along Z axis are graphically shown in figure 150.
(a) Z axis coordinates error
(b) Yz axis coordinates error
Figure 150. Z axis coordinates error
The MAEs for the comparative method, for all positions along both X axis and Z axis as well
as the respective STDs are shown in table 15.
Small vessel tests
MAE(cm)
STD(cm)
Large vessel tests
X
Yx
Y
Yz
Z
X
Yx
Y
Yz
Z
4.13
3.3
8.2
6.7
9.81
6.15
11.44
8.3
5.4
5.1
51
24.64
5.66
3.8
18
21.86
30.33
40.1
5
0
Table 15: MAE and STD for the comparative method
The comparative method is the only one, from the seven new approaches discussed in this
chapter, that follows an iterative approach towards estimating the position of the transmitting
node. Nevertheless, although the average execution time for a single run of this algorithm
(~3s) is 20 times the average execution time for the rest of the algorithms (~0.15s); the
.
155
accuracies observed when using the comparative method are favourably comparable with the
rest of the methods especially when regards to the MAE for the large vessel tests.
6.7 Analytical localisation method
The analytical localisation method unlike the previous methods is based solely on the
solution of the equations giving the distance difference profiles. This new localisation method
essentially computes the position of the transmitter through a derivation of closed form
solution of sets of non-linear equations. Equations consist of known (from experiments)
distance difference values, Di, Dj, the coordinates of a reference receiving antenna (X0, Y0 ,
Z0) and the coordinates (Xn, Yn, Zn) of subsequent n receiving antennas as seen in figure 151.
Reference receive antenna
(X, Y,Z)=(0,0,0)
D
Figure 151. Receive and transmit position coordinates
For an arbitrary pair of receiving antenna, the non-linear equation relating distance difference
data to the coordinates of both of the transmitting node and the receivers is given by the
familiar equation (75). Here it is important to note that the measuring arrangement discussed
in chapter 4 imposed that TDoA and consequently distance difference data are derived in 2D,
i.e. there are always two sets of TDoA measurements one for X axis and another for Z axis.
For the analytical method the familiar distance difference equation in 2D is given as follows:
D xi 
X   X t 2  Y  Yt 2  X 0  X t 2  Y0  Yt 2
(76)
where ζ is either i or j
.
156
In equation (76) there are two unknowns, the coordinates of the transmitting node (Xt, Yt),
given that parameter Dxi is derived from the experiments whilst the coordinates of the
receiving antennas are measured by the user. Solving equation (76) for Xt and Yt respectively
will result in the location of the unknown node. Note that analysis proceeds with presentation
of X axis equations since these will be the same for Z axis as well.
As in all localisation methods presented in chapter 4 onwards the measurement plane on
which receiving antenna lie, seen in figure 151, is taken as Y=0 thus in all cases Yi=0.
Further the reference antenna (seen in figure 151) is located at the edge of the testing vessel
thus defining the origin of the coordinates i.e. (X0,Y0,Z0)=(0,0,0). As therefore, with X0, Y0
and Yi, in equation (76), all set to zero, equation (76) can be solved for Xt , for ζ=i as follows:




D
 2
xi
2
 D xi  X i  2  D xi 
2








Xt  




D
 2
2
 xi
D

X

2

D
i
xi
 xi
 2







Xi 

2  Xi
Xi 

2  Xi
X i  D 2x i  4  Yt2  


D 2x i  X i2

2






X i  D 2x i  4  Yt2  


D 2x i  X i2

2






(77)
Similarly, solving equation (76) for Yt , for ζ=j, with both X0 and Y0 set to zero gives:
D x j  X t 2  Yt 2 
X j  X t 2  Yt 2
(78)
Squaring both sides of equation (78) gives:
D 2x j  2  D x j  X t 2  Yt 2  X t 2  Yt 2  X j 2  2  X j  X t  X t 2  Yt 2
(79)
Simplifying (79), rearranging and squaring results in:



4  D 2x j  X t 2  Yt 2  X j 2  X j  2  X t
2  2  D 2x
j


 X j  X j  2  X t  D 4x j
(80)
Rearranging equation (80) gives:
4  D 4x j  Yt 2  X j 4  4  X 3j  X t  4  X 2j  X 2t  2  D 2x j  X 2j  4  D 2x j  X j X t  D 4x j  4  D 4x j  X t 2
.
(81)
157
Solving equation (81) for Y results in:


 
 
 
  D  X    D  X    D  X  2  X    D  X  2  X  
x
j
x
j
x
j
t
x
j
t





 
 
j
 
j
j
j
 
 
 

2  Dx


j


Yt  










D  X D  X D  X  2X D  X  2X  
j  x j
j  x j
j
t   xj
j
t

 xj
 
 
 

 


2  Dx
j


(82)
Substituting Yt from equation (82) to equation (77) eliminates Yt:
 D x  X 2j  D  D 2x  D 2x  D x  D x  X i2 
xi
j
i
j
j
 i



2  D xi  X j  2  D x j  X i
Xt   2

2
2
2
2
 D xi  D x j  D xi  D x j  D xi  X j  D x j  X i 


2  D xi  X j  2  D x j  X i


(83)
Equation (83) contains only known quantities thus allowing Xt to be calculated. A similar
approach to the derivation of equation (83) results in equation (84) that gives the Z
coordinates of the transmitter.
 D z  Z 2j  D  D 2z  D 2z  D z  D z  Z i2 
zi
j
i
j
j
 i



2  D zi  Z j  2  D z j  Z i
Zt   2
2
2
2
2
 D zi  D z j  D zi  D z j  D zi  Z j  D z j  Z i 


2  D zi  Z j  2  D z j  Z i


(84)
Once Xt and Zt, for each experiment, were derived, these were substituted in equation (82) to
derive Yt estimates.
Solving sets of non-linear equations as described above results in multiple Xt, Zt and Yt
estimates. In fact there are as many solutions as the number of the receiving antennas. To
derive single positions out of a k number of Xt, Yt and Zt values these were substituted back to
equation (76) to derive analytical-based distance differences Danl that are directly comparable
to the original distance differences, Dorig, giving the analytical localisation results in the first
place. As therefore a least square approximation, in equation (85), was used whereas the Xt,
Yt and Zt coordinates, from the analytical algorithm, that produce Danl values minimise
equation (85) are chosen as the final Xt, Yt and Zt coordinates of the transmitting nodes.
.
158
X t , Z t , Yt   min D fanl  D forig 
k
2
(85)
f 1
6.7.1 Experimental evaluation
The analytical localisation algorithm was examined at a practical level through all available
experimental distance difference values.
Figure 152 shows the measured positions as opposed the estimated positions along X axis.
(a) Small vessel tests
(b) Large vessel tests
Figure 152. Measured and estimated X axis positions
The AEs for data along X axis are shown in figure 153.
(a) X axis coordinates error
(b) Y axis coordinates error
Figure 153. X axis coordinates error
Figure 154 shows the measured positions and the estimated positions, computed using the
analytical method, along Z axis.
.
159
(a) Small vessel tests
(b) Large vessel tests
Figure 154. Measured and estimated Z axis positions
The AEs for data along Z axis are shown in figures 155.
(a) Z axis coordinates error
(b) Yz axis coordinates error
Figure 155. Z axis coordinates error
The MAE for the comparative method, for all positions along both X axis and Z axis as well
as the respective STDs are shown in table 16.
Small vessel tests
MAE(cm)
STD(cm)
Large vessel tests
X
Yx
Y
Yz
Z
X
Yx
Y
Yz
Z
5.32
4.2
9.3
6.4
9.4
6.15
9.5
7.8
6.43
4.96
37.22
27.7
36.2
37.56
36.8
19.38
37.4
33.7
17.93
19.92
Table 16: MAE and STD for the Analytical method
The analytical method proved, once again, that using all distance difference values, in each
experiment, to compute the position of the transmitter is best way to move forward.
Accuracies in the analytical method were as good as the accuracies observed in the advanced
IALED, BMLS and in ACL.
.
160
6.8 Summary
This chapter presented the theory behind seven novel TDoA-based localisation methods.
These methods were evaluated using experimental distance difference values; obtained using
the IALED method that provided the most accurate results (discussed in chapter 5).
The MAE from all localisation methods, using data from tests in the small vessel, discussed
in chapter 4, 5 and 6 are summarised in table 17.
Algorithm
Prototype (RGLED)
Prototype (BALED)
Prototype (IALED)
BMSF
BMLS
BMLSD
BCL
ACL
Comparative
Analytical
X
5.81
4.3
5.6
9.63
4.31
5.56
6.9
6.98
4.13
5.32
Small vessel MAE (cm)
Yx
Y
Yz
18.9
19.02
19.14
7.14
10.3
13.41
13.35
12.83
12.3
27.96
37.2
44.42
7.14
10.3
13.41
13.13
12.72
12.3
27.45
22.83
18.2
8.6
9.5
10.35
8.2
9.92
11.44
9.3
9.4
9.5
Z
7.4
9.6
4.71
9.42
4.62
4.71
5.94
7.91
5.4
6.43
Table 17: MAEs for all localisation methods in the small vessel
A detailed analysis of the performance of the prototype localisation algorithm in chapters 4
and 5 revealed that the accuracy of LE detection methods and consequently the accuracy of
distance difference values determine the precision of the localisation results. Furthermore, a
close examination of the distance difference profiles showed that errors are distributed
amongst all values randomly, that is no individual distance difference value at any receiving
antenna-transmitter combination shown to be consistently more accurate than any other case.
The above observations can be used to interpret the MAEs in table 17. The prototype
RGLED-based algorithm provided highly erroneous results since the method used to detect
the LE resulted in erroneous distance difference values. The implementation of two advanced
LE detections methods (BALED, IALED) especially of the IALED method improved
localisation accuracies (see chapter 5).
The BMSF and the BCL methods employed only a limited quantity of the available distance
difference values, in each experiment. Here, the exclusion of some values couldn‘t guarantee
that the remained values were not affected by noise to the extent that their usage could
provide accurate results. In fact, the design of the BMSF and of the BCL algorithms was
.
161
based on observations from analysis of theoretical distance difference profiles and didn‘t take
into account the presence of noise. As such, the MAEs for those two methods especially the
once concerning Y estimates, in table 17, are representative of the deficiencies of the
algorithms. Note that the rest of the new algorithms employed all the distance difference
values obtained at each experiment and implemented advanced methods to deal with errors in
distance difference values.
A review of all MAEs in table 17 points out that Y estimates in all algorithms are significant
higher that X and Z position estimates. This is a direct result of the use of equation 71 to
estimate Yt that is deeply dependent on the accuracy of the minimum value in the distance
difference profile. Note that the minimum value in the distance difference profile appears
both in the numerator and in the denominator of equation (71) thus any errors in these values
can have significant impact on Yt. This can be explained through an example as follows.
Consider a transmitter, in a 2D plane, at Xt=42cm, Yt=72cm in the setup seen in figure 104,
giving a minimum distance difference value of -11.3cm. If X in equation 71 remains constant
at 42cm but the minimum of the distance difference value is altered in θ steps (0.5cm up to
6cm) then this introduces errors in the computation of Yt. A plot of the absolute distance
difference errors (i.e. |-11.3-θ|) against the absolute errors in Yt (|72- Yt from equation 71|) is
shown in figure 156.
Figure 156. Errors in profiles Vs errors in Yt
Figure 156 demonstrates that even a small increase in the distance difference error of 0.5cm
results in an increase in Yt errors of 10cm, reaching an increased Yt error of 90cm even with
6cm error in the minimum distance difference value. These observations verify the
.
162
dependency of Yt on the value of the distance difference and justify the increased errors in Yt
estimates even when errors for Zt and Xt are low.
Similar trends in the MAEs, to the once characterised localisation in the small vessel, were
observed in the MAEs for localisation inside the large vessel as seen in table 18.
Algorithm
RGLED
BALED
IALED
BMSF
BMLS
BMLSD
BCL
ACL
Comparative
Analytical
X
33.75
25.4
26.45
30
37.57
37.7
24.67
28.6
51
37.22
Large vessel MAE (cm)
Yx
Y
Yz
69.44
78.32
87.2
42.4
42.5
42.6
13.65
33.425
53.2
76.3
124.9
173.5
64.31
39.94
15.6
90.31
87.58
84.86
64.96
89.67
114.4
62.97
66.72
70.46
5.66
18
30.33
36.2
36.8
37.4
Z
14.3
7.1
13.2
15
11.91
15
15
16.5
5
17.93
Table 18: MAEs for all localisation methods in the large vessel
The MAEs depicted in table 18, in particular those for Y estimates were significantly higher
than the MAEs in table 17. Here, the increase in the errors is directly related to the increased
errors attained in the estimation of distance differences as discussed in chapters 4 and 5. At
this stage it should be noted that large scale tests were only preliminary and substantially
more measurements are required to allow expressing solid conclusions on the impact of
placing the transmitters in large distances (>2m) from the receiving antennas.
.
163
CHAPTER 7
NETWORK-BASED LOCALISATION
7.1 Introduction
The UWB-based localisation methods, discussed in the previous three chapters, are all range
based, by the classification introduced in chapter 2. Special-purpose hardware is used to
estimate range (actually range/distance difference), which is then used in the determination of
the position of nodes. Although the investigations of the previous chapters showed that a
positional accuracy of less than 5cm can be achieved under laboratory conditions, there are at
least two reasons to believe that this accuracy might be problematic in very deep silos.
The first reason for believing that it might not be possible to obtain as good accuracy on large
silos, as in the laboratory, concerns the results of measurements, taken in the larger vessel and
reported throughout the last two chapters. From the limited data available it appears that for
the transmission powers used in the tests, the maximum transmission range of the UWB
pulses from the nodes is approximately 2m. The second reason is a corollary of Theorem 5
from chapter 6. For fixed Xt and Zt, distance difference reduces with increasing Yt. This
means that a fixed-size error in distance difference estimate will lead to increasing errors in
the estimate of Yt as the depth of the transmitter inside a vessel increases. The quality of the
estimates for Xt and Zt will also be affected, since according to Theorem 5 (in chapter 6),
distance difference profiles become flatter as transmitter depth increases, making the
estimation of the position of minimum distance difference increasingly error prone.
On the basis of the above reasoning, it was decided to investigate the applicability of
network-based techniques for localisation, of the type discussed in chapter 2 and which used
the sensor node‘s NB RF transceivers. The RSS data were collected during a set of in-silo
networking experiments and used to evaluate the feasibility of network-based positioning.
.
164
The WSN4IP node hardware and software are briefly reviewed in section 7.2 and issues
concerning antenna and node radiation are considered in section 7.3. A set of networking
experiments, conducted in a large vessel are briefly summarised in section 7.4 and section 7.5
contains preliminary results obtained using some of this data with a MATLAB
implementation of two network-based localisation methods. The data used were produced by
another member of the WSN4IP team and only became available at a later stage of the
current work. Hence only preliminary results are included.
7.2 WSN4IP nodes
In this section the basic hardware and software of the WSN4IP nodes is introduced briefly.
The radiation patterns of the node‘s antenna and of the complete node are also discussed and
the impact of grain on propagated signals is thoroughly examined.
7.2.1 WSN4IP hardware and software
Sensor node hardware and software were developed in the WSN4IP project. The node
hardware was based around a backplane organisation that included a number of printed
circuit boards (PCB), a microcontroller PCB, an RF transceiver PCB, a sensor PCB, an UWB
transmitter PCB and a battery board as illustrated in figure 157.
Figure 157. WSN4IP sensor node integrated parts [28]
The microcontroller PCB was based on a PIC32 microcontroller with an on-board RAM and
2GB flash memory. This board executes system software, principally the communication
protocol stack (see below) and application-level software. The RF transceiver board is based
.
165
around a Chipcon CC1101 transceiver which operates in the 868 MHz range of unlicensed
ISM bands. This particular device was chosen because it is highly configurable in terms of
many important radio parameters such as transmission power, frequency bands etc. This
degree of configurability was necessary due to the lack of knowledge of the radio channel in
grain. The board also contained a quarter wavelength helical antenna.
antennas
PCB
Figure 158. WSN4IP sensor node
The UWB circuit has already been discussed in chapter 4 and the sensor PCB was not used in
this work and is not discussed further. It should be noted that although all the elements
discussed above were built and tested, the complete node shown in figure 157 was never
integrated within the lifetime of the WSN4IP project. In particular, the UWB positioning
system has not been integrated with the rest of the node hardware.
The generation of UWB pulses was controlled by hardware that was independent of the
software running on the PIC microcontroller. However, in the future it is anticipated that the
PIC microcontroller will control both data communication and UWB propagation. The
employed UWB localisation system was discussed in greater detail in chapter 4.
Tests were carried out in large scale silos (2.5m deep, 1m across) filled with grain. Further
details of the testing arrangements can be found in chapter 4.
7.3 Node radiation characteristics
In network-based localisation using RSS, the antennas employed in the system have a key
role in determining localisation accuracy [52]. It has been observed that the use of antennas in
non-air medium as well as the orientation between receiving and the transmitting antennas
.
166
can have significant impact on RSS [52, 18,38, 39]. To investigate the behaviour of the
antennas used in the WSN4IP node a programme of measurements was undertaken by the
author and other members of the WSN4IP team.
Measurements were taken inside an anechoic chamber, in air and with the antennas immersed
inside grain using a Vector Network Analyser (VNA). Note that prior to any measurements
with the VNA a full calibration procedure was followed to remove systematic errors resulting
from external parameters such as the cables connecting the antenna to the VNA.
7.3.1 Antenna response in air
Prior to any tests, the radiation pattern of the antenna was derived through measurements
inside an anechoic chamber using the arrangement seen in figure 159.
Helical
antenna
Ground
plane
Rotating
platform
Figure 159. Errors in profiles Vs errors in Yt
The radiation pattern of the antenna was measured on an elevation and horizontal plane as
illustrated in figure 160.
(a) Elevation
(b) Azimuth
Figure 160. Quarter wave monopole antenna
The radiation pattern of the antenna alone for the ZY plane with the antennas facing each
other is shown in figure 161.
.
167
Figure 161. Elevation ZY radiation pattern
On the other hand the radiation pattern of the antenna alone for the XY plane with the
antennas facing each other is shown in figure 162.
Figure 162. Azimuth XY radiation pattern
Figures 161 and 162 indicate that the antenna‘s radiation pattern is almost isotropic in planes
perpendicular to the axis of the antenna, but significant directionality exists, with strong side
lobes, in planes containing the antenna‘s axis.
The next issue to be considered was the effect of the antenna sheath. The antennas were offthe-shelf components that were supplied in a protective plastic sheath. However the sheath
had to be removed to allow the antennas to fit into the enclosing mortar shells. Information on
the characteristics of the antennas and in particular its frequency were specified by the
.
168
manufactured through tests in air. Tests were carried out using a VNA to examine the effect
of removing the sheath on the helical antenna coil. In the above measurements, a single
antenna was connected, via a cable to the VNA, in order to measure the input reflection
coefficient, S11, at 868MHz. The reflection coefficient S115 provides an indication of how
well an antenna is tuned to work in a specific environment; the smaller the value of S11 the
better the antenna works since it shows that the overall reflected voltage from the antenna
into the VNA is less than the voltage going into the antenna.
S11 measurements with (covered) and without (bare) the sheath, are shown in figure 163.
These results revealed that the removal of the plastic sheath shifts the resonance frequency
from the nominal value of 868MHz (measured resonance in figure 163 was 872.5 MHz) to
about 925MHz [114].
Covered
antenna
Bare
antenna
Figure 163. Antenna response in air [114]
The shift in the resonance frequency is caused by the different dielectrics surrounding the
metal antenna coil. Detuning of antenna in grain is discussed in the following section.
7.3.2 Antenna response inside grain
Research in the GlacsWeb project [8-9] and in the networking experiments in silage [18-19]
revealed that antennas detune when operating in a non-air medium. In this connection, tests
were carried out by the author and other members of the WSN4IP team to establish the
impact of grain on the antennas.
5
S11 is a measure of the ratio of voltage reflected from the antenna back to the VNA over the voltage delivered
to the antenna. The smaller the S11 the better the antenna works.
.
169
Initial measurements were taken with the single antennas covered with the protective sheath
being buried in grain. These results are shown in figure 164.
(a) Antenna in air
(b) Antenna inside grain
Figure 164. Antenna response in air and in grain
Figure 164a shows that S11 is at -16dB at 868MHz when the antenna is in air whilst figure
164b shows that S11 goes to -25dB when the same antenna is immersed inside grain. The
frequencies at which the antenna is most responsive shifted from 868MHz in air to about
633MHz in grain.
The shifts in resonance frequencies are attributed to the high dielectric constant ε' of grain
(ε'=3) compared to only ε'= 1 for air that causes a reduction in the speed of the propagating
radio signal. The speed of propagation equals to the inverse of the square root of the
medium‘s dielectric constant [43, 45]. An increase in the dielectric constant will reduce the
speed of the propagating signals inside silage thus reducing the frequency of the signals.
The decrease in the speed of the propagating wave reduces the wavelength  of the radio
signal as depicted in equation (1). In this equation the frequency of the radio signal is fixed,
thus a change in the speed of the radio signal results in a change of the wavelength. The
length of the antennas used in the experiments was quarter wavelength long, tuned to work at
free space.
To alleviate the problems caused by the detuning of the antennas inside grain, the length of
the antennas was systematically reduced, to match up the wavelength inside grain, until
resonance returned to the nominal value. Figure 165 illustrates a tuned 868MHz antenna
operating inside grain
.
170
Figure 165. Tuned antenna inside grain
Figure 165 shows the frequency response of 868MHz antenna after tuning. It can be seen that
the S11 went up to -37.5dB which is an indicative of the reduction of the reflected signals
from the antenna which is a significant improvement. Further the resonance frequency
returned at 868MHz.
7.3.3 Antenna response whilst inside mortar shells
Results presented in the previous section were obtained with the antenna in physical contact
with the grain, which is not representative of the WSN4IP nodes, where the antenna will be in
contact with the air inside the node‘s mortar shell. As therefore it was anticipated that the
antenna will behave differently from what was discussed in section 7.2.2. To investigate how
the antenna responds whilst inside the mortar shell further tests were carried out.
For these tests, the antenna SMA coaxial connector was disconnected from the CC1101
transceiver on the WSN4IP node and connected to the VNA via a cable as seen in figure 166.
Figure 166. Antenna inside the protective sphere [114]
Figure 167 compares the response of the antenna (S11) when inside the sphere enclosing the
WSN4IP nodes (figure 166) and when outside.
.
171
Figure 167. Antenna response inside the protective sphere [114]
Figure 167 shows the antenna response in three situations a) using the single antenna in air
(as in section 7.2.1), b) when antenna was inside the mortar shell in air (see figure 166) and c)
when the antenna was inside the protective mortar shell that was buried inside grain. What is
shown in figure 167 is that there are no major variations in the antenna response in any of the
test cases and that the antennas behave in the same way as in an air environment. This can be
explained based on the fact that the VNA measures S11 by sending a very small amount of
voltage and measuring how much of this is reflected back to it, that is S11 represents a
measure of the immediate surrounding area (smaller than near field region) of the antenna
that is limited within the area bounded by the volume of the protective sphere.
7.4 The effect of antenna orientation
The effect of antenna orientation is of interest, since it is expected that nodes inserted in a silo
will assume a random orientation with respect to their antennas. The radiation pattern in figure
162 gave a first indication that the orientation of the antennas is a major factor that needs to be
considered carefully in the effort for RSS-based localisation [52].
A preliminary study, conducted by the author, showed that antenna orientation can have
significant impact on RSS. To this extent, measurements were carried out using a VNA and
two 868MHz helical antennas.
Prior to any measurements, using the VNA, a full calibration procedure was followed.
Calibration is a procedure that improves the accuracy of VNA results by removing systematic
errors resulting from external parameters such as cables and any other equipment connected
.
172
to the VNA. The magnitude and the phase of these errors cannot be predicted and therefore it
is considered that they add a certain degree of uncertainty to the measurements, thus making
calibration even more essential [129]. Further the averaging function of the VNA was
enabled so that the average of 128 measurements was recorded for each of the S parameter
readings, thus increasing in the accuracy of the results [129].
Two antennas, one transmitting and one receiving were connected to the VNA to measure
received signal strength (S21) for antenna orientations seen in figure 168.
VNA
antenna
(b) Azimuth orientation
(c) Elevation orientation
Figure 168. Antenna orientation positions
The transmitting power was set at 10dBm throughout the experiments and measurements
were taken at 868MHz. Measurements in air were taken using antennas that were tuned to
operate in air. On the other hand measurements inside grain were taken using antennas tuned
to operate in air as well as antennas tuned to work inside grain. Tuning of the antennas was
achieved by shortening the length of their metal coil to bring the wavelength of the antenna to
a level that allowed antennas propagate at 868MHz inside grain. Results from these
measurements are shown figure 169.
(a) Elevation orientation
(b) Azimuth orientation
Figure 169. Antenna tests inside grain and in air
.
173
What can be seen in figure 169 is that RSS reduces as the distance between the receiving and
transmitting antennas increases. Further, antenna orientation and specifically elevation
orientation caused significant attenuation with a reduction of 10dBm in RSS when compared
with the azimuth antenna orientation.
7.5 The effect of WSN4IP antenna orientation
The radiation patterns of the antenna in orthogonal planes are discussed briefly in section
7.2.2. However, since the antenna is partly located within the backplane structure of the node
(see photograph in figure 170a), its radiation pattern is likely to be affected by this. In
addition, it is expected that during grain flow node‘s antenna will assume a random
orientation.
The radiation pattern of the helical antenna, discussed in section 7.2.1, showed that this
changes according to the angle between transmitting and receiving antennas. This effect is
anticipated to be more intense in the case of the antenna inside the WSN4IP node; given a
significant portion of the antenna is surrounded by the metallic backplanes that include live
components; adding to the causes of possible distortions to the antenna radiation pattern.
To examine the impact of the WSN4IP backplanes on the radiation pattern of the antenna, a
programme of measurements was undertaken in the anechoic chamber utilising a fixed dipole
antenna, acting as a transmitter. The node itself was not operational during these experiments
and although the battery was inside the node this was not connected to the node. The output of
the WSN4IP antenna was fed via a cable to a VNA to acquire the radiation pattern. The node
was placed on a turntable as shown in figure 170 allowing it to be rotated in small increments
about the axis between measurements.
240cm
(a)WSN4IP node rotating anticlockwise
(b)Dipole 868MHz antenna
Figure 170. WSN4IP node Vs dipole in anechoic chamber starting at 0o
.
174
The radiation pattern of the WSN4IP node antenna at an angle to the dipole of 0o is shown in
figure 171.
Figure 171. Radiation pattern for the WSN4IP node at 0o
The radiation pattern in figure 171 indicates there are two side lobes with the stronger lobe
resulting when the WSN4IP antenna and the dipole face each other from the side of the
WSN4IP node that there is no backplane.
To further understand the effect of random orientation of WSN4IP nodes, while these move
with grain flow, a more detail 3D radiation pattern of the WSN4IP antenna was derived by the
author of [114] over numerous angles (shown on each axis) as seen in 172.
Figure 172. WSN4IP antenna 3D radiation pattern [114]
The WSN4IP antenna radiation pattern, shown in figure 172, is severely asymmetric with a
number of significant null points. These are related mainly to the backplane structure and
partly to the manual method used to place the node at different angles that introduced a level
.
175
of errors [114]. Figure 172 serves to indicate that antenna gain can change very rapidly for a
small angle variation (1o to 2o) [114]. As such in cases were simple log-normal law is used to
interpret RSS to distance, strong RSS could be perceived as nodes being close to each other
were in fact nodes are far apart and vice versa.
The author of [144] investigated the possibility of using information from the 3D pattern to
implement corrections to RSS, assuming that the orientations of received and transmitter are
known. This appears to be possible, however, as noted in [114], it is difficult to ensure an
error free radiation pattern that will cover every possible angle thus any compensation on
RSS based on the antenna patterns could not ensure 100% effectiveness.
Overall, these results demonstrate that for an efficient localisation approach using RSS
information it will be necessary to take into account the relative orientation of the receiving
and the transmitting antenna. Similar observations were recorded in extensive RSS
measurements in [52] where it is also stated that antenna orientation is a major factor that
needs to be examined carefully when considering RSS-based localisation.
7.6 Networking experiments in grain
A programme of networking experiments was designed by another member of the WSN4IP
team in which the ability of nodes immersed in grain to form network links was investigated.
This study used ten nodes and produced data about the variation of RSS with transmission
power and position, and the influence of transceiver variables such as transmitter power,
carrier frequency, data rate etc on parameters such as bit and packet error rates [114].
Since the RSS data were needed for the preliminary assessment of network-based
localisation, the author assisted in the design and implementation of the experiments.
7.6.1 Experimental configuration
A total of ten nodes were placed at predetermined positions inside the large silo that was
introduced in chapter 4. Nine nodes were placed in a rectangular grid in a plane in the centre
of the vessel. A 10th node was located at the top of the silo and was used for the purpose of
providing timed transmissions to control the other nine nodes and the progress of the
experiments. The experimental arrangement is shown in figure 173.
.
176
1
0cm
2
15cm
5
4
3
35cm
35cm
6
50cm
15cm
7
y
150cm
x
8
9
10
180cm
Anchor
Figure 173. Node's positions inside silo
The experiments were performed according to a pre-programmed schedule. For a particular
combination of transceiver parameters each of the nine nodes in the grid in turn took the role
of transmitter, whilst all other nodes were placed in receiver mode. Nodes stored received
data in their own flash memory for later retrieval and processing.
The results discussed in section 7.2.3 showed RSS can vary rapidly with angle at fixed
transmission power and distance. It was anticipated that the 3D radiation pattern shown in
figure 172 could be used to associate RSS to a particular direction thus dealing with random
fluctuations in RSS due to orientation. However, this required the relative orientation of the
transmitter and receiver to be known. Based on the intended arrangement of nodes specified
above, it was necessary to ensure that the nodes were all in a fixed orientation. This was a
difficult practical problem to solve given that any accurate placement of nodes in the silo had
to be followed by the insertion of grain, which could easily disturb the position and
orientation of the nodes.
After many different options had been considered, a system which fixed the nodes in position
in an empty silo and which kept their position during filling of the silo was chosen. A node
suspension ‗structure‘ was created from high tensile-strength fishing line and plastic tubing
both of which were unlikely to affect the RF transmissions occurring during the experiments.
The structure was supported from a metal bar across the top of the silo. Each mortar shell had
eye-hooks glued to the top and bottom to allow the assembly of the node array, as can be seen
in figure 174. Plastic tubes were inserted through the bottom of the spheres to prevent the
.
177
shells from rotating and helping them to maintain a fixed node orientation. The whole
assembly was attached to another mortar shell (anchor) that was placed in the bottom of the
silo to guarantee nodes did not move with the flow of grain whilst the silo was filled. The
anchor was fixed at the bottom of the silo thanks to the heaviness of grain and it balances the
whole apparatus through the filling process.
The whole assembly was lowered into the silo and measurements taken prior and after the
silo was filled with grain. Lowering of the assembly into the silo required significant effort
that was made even more difficult by the presence of a safety net covering the silo opening
and from the heaviness of the whole assembly. The author worked along a co-researcher from
the WSN4IP project and devised a method that allowed the lowering the assembly in steps
and the attachment of the plastic bar to sets of three nodes at a time until all nodes were
placed inside the silo and attached to the plastic bars.
The complete experimental assembly with the nodes attached to the plastic bars and the
fishing wire is shown in figure 174.
Metal bar
Plastic bar
Mortar shell
Figure 174. Experimental apparatus
Measurements, using the apparatus in figure 174, were conducted both in air and whilst the
silo was filled with grain at six different data rates: 1.2, 2.4, 4.8,10, 28.4 and 76.8 in kb/s and
at a frequency range at 50kHz intervals between 867MHz to 869MHz. Further for each data
rate, nine transmitting powers were examined: 0.62, -3.52, 5.76, 7.67, -8.02, 10.15, -13.24, 18.27, -25.25 in dBm [114].
.
178
The RSS register, on each transceiver (Chipcon CC1101), is eight bits wide and is accurate to
±0.5dBm. As such another WSN4IP researcher attempted to calibrate RSS to improve its
accuracy [114]. Calibration involved using a signal generator to generate signals of known
power and frequency to allow measuring RSS. In the first stage, the signal generator was
connected, via a cable, to the antenna input of the node‘s transceiver. In the other stage, the
signal generator was connected, via a cable, to a power meter thus providing a more accurate
RSS measurement. These tests revealed that, in both cases, variations in frequency didn‘t
have significant impact on RSS. On the other hand, tests with different transmit powers
shown that the CC1101 transceiver could accurately detect signals up to certain transmit
powers [114]. The information from RSS calibration was used to process the experimental
results accordingly. More details on this work can be found in [114].
The following sections present networking and localisation results based on a data rate of
4.8kb/s at a transmit power of -3.52dBm at 868MHz. These values were chosen as being
representative of typical network behaviour of the obtained results.
7.6.2 Networking results
The networking experiments produced an extremely large amount of data about networking
in grain. However, the general trends relevant to this work can be summarised as follows.
RSS does appear to reduce with increase distance and over the frequency range investigated;
carrier frequency has no impact on RSS [114]. Hence in most experiments, the network had
full connectivity, as shown in the sample plot of figure 175. However, at highest datarates
and lowest transmission powers, only partial connectivity was observed [114].
Figure 175. Network connectivity
.
179
The experimental data only became available late in this project and so there was not time to
perform an extensive study of network-based localisation using all data. It was decided to
consider a single set of experimental results and to apply two simple network-based
positioning algorithms to the data. The selected results were obtained under the same
conditions as figure 175 i.e. a transmit power of -3.52dBm, a carried frequency of 868MHz
and a datarate of 4.8kb/s. This particulate condition was chosen because it was representative
of many others.
Figure 176 shows characteristic RSS plots against distance for nodes 3 and 8. These figures,
and indeed all of the figures showing RSS over distances, indicate that although RSS varies
randomly with distance the overall trend is for RSS to reduce as distance between a transmit
and a receive node increases.
1
1
2
3
4
2
3
4
5
2
8
2
6
7
6
7
9
2
10
2
5
2
8
2
9
2
10
2
(a) RSS for node 3
(b) RSS for node 8
Figure 176. RSS Vs distance for nodes 3 and 8
The variability in RSS is mainly attributed to the WSN4IP backplanes, surrounding the
antenna, that alter the radiation pattern of the antennas as discussed in section 7.5. A
characteristic example of the impact of the backplanes is the overall observation that nodes
on the same horizontal level had higher RSS amongst them than any other orientation. This
observation is supported by the antenna radiation pattern shown in figure 171 where WSN4IP
antenna radiation was stronger on the sides were backplanes didn‘t obstruct LOS of the
antenna.
An additional cause of random RSS fluctuations is the fact that signals propagate from an air
environment inside the protective shells, through grain and back to air inside another mortar
.
180
shell. This inevitable gives rise to multipath due to multiple reflections and refractions thus
causing random fluctuations in RSS.
The variability of these results, which are typical of the RSS data reported in [114], means
that it is not possible to derive reliable log law parameters (e.g. path loss). In turn, this means
that it is not feasible to use any network-based positioning approach which relied on the log
law described in section 2.2.2.
Based on the above, it was decided to use two localisation algorithms that are relatively
robust to random RSS fluctuations
7.6.3 Localisation results
Network based localisation was implemented using information from RSS signals derived in
the experiments described above. Note that in any one time a node on the grid placement is
considered to be the unknown node whose position is to be estimated. At the same time the
rest of the nodes are considered to be anchors and are used as such in the localisation process.
The first attempt for localisation employed the centroid approach discussed in section 2.4.6.
In this case, given all nodes can communicate with each other (figure 175), in principle, if
anchors can broadcast their coordinates to the unknown then the latter can compute its
position as the centroid of all anchors. However, this is a very simplistic and inaccurate
approach. A more accurate localisation method considered employing information from
experimental RSS data.
Given information from RSS signals and specifically given the fact that RSS values are
significantly stronger for nodes on the same horizontal level (figures 176a-176b) then it is
possible to associate RSS with specific node positions and use this information to compute
the positions of the unknown nodes. To be more specific if an unknown node can receive the
locations of a number of anchors along with respective RSS from those anchors then it can
sort anchors with respect to RSS. In this way an unknown can take the centroid of those
anchors giving the strongest RSS to compute its position. This approach incorporated in an
algorithm running in MATLAB and executed using experiments RSS data.
.
181
Figure 177a shows the estimated positions for all nodes in the system. In this case, since RSS
was strongest amongst nodes on horizontal level and given there were only two anchors for
one unknown, position estimates were derived using the centroid of only two anchors. On the
other hand, figure 177b shows an implementation of the same method but this time using four
anchors, in the localisation process, with the highest RSS to the unknown.
1
2
3
4
5
2
8
2
6
7
9
2
10
2
(a) Two anchors centroid
1
2
3
4
5
2
8
2
6
7
9
2
10
2
(b) Four anchor centroid
Figure 177. RSS based centroid localisation
The MAE for position estimates using the centroid of only two anchors was 0.31m for X
positions whilst MAE for Y estimates was 0.18m. These results are justified by the fact that
the unknown nodes are on the same Y level as the anchors and as such Y estimates errors are
low. Note that the highest Y error occurs for node 1 since this doesn‘t have any nodes on the
same horizontal level. In Figure 177b it is obvious that an increase in the anchors in the
centroid causes position estimates to be drawn closer to the centred of the experimental setup
with subsequent increase in localisation error. In fact the MAE for four anchors in the
centroid was 0.33m for X positions and 0.305m for Y positions.
The proposed centroid-based localisation approach depends on the principle that RSS values
will follow specific trends based on the orientation amongst sensor‘s antennas. To this extent,
it serves to indicate that associating RSS to antenna orientation can be useful tool in
localisation especially in cases were nodes might be inside dynamic processes causing
random changes in nodes orientation.
Network-based localisation was also implemented using the Ecolocation algorithm discussed
in section 2.4.8. Figure 178a and 178b shows the estimated positions using Ecolocation
against the measured positions when using two anchors and with four anchors respectively.
.
182
Ecolocation results using experimental RSS readings, in grain tests, shown to reach
accuracies between 0.1m and 1m [125].
1
1
2
3
4
2
3
4
5
2
8
2
6
7
6
7
9
2
10
2
5
2
8
2
9
2
10
2
(a) Two anchors
(b) Four anchors
Figure 178. Ecolocation algorithm results
These results compare favourably to the experimental investigation of Ecolocation presented
in section 2.4.8 where for indoors tests using 12 nodes over an area of 120 m2 the localisation
errors were at about 100cm.
7.7 Summary
Work presented in this chapter examined network based localisation within the context of
localisation of sensors inside vessels filled with grain.
Based on the knowledge acquired from literature review on networking inside non air
medium, presented in sections 1.3 and 1.6 preliminary tests were conducted to determine the
impact of grain on the sensor nodes. These results revealed that as long as the antennas are
inside the protective mortar shell then grain doesn‘t cause detuning. However, the fact that
signals need to propagate from air (inside mortar shells) through grain and back to air raises
certain concerns since it inevitably gives rise to reflections and refractions all contributing to
multipath that makes localisation based on RSS challenging.
Analysis of the antenna radiation pattern, whilst inside the WSN4IP node, shown that this is
affected by the PCB boards. The correlation between antenna orientation and the node‘s
PCBs is a critical factor that needs to be considered carefully especially in RSS based
localisation. To this extent, the two algorithms examined in this chapter showed that RSSbased localisation depends greatly to all of the parameters affecting RSS.
.
183
CHAPTER 8
DISCUSSION AND FUTURE WORK
8.1 Discussion
This chapter begins with a summary of the work that has been carried out during this research
project and the results that have been obtained. A number of issues that have been raised by
the UWB-based localisation work are considered in section 8.2 with issues related to
network-based localisation being discussed in section 8.3. Proposals for further work on all
localisation methods are discussed in the last two sections of this chapter.
8.2 Thesis summary
The work presented in this thesis concerns the localisation of wireless sensor nodes that are
immersed in an industrial process contained inside vessels. Chapter 1 discussed the
motivation for the work and explained that the techniques developed were to be applied to the
localisation of sensor nodes within grain silos. The basic theory behind localisation, and
network-based localisation algorithms developed for WSNs, are discussed in chapter 2. Since
this work, for reasons discussed in section 1.5, has been concerned with further development
of a UWB-based localisation system, a review of UWB-based localisation systems is
presented in chapter 3.
Chapter 4 contains a discussion of the prototype UWB-based localisation system that formed
the basis of this work. The system consists of an antenna array mounted above the process
vessel which receives UWB pulses, transmitted by WSN nodes that need to find their
positions. Received signals from the antenna array are digitised by a DSO and processed by a
MATLAB program running on a PC. The software identifies the LEs of the signals arriving
at each antenna and used this information to derive TDoA data. TDoA data are then used in
an iterative, positioning algorithm, to calculate an estimate of the transmitter‘s position.
The hardware and software of the prototype localisation system are outlined in chapter 4. The
developers of this system did not perform a detailed experimental investigation of the
performance of this system, and so this was undertaken as an early part of this research. The
.
184
results are presented and discussed in sections 4.7 to 4.9. An analysis of the results obtained
from this experimental study using the RGLED method revealed that large localisation error
were often (69% of all position estimates) caused by an incorrect estimate of the LE
positions. To be more specific, in a total of 64 position estimates for X, Z and their respective
Y coordinates, 48% of positions had errors above 10cm. In addition 47% of the distance
difference profiles derived from the RGLED method had errors above 5cm. The errors arose
because the LE identification procedure was based on user observation of signals and was
difficult to apply consistently.
In response to the problems described in chapter 4, two automated LE detection algorithms
the BALED and the IALED were developed and thoroughly examined in chapter 5. Results
showed that 40% of position estimates had errors above 10cm for the BALED method whilst
only 19% of position estimates had errors above 10cm when using the IALED-based
localisation method. The respective data for distance differences for the BALED method
resulted in 31.25% of errors above 5cm with the errors in IALED almost halved with only
18.75% of errors above 5cm. Data from the IALED method proven to be the most accurate
from all LE detection methods with errors halved when compared to the prototype RGLED
method presented in chapter 4.
Chapter 6 concerned with using the properties of distance difference profiles (equivalent to
TDoA profiles) measured by an antenna array mounted above the process vessel of interest,
to determine the position of the transmitter. The chapter begins with a number of simple,
mathematical proofs of the properties of distance difference profiles and is followed by
proposals for seven localisation algorithms, all of which are related to the distance difference
approach. The results of the IALED method giving the most accurate distance difference
profiles, in chapter 5, were reprocessed with each of these new localisation algorithms.
Review of MAEs for all the new methods showed that these range from 4cm up to 10cm for
X and Z position estimates. However MAEs for Y positions, in cases reached up 30cm. The
reason for this, is the direct dependency of the mathematical equation, giving Y estimates, on
distance differences as described in section 6.7.
The average execution time for all algorithms was measured using a stop watch function in
MATLAB (tic-toc). The execution time for the new localisation algorithms presented in
Chapter 6 was about 0.15s except for the comparative method for which the execution time
.
185
reached 3s. These times compare favourably with the average execution time of the prototype
localisation algorithm (discussed in chapter 4) that was 0.25s. Note that the time it takes to
compute TDoA using cross-correlation is about 23s, and so reducing the execution time of
the localisation algorithm only has a small impact on the overall time needed to compute the
position of the transmitter.
Network-based localisation is re-introduced in chapter 7. Data for its implementation became
available at a late stage of this research, and so there was only time for a limited study. The
data was obtained from experiments that were designed for the purpose of obtaining a sound
understanding of the RF channel in grain and investigating networking issues such as link
formation [114]. However, RSS data produced from these experiments provided data that
could, in principle, be used with a variety of network-based localisation schemes that use
RSS for some kind of range estimation (typically via the log-normal law- see section 2.2.2).
Unfortunately, the hardware design of the wireless sensor nodes, along with the location of
the antenna in the middle of the node electronics, gave rise to a severely non-uniform node
radiation pattern (seen in section 7.2.4). Experimental RSS results from [114] show some of
the expected trends (e.g. falling RSS with increasing received-transmitter separation), but are
not consistent enough to enable trends (e.g. the log normal law in section 2.2.2) to be clearly
identified. The results show noticeable variation with direction relative to the transmitter‘s
antenna.
Given the difficulties with the experimental data, the Centroid [58,63] and Ecolocation [124]
algorithms were selected for application to the experimental results, since the Centroid
approach merely required the existence of links to anchors, and the Ecolocation algorithm
works by ranking the RSS values at the receiver due to transmissions from all in-range
anchor nodes. This makes the method robust to issues such as multipath [124].
Due to time constraints, the algorithms were applied to a single test case, selected on the basis
that it is representative of other results obtained under a variety of experimental conditions.
Applying the Centroid method gives localisation errors of less than 20cm for Y positions and
up to 40cm for X and Z position estimates. Results obtained with the Ecolocation algorithm
had localisation errors ranging from 5cm up to 100cm. These compare favourably with the
results obtained in an indoors measurement procedure, in [124], in artificial circumstances
(e.g. the anchors are, in most cases, very close to the unknown). Considering that localisation
.
186
errors can be as low as 5cm then further investigation of these techniques could be
worthwhile.
8.3 Future work on UWB-based localisation
UWB-based localisation was shown to work under laboratory conditions with accuracies
reaching few centimetres. However, implementation of this method in large-scale industrial
settings will require certain adaptations concerning mainly the power consumption of the
sensor nodes, the use of DSO with respect to its cost and practicality and the implementation
of the distribution of receive antennas.
8.3.1 UWB-receivers
Experimental work on localisation, presented in chapters 5 and 6, was based on the use of a
DSO to capture receiving signals from a single transmitting node. Here the use of the DSO
arose from the need to capture very short pulses (~ns) with adequate resolution that will then
allow the processing of these signals in order to derive TDoA. The use of an expensive DSO
in industrial premisses is undesirable not least since these can be damaged thus costing lots
money to be replaced. To this end, further work is required to make the receive part of the
localisation system more practical with one method being replacing the DSO with cheap and
less energy demanding alternatives that, will nevertheless, allow capturing received signals
with adequate resolution.
Literature in [116-117] pointed out that sampling received UWB pulses using fast ADCs with
sampling rate of the order of giga samples per second (GS/s), can be very expensive both in
money and in energy consumption and severy limits application areas. Research in [117]
pointed out that there are techniques that employ numerous low speed ADCs in order to
achieve satisfacory sampling rates. However, these techniques consume a lot of energy and
cause an increase of noise in the sampled signal.
To deal with issues arising from the need to use high sampling rates, research in [116-117]
proposed using sub-sampling; thus allowing recording UWB pulses at a lower rates, that
nevertheless, allow for a satisfactory reproduction of the received signals..
.
187
Work in [116] proposed a sub-sampling technique whereby sets of circuits was used to widen
the duration of the received UWB pulses, whilst the basic pulse shape amplitude remained
relatively unchanged. Extending the duration of the pulses allowed using slow ADCs to
capture the received pulses. The functionality of the sub-sampling circuit was tested using a
low sampling oscilloscope with 500 MS/s sampling rate. Results using a 0.3ns signal showed
that the sub-sampling circuit can produce as good representation of the received signal as in
the case when the signal is fed directly to an oscilloscope with a sampling frequency of
70GS/s. The subsampling circuit in [116] was used in a localisation method developed by the
authors in [119] with reported localisation accuracies reaching the milimiter range.
A preliminary study of experimental TDoA used in this thesis revealed that it is possible to
use a lower number of samples to capture a received signal whilst maintaining the same
degree of accuracy as in the case of using the maximum sampling rate. A received signal
captured by the DSO (sampling at 5GS/s) consisted of 131036 samples with the gated signal
containing 520 samples. The following table summarises the MAE for distance differences
for a total number of 48 individual distance difference values.
Transmitting
coordinates
Xt=28
Yt=89
Zt=28
Xt=42
Yt=72
Zt=42
Xt=53
Yt=89
Zt=61
Sub sampling
Factor
MAE for Xt
2
0.59
5
0.62
10
50
100
*
2.89
1.5
*
7.14
MAE for Zt
1.49
1.66
7.92
1.3
5.6*
MAE for Xt
2.26
2.21
2.2
14.4*
3.97
MAE for Zt
1.41
1.37
1.48
1.84
1.2
MAE for Xt
0.45
0.5
2.47
0.5
6.3*
MAE for Zt
3.65
3.63
8.4*
2.36
5.26*
Table 19: Sub-sampling MAEs in cm
Table 19 demonstrates that sub-sampling can give as accurate results as in the case of using
maximum amount of samples. This observation indicates that it is possible to use an analogue
to digital converter (ADC) with low sampling rate in the order of mega samples per second
(MS/s) to capture UWB pulses.
*
MAE exceeding 4cm indicate the presence of one or more large outliers caused by failure of the LE detection
methods to detect the correct LE
.
188
Building on knowledge from the literature and on the indications that sampling at a lower rate
can give accurate results as seen in table 19, future work should consider employing a circuit
build of a number of ADCs to sample incoming UWB pulses instead of using a DSO.
8.3.2 Placement of receiving antennas
In the experiments undertaken in this work a cross receiving antenna formation was situated
on the top of a round and square vessels (see section 4.6.2). Although this appeared to
provide satisfactory results this particular antenna formation might not be practical in real
world scenarios. Indeed, the very presence of the receiving antennas in the middle of the
large silo opening caused problems in filling of the silo with grain. The antennas had to be
removed during the filling in process to minimise spilage of grain kernels whilst these hit the
the antenna formation and to protect the antennas themsevles from falling appart.
Other than practicality issues, it is of interest to explore alternative placements of receiving
antennas with respect to localisation accuracy. To this extent, simulations presented in [123]
based a TDoA-based localisation scenario, similar to the research presented in this thesis,
considered placing the receiving antenna at different positions inside a cylindrical vessel.
Results in [123] showed that placing the receivers at the extremities (top, bottom and middle
i.e. as shown in figure 179) of the vessel can improve localisation accuracies. This is justified
by the fact that the largest the separation between the receiving antennas and the transmitter
the higher the spatial diversity that is achieved thus improving the probabilities of receiving
signals with reduced noise parameters.
Figure 179. Alternative receiving antenna placements
Results obtained from the experiments conducted in the larger silo (see chapters 5 and 6)
seemed to indicate difficulties in the reception of signals from transmitters buried at depths
.
189
below about 150cm. However, there is another potential issue in localising nodes buried at
large depths within a silo or other vessel, and that concerns the effect, described in Theorem
5, section 6.2.3, that deviation between distance differences reduces significantly as
transmitter depth increases.
For example, consider the following table, obtained by using equation 67 to compute the
minimum value in a distance difference profile. The calculations were performed on the basis
of a cylindrical silo of height 80m and diameter 30m, dimensions that are typical of the
largest silo in the world [127]. Assume that the coordinate system is the same as that
discussed in chapter 4 and used throughout chapters 4 to 6, and that a transmitter is located on
the silo‘s axis of symmetry, with the reference antenna located at (0m,0m,15m). Under these
assumptions, the value of K in equation 67 is 15, and the variation of the minimum value of
distance difference, Dmin, occurring in profiles at different depth is shown in table 20.
Yt(m)
Dmin(m)
20
-5
40
-2.72
60
-1.85
80
-1.39
Table 20: Depth accuracy
This shows that the issue of the flattening of distance difference profiles with increasing
depth is not likely to cause significant difficulties even in the largest silos since Dmin is still
distinguishable at large depths.
To summarise the above discussion, it appears that they key limitation to localisation in large
vessel is the distance over which UWB pulses can be transmitted to enable them to be
received and processed to find the LE. Clearly this could be addressed by building arrays of
receiving antennas into the walls of a large silo. Wall location is necessary to avoid the
damage that would inevitably occur if antennas were mounted on struts that project into the
body of the material within the silo. Building antennas into the walls of new silos is
potentially expensive, and in silos of large diameters still may not enable signals from the
transmitters near to the centre of the silo to be received by the antenna array. The distance
.
190
over which UWB signals can be transmitted in grain and other types of material stored in
silos needs to be studied to establish the applicability of the type of UWB-based positioning
discussed in this thesis to the largest silos.
Nevertheless, there many silos and storage facilities that are significantly smaller than the
maximum size discussed above. In particular, flat storage facilities are fairly common in the
grain industry where floor areas are large and depths are much smaller than in large silos.
This is the kind of silage storage that is discussed in reference [18, 19] and was reviewed in
section 1.3.3. The type of UWB-based positioning techniques discussed in this thesis could
be adapted for such environments.
8.4 Future work on network-based localisation
Time did not allow an in depth study of network-based localisation. However, some
preliminary results were obtained using RSS data from within the larger vessel described in
section 7.6.2. The main issues of concern regarding this data were the position of the antenna
inside the WSN4IP node and the severely non-uniform radiation pattern produced by the
node. This means that the node orientation, which is difficult to predict, if a node moves
freely with the flow of grain, as seen in figure 180, is likely to have an impact on the ability
of the node to acquire meaningful RSS data from its neighbours.
Antenna
Sensor
node
Silo
.
Figure 180. Random antenna orientation
Many WSN applications are essentially 2D, or closely approximate this, and so it is relatively
straightforward to ensure an approximately uniform radiation pattern in the plane. See for
.
191
example, the whip antenna polar plot in figures 161 and 162. However, industrial processes
are essentially 3D and if network-based localisation is to be explored further in this context,
then it is important for a node to have an approximately uniform radiation pattern in 3D.
One way to achieve a uniform radiation pattern is trhrough the use of patch antennas as
discussed in [115] where four directional (~90o coverage) patch antennas were placed around
a sensor node as shown in figure 180. The antennas were controlled by a switch network that
allowed the switching of antennas for transmitting or receiving according to a control signal
issued by the node microcontroller. Note that at any one time only one antenna was active
however the switching time was set at 150ns that according to the authors gave enough time
for the antennas to transmit or receiving signals [115].
Figure 181. Angular antennas around a sensor node [115]
The objective of the work in [115] was to use mutiple antennas to read RSSI whilst the node
was taking different orientations with respect to a static transmit node that was equipped with
a quarter wavelength whip antenna. This allowed determining the orientation of the node in
the testing space given that, according to the authors, strongest RSSI values occured only
when the radiation lobes of the directional antenna meet the radiation lobes of quarter wave
whip antenna [115].
Similarily work in [123] attempted to deal with the rotation of the nodes under water and the
assumption of random orientations, between the acoustic transducers and the receivers, using
multiple transducers attached on the outer surface of round nodes. Extensive analysis showed
that using eight transducers could deal effectively with orientation problems [123].
Using patch antennas on the WSN4IP node would require the removal of the whip antenna
and an investigation of the number and placement of patch antennas around the periphery of
.
192
the node, in order to provide as uniform radiation pattern as possible, whilst keeping cost,
complexity and power consumption within reasonable limits.
Returning to the issue of large silos discussed in the previous section, the possibility of
combining UWB-based positioning with network-based positioning could be explored. In
such cases, nodes that are within range of a receiving antenna array can have their position
established relatively accurately using the UWB-based system. The position of such a node
could be transmitted from the external PC via the network back to the node itself, which
could then act as an anchor for nodes that are out of range of a receiver array.
8.5 Future work on the node’s power supply
In principle, nodes could remain immersed in a process for a considerable time and so
conserving energy to prolong battery life becomes important.
The electronics inside the nodes, for both UWB and network-based localisation, were
powered by rechargeable batteries. The batteries, powering the WSN4IP nodes, used for
network-based localisation, lasted for one day because the nodes undergone numerous tests
using different transmit powers and datarates [114]. Similarly the batteries for UWB-tests
lasted for about a day of tests. In all cases the batteries had to be recharged for many hours
thus causing delays in the experimental procedures.
The requirements for monitoring industrial processes such as grain inside silos necessitate
that sensors nodes should be able to operate for long periods of time ( months). Indeed the
software designed for the WSN4IP nodes utilises a sleep-wake up schedule that allows nodes
to be active only for specific small time intervals whilst remaining in a low power mode for
most of the time [114].
The proposed sleep-wake up schedule in [114] entails that nodes are synchronised to transmit
and receive at predetermined time intervals based on a clock provided by the microcontroller.
However the use of microcontroller‘s clocks and possible jitter that might occur have been
reported to cause discrepancies the data acquired by the nodes [114, 76]. To deal with those
issues, future work could consider using RF signals transmitted by nodes to trigger receive
nodes to wake up thus avoiding the need for clock synchronisation [68-69].
.
193
Future work could also examine how long a network of WSN4IP nodes could operate using
energy saving techniques and also investigate techniques that will allow nodes to harvest
energy from their surrounding environment. Energy harvesting will allow charging the
batteries without the need for the nodes to be extracted from the monitoring process thus
providing extra monitoring time [70, 118,126].
.
194
REFERENCES
[1] C.S Raghavendra, K. M. Sivalingam and T. Znati., ―Wireless Sensor Networks‖, Boston
: Kluwer, ISBN 1402078838, 2004.
[2] College of the Atlantic ―In-Situ Sensor Networks for Habitat Monitoring‖ available:
http://www.coa.edu/html/motes.htm
[3] M, Cardei, I. Cardei, D. Du, ―Resource Management in Wireless Networking‖, Vol. 6,
pp. 507-542, Springer US, 2005.
[4] P. Gober, A. Ziviani, P. Todorova, M. D. de Amorim, P. HÄunerberg, and S. Fdida,
―Topology control and localization in wireless ad hoc sensor networks, Ad Hoc & Sensor
Networks‖, pp. 301-321, 2005.
[5] Dr P. Green, ― Ad Hoc and Sensor Networks-Positioning in Wireless Sensor Networks‖,
lecture notes, University of Manchester, 2009.
[6] James Cook, available: http://en.wikipedia.org/wiki/James_Cook>
[7] B. Zhang, G. Sukhatme, A. Requicha, ―Adaptive Sampling for Marine Microorganism
Monitoring‖, IEEE/RSJ, Intelligent Robots and Systems, 2004.
[8] K. Martinez, J. K. Hart, R. Ong, ― Deploying a wireless sensor network in Iceland‖,
LNCS, vol. 5659/2009, Springer Berlin / Heidelberg, 2009.
[9] K. Martinez, J. K. Hart, R. Ong, ―Environmental Sensor Networks‖, IEEE Computer, 37
(8). pp. 50-56.1, 2004.
[10] P. Padhy, R. Dash, K. Martinez, N. Jennings, ―A utility-based adaptive sensing and
multi-hop communication protocol for wireless sensor networks‖, ACM transactions on
Sensor Networks, 2010.
[11] G.E. Rolader, J. Rogers, J. Batteh, Self-healing minefield, in: Proceedings of the SPIE –
The International Society for Optical Engineering, Battlespace Digitization and
Network-Centric Systems IV, 13–15 April 2004, vol. 5441, no.1 pp. 13–24, 2004.
[12] T. Alhmiedat, A, Taleb, M. Bsoul, ―A Study of Threats Detection and Tracking
Systems for Military Applications using WSNs‖, International Journal of Computer
Applications, Vol. 40, No.15, 15 February 2012.
[13] M.A. Hussain; P. Khan; K.K. Sup;, "WSN Research Activities for Military
Application," 11th International Conference on Advanced Communication Technology,
2009(ICACT 2009), vol.01, pp.271-274, February 2009.
[14] S. Nawaz‚ M. Hussain‚ S. Watson‚ N. Trigoni, P. N. Green, ―An Underwater Robotic
Network for Monitoring Nuclear Waste Storage Pools‖, 1st International ICST
Conference on Sensor Systems and Software (SCUBE), 2009.
.
195
[15] S. Nawaz, N. Trigoni, ―Robust Localization in Cluttered Environments with NLOS
Propagation‖, 7th IEEE International Conference on Mobile Ad−hoc and Sensor
Systems (IEEE MASS), November 2010.
[16] M.Hussain‚N.Trigoni,‖Distributed Localization in Cluttered Underwater
Environments‖, Fifth ACM International Workshop on Underwater Networks
(WUWNet), 2010.
[17] P.N.Green,P.R. Green, M. Hussein, S. Narwaz, A. Phasouliotis, Z. Qu, N. Trigoni, S.
Watson and T. York ―Mapping legacy storage ponds‖, Nuclear Future, Volume 7 Issue
5, September 2011.
[18] J. Larsen, O. Green, E. Nadimi, T.Toftegaard , ―The Effect on Wireless Sensor
communication When Deployed in Biomass‖, 8295-8308, Sensors 2011.
[19] O. Green, E.S Nadimi,V. Blanes-Vidal, R.N. Jorgensen, I.M.D Storm, I.M.D.; C.G.
Sorensen, ―Monitoring and modeling temperature variations inside silage stacks using
novel wireless sensor networks‖, Comput. Electron. Agr. ,pp. 69, 149–157,2009.
[20] D. D. Perkins, R. Tumati, H. Wu, and I. Ajbar. Localization in Wireless Ad Hoc
Networks. In Resource Management in Wireless Networking, Springer US, volume 16,
July 2006.
[21] J. Heidemann, N. Bulusu and D. Estrin, ―GPS-less low cost outdoor localization for very
small devices, IEEE Personal Communications Magazine Vol. 7 No. 5, pp. 24-28,
October 2000.
[22] P. Coronel, S. Furrer, W. Schott, B. Weiss,‖Indoor Location Tracking Using Inertial
Navigation Sensors and Radio Beacons‖, IOT, Springer-Verlag Berlin Heidelberg ,2008.
[23] C. Wann, H. Chin, ―Hybrid TOA/RSSI Wireless Location with Unconstrained
Nonlinear Optimization for Indoor UWB Channels‖, IEEE,WCNC, 2007.
[24] T. He, C. Huang, B. Blum, J. Stankovic, T. Abdelzaher,‖ Range-Free Localization
Schemes for Large Scale Sensor Networks‖, MobiCom , ACM, San Diego, California,
USA, 2003.
[25] Q. Huang, S. Selvakennedy ―A Range-Free Localization Algorithm for Wireless Sensor
Networks‖, University of Sydney, NSW, Australia, 2006.
[26] G Davis, R. Sloan, A.J. Peyton, and M. Bilal, ―A Positioning Algorithm for Wireless
Sensors in Rich Multipath Environments‖, IEEE Microwave and Guided Wave Letters,
September 2008.
[27] G. Parkinson, M.C. Boon, J.G. Davis and R. Sloan, ―3D positioning using spherical
location algorithms for networked wireless sensors deployed in grain‖,IEEE MTT-SInt.
Microwave Symp. Dig., Boston, MA, pp. 1417-1420, June 2009.
.
196
[28] G. Parkinson, D. Crutchley, P. M. Green, M. Antoniou, M. Boon, P. N. Green, P. R.
Green, R. Sloan, T. York, ―Enviromental Monitoring in Grain‖, IEEE, I2MTC, 2010.
[29] M. Antoniou, M.C.Boon, P.N.Green, P.R.Green and T.A.York, ―Wireless Sensor
Networks for Industrial Processes‖, Sensors Applications Symposium (SAS), IEEE,
USA, February,2009.
[30] J. Davis, R Sloan, A.J. Peyton, "A 3-D Positioning Algorithm for Networked Wireless
Sensors", IEEE Transactions on Instrumentation & Measurement, April 2011.
[31] T. Weis, ―The Global Food Economy: The battle for the Future of Farming, Zed Books,
July 2007.
[32] SAFEGRAIN, Inc., SafeTrack-Grain Temperature Monitoring Sytem,
Brochure,available: http://www.safegrain.com
[33] Banner Engineering Corp., SureCross product line app notes, available:
http://www.bannerengineering.com
[34] TempuTech, Inc.,available: http://www.temputech.com/page3.html
[35] N. Magan, D. Aldred, K. Mylona, RJM Lambert, ―Limiting mycotoxions in stored
wheat‖, Food Additives & Contaminants: Part A Chemistry, Analysis, Control,
Exposure & RiskAssessment 27, 644–50, 2010.
[36] K. Coggins, J. Principe, ― Detection and Classification of Insect Sounds in a Grain Silo
using a Neural Network‖, The IEEE World Congress on Computational Intelligence
and The IEEE international Joint Conference on Neural Networks, 1998.
[37] R. Hickling, W. Wei, ―Sound transmission in stored grain‖. Appl. Acoust..45,1-8, 1998.
[38] R. Hickling, W. Wei,D. Hagstrum, ―Studies of sound transmission in various types of
stored grain for acoustic detection of insects. Applied Acoustics 50, 263–278, 1997.
[39] P. Armstrong, ‗Wireless data transmission of networked sensors in grain storage, 2003,
ASAE Annual Meeting, ASAE Riviera Hotel and Convention Center, Las Vegas, USA,
27- 30 July 2003.
[40] J. Jia, J. Kuang, Z. He, Y. Mu, ―Design of Monitor System for Grain Depots Based on
Wireless Sensor Network‖ Proceedings ofthe 2009 IEEE, International Conference on
Mechatronics and Automation, Changchun, China, 2009.
[41] H. Zhou, F. Zhanq, J. Liu, F. Zhanq, ―A Real-time Monitoring and Controlling System
for Grain Storage with ZigBee Sensor Network‖, Fifth International Conference on
Wireless Communication, Networking and Mobile Computing, WiCom, 2009.
[42] S. O. Nelson, ―Agricultural Applications of Dielectric Measurements‖, IEEE
Transactions on Dielectrics and Electrical Insulation Vol. 13, No. 4, August 2006.
.
197
[43] Balanis, C. Antenna Theory : Analysis and Design, 3rd ed.; John Wiley & Sons:
Hoboken, NJ, USA, 2005.
[44] Ponne, C.; Bartels, P. Interaction of electromagnetic energy with biological material
relation to food processing. Radiat. Phys. Chem., 45, 591–607, 1995.
[45] J. J. Carr, ―Practical Antenna Handbook‖, 4th edn, McGraw-Hill, ISBN: 071374353,
2001.
[46] G. Mao, B. Fidan and B.D.O. Anderson, Wireless sensor network localization
techniques, Elsevier/ACM Computer Networks 51 (10) , pp. 2529–2553, 2007.
[47] T. Callaghan, N. Czink, A. Paulraj, and G. Papanicolaou, Correlation-based radio
localization in an indoor environment, submitted, available at URL:
http://math.stanford.edu/˜papanico/pubftp/corrbasedlocalization.pdf
[48] S. Gezici, Z. Tian, G.B. Giannakis, H. Kobayashi, A.F. Molisch, H.V. Poor, and Z.
Sahinoglu, "Localization via ultra-wideband radios", IEEE Signal Processing Mag.,
vol. 22, no. 4, pp. 70 - 84, 2005.
[49] A. Bensky ―Wireless Positioning-Technologies and applications‖, Artech house, 2008.
[50] C. Savarese and J. Rabaey, ―Robust positioning algorithms for distributed ad-hoc
wireless sensor networks‖, USENIX Annual Technical Conference, Monterey, CA,
June 2002.
[51] X. Cheng , A. Thaeler , G. Xue , D. Chen ,‖TPS: A Time-Based Positioning Scheme for
Outdoor Wireless Sensor Networks‖, IEEE INFOCOM, 2004.
[52] D. Lymberopoulos, Q. Lindsey, A. Savvides, ―An Empirical Characterization of Radio
Signal Strength Variability in 3-D IEEE 802.15.4 Networks Using Monopole
Antennas‖ EWSN, LNCS 3868, pp. 326–341, 2006.
[53] Rappaport, T.S.: Wireless Communications. Prentice Hall, New Jersey, 1996.
[54] S. Tian, X. Zhang, P. Liu, P. Sun, X. Wang, ―A RSSI-based DV-hop Algorithm for
Wireless Sensor Networks‖, IEEE, 2007.
[55] N. Patwari and A. O. Hero III, ―Using proximity and quantized RSS for sensor
localization in wireless networks‖, in IEEE/ACM 2nd Workshop on Wireless Sensor
Nets & Applications, September 2003.
.
198
[56] N. Patwari, A. O. Hero III, M. Perkins, N. Correal, and R. J. O‘Dea, ―Relative location
estimation in wireless sensor networks‖, IEEE Trans. Sig. Proc., vol. 51, no. 8, pp.
2137–2148, August 2003.
[57] K. Sohrabi, B. Manriquez, G. Pottie, ―Near-Ground Wideband Channel Measurements‖
IEEE Proc. VTC, New York, 1999.
[58] J. Blumenthal, R. Grossmann, F. Golatowski, D. Timmermann ―Weighted Centroid
Localization in Zigbee-based Sensor Networks‖, Folien IEEE International Symposium
on Intelligent Signal Processing, WISP 2007.
[59] A. Savvides, C.C. Han, and M. Srivastava, ―Dynamic Fine-Grained Localization in AdHoc Networks of Sensors‖, Proceedings of ACM Mobicom, Rome, Italy, pp. 166- 179,
July, 2001.
[60] S. Qin, H. Hong, F. Tao, L. De-Ren ―Using Linear Intersection for Node Location
Computation in Wireless Sensor Networks‖ Acta Automatica Sinica, Vol. 32, No. 6,
November 2006.
[61] A. Savvides, H. Park and M. B. Srivastava, ―The NHop Multilateration Primitive for
Node Localization Problems‖, Mobile Networks and Applications 8, pp. 443–451,
Kluwer Academic Publishers, 2003.
[62] D. Niculescu and B. R. Badrinath, ―Ad hoc positioning system (APS)‖, IEEE
Globecorn ,San Antonio, November, 2001.
[63] J. Heidemann, N. Bulusu and D. Estrin, ―GPS-less low cost outdoor localization for very
small devices, IEEE Personal Communications Magazine Vol. 7 No. 5, pp. 24-28,
October, 2000.
[64] J. Blumenthal, F. Frank Reichenbach, D. Timmermann ―Precise Positioning with a Low
Complexity Algorithm in Ad hoc Wireless Sensor Networks‖, PIK - Praxis der
Informationsverarbeitung und Kommunikation J., vol. 28, pp. 80–85, 2005.
[65] J. C. Adams , W. Gregorwich , L. Capots and D. Liccardo "Ultra-wideband for
navigation and communications", Proc. IEEE Aerosp. Conf., vol. 2, p.2/785 , 2001.
[66] R. J. Fontana,"Experimental results from an ultra wideband precision geolocation
system", EuroEM 2000.
[67] Fontana, R.J. and S.J. Gunderson, ―Ultra Wideband Precision Asset Location System,‖
Proceedings 2002 IEEE Conference on Ultra Wideband Systems and Technologies,
Baltimore, MD, May 2002.
.
199
[68] L. Gu, J. A. Stankovic, ―Radio-Triggered Wake-Up for Wireless Sensor Networks‖,
proceeding of 10th IEEE Real-Time and Embedded Technology and Applications
Symposium, 2004.
[69] M. Antoniou, ― A Wake-Up Radio for Wireless Sensor Networks‖, MSc
Communication Engineering thesis, The University of Manchester, September 2006.
[70] D. Harrist, ―Wireless battery charging system using radio frequency energy harvesting‖,
MSc thesis, University of Pittsburg, July 2004.
[71] J. Chuang, N. Xin, H. Huang, S. Chiu and D. G. Michelson, ―UWB radiowave
propagation within the passenger cabin of a Boeing 737- 200Aircraft,‖ in Proc. IEEE
VTC 2007, pp. 496-500, April 2007.
[72] J. Jemai et al., ―UWB channel modelling within an aircraft cabin,‖ in Proc. IEEE
ICUWB, pp. 5-8, September 2008.
[73] Q. Bonds, T. Weller, H. Arlsan, ― An Ultra-Wideband (UWB) Pulse Dispersion Study
for Antenna in Sensor Network Applications‖, IEEE, 2007.
[74] G. Wang and W. Kong, ―Angle-dependent pulse distortion in UWB radiation and its
impact on UWB impulse communications,‖ Electron. Lett., vol. 41, pp. 1360–1361,
December 2005.
[75] R.C. Qiu, C. Zhou, and Q. Liu, ―Physics-based pulse distortion for ultrawideband
signals,‖ IEEE Trans. Veh. Technol., vol. 54, no. 5, pp. 1546–1555, September 2005.
[76] J. Schroeder, S. Galler, and K. Kyamakya, "A Low-Cost Experimental Ultra-Wideband
Positioning System," Proceedings of IEEE International Conference on UltraWideband, Zurich, Switzerland, pp. 632-637, 2005.
[77] M. Vossiek, M. Wiebking, L. Gulden, P. Weighardt, and J. Hoffmann, ―Wireless local
positioning—Concepts, solutions, applications,‖ in Proc. IEEE Wireless Commun.
Netw. Conf. , pp. 219–224, August 2003.
[78] H. Liu, H. Darabi, P. Banerjee, J. Liu, ―Survey of Wireless Indoor Positioning
Techniques and Systems‖, IEEE transactions on systems, man and cybernetics-Part C:
Applications and reviews, vol.37, No.6, November 2007.
[79] D. Dardari, A. Conti, U. Ferner, A. Giorgetti, M.Z. Win, ―Ranging With Ultrawide
Bandwidth Signals in Multipath Enviroments‖, Proceedings of the IEEE, Vol.97, No. 2,
February 2009.
[80] J.Zhang, P. Orlik, Z. Sahinoglu. A.F. Molisch, P. Kinney, ―UWB systems for wireless
sensor networks‖, Proc. IEEE (Special Issue on UWB Technology and Emerging
Applications), Vol. 97, No. 2, February 2009.
.
200
[81] IEEE Standard for Information Technology – Telecommunications and Information
Exchange Between Systems- Local and Metropolitan Area Networks- Specific
requirement Part 15.4: Wireless Medium Access Control (MAC) and Physical Layer
(PHY) Specifications for Low-Rate Wireless Personal Area Networks (WPANSs),
IEEE Std 802.15.4a-2007, 1-203, August 2007.
[82] T.Kaiser, ―UWB Communication Systems: A Comprehensive Overview‖,Hindawi
Publishing Corp., May 2006.
[83] A. Azim, M. A Matin, A. Amin, and N. Amin, ―UWB Technology for WSN
Applications, Novel Applications of the UWB Technologies‖, Boris Lembrikov (Ed.),
ISBN: 978-953-307- 324-8, InTech publisher, 2011.
[84] L. E. Miller, ― Models for UWB pulses and their effects on narrowband direct
conversion
receivers‖, Wireless Communication Technologies Group, National
Institute of Standards and Technology, Gaithersburg, Maryland, USA
[85] L. E. Miller, ―Why UWB? A review of ultra wideband technology,‖ Wireless
Communication Technollogy Group,Gaitherburg, Maryland, April 2003.
[86] G. MaCGougan, K. O‘Keefe, R. Klukas,―Ultra-wideband ranging precision and
accuracy‖, Measurement science and technology, 2009.
[87] R.C. Qiu, C. Zhou, and Q. Liu, ―Physics-based pulse distortion for ultrawideband
signals,‖ IEEE Trans. Veh. Technol., vol. 54, no. 5, pp. 1546–1555, 2005.
[88] K. Rambabu , A. Tan , K.Chan , M. Chia, S. Leong "Study of antenna effect on UWB
pulse shape in transmission and reception", Proc. ISAP 2006.
[89] K. Rambabu , A. Tan , K.Chan , M. Chia, "Estimation of antenna effect on ultrawideband pulse shape in transmission and reception", IEEE Trans. Electromagn.
Compat., vol. 51, no. 3, pp.604- 610, 2009.
[90] I. Oppermann, M. Hämäläinen, J. Iinatti ―Introduction, in UWB Theory and
Applications‖, John Wiley & Sons, Ltd, Chichester, UK.
[91] J. Liang, C. C. Chiau, X. Chen, and C. G. Parini, ―Study of a printed circular disc
monopole antenna for uwb systems,‖ IEEE Trans. Antennas Propagat., vol. 53, no.
11, pp. 3500–3504, November 2005.
[92] W. Wiesbeck, G. Adamiuk and C. Sturm "Basic Properties and Design Principles of
UWB-Antennas", Proceedings of the IEEE, vol. 97, no. 2, pp. 372-385, 2009.
[93] J. Zhang, J. Wang, W. Hu,‖ Analysis of UWB signal distortion in transmitting/receiving
antenna Systems‖, IEEE, 2010.
[94] K. Rambabu , A. Tan , K.Chan , M. Chia, "Experimental verification of Ultra-WideBand pulse radiation from monopole array antennas‖, Progress In Electromagnetics
Research C, Vol. 17, 39-53, 2010.
.
201
[95] M. Mahfouz, A. Fathy, M. Kuhn, Y. Wang, "Recent trends and advances in UWB
positioning," Wireless Sensing, Local Positioning, and RFID, 2009. IMWS 2009. IEEE
MTT-S International Microwave Workshop on , vol., no., pp.1-4, 24-25, 2009.
[96] D. Clarke, A. Park, "Active-RFID System Accuracy and Its Implications for Clinical
Applications," cbms, pp.21-26, 19th IEEE Symposium on Computer-Based Medical
Systems (CBMS'06), 2006.
[97] T. Ye, M. Walsh, P. Haigh, J. Barton, A. Mathewson, B. O‘Flynn, ―Experimental
Impulse Radio IEEE 802.15.4a UWB Based Wireless Sensor Localization Technology:
Characterization,Reliability and Ranging‖, ISSC 2011, Trinity College Dublin, 2011.
[98] J-S. Lee, Y. Su, and C-C. Shen, ―A Comparative Study of Wireless Protocols:
Bluetooth, UWB, ZigBee, and Wi-Fi,‖ Proc. 33rd Annual Conf. IEEE Industrial Elect.
Soc,Taipei, Taiwan, 2007.
[99] C. S. R. Murthy and B. S. Manoj, ―Ad Hoc Wireless Networks: Architecture and
Protocols, ch. Recent Advances in Wireless Networks‖, pp. 773, Prentice Hall
Communications Engineering and Emerging Technologies Series, New Jersey:
PrenticeHall Professional Technical Reference, 2004.
[100] K. Siwiak, D. McKeown,―Ultra-Wideband Radio Technology‖. John Wiley and Sons,
2006.
[101] T. Ye, M. Walsh, P. Haigh, J. Barton, A. Mathewson, B. O‘Flynn,‖ Experimental
Impulse Radio IEEE 802.15.4a UWB Based Wireless Sensor Localization
Technology: Characterization, Reliability and Ranging‖, ISSC 2011, Trinity College
Dublin, 2011
[102] J. Ni, D. Arndt, P. Ngo, C. Phan, K. Dekome, and J. Dusl, ―Ultra-Wideband TimeDifference-OfArrival High Resolution 3D Proximity Tracking System,‖ in Proc.
IEEE Position Location and
Navigation Symposium (PLANS), Indian Wells, CA,
USA, pp. 37 - 43, May 2010.
[103] J. Ni and R. Barton, ―Design and Performance Analysis of a UWB Tracking System
for Space Applications,‖ in Proc. IEEE/ACES Int. Conf. on Wireless Communications
and Applied Computational Electromagnetics, pp. 31- 34, April 2005.
[104] A. Maali, G. Baudoin, H. Mimoun, A. Ouldali,‖ UWB Localization System with
TDOA Algorithm Using Experimental Measurements‖, The Third International
Conference on Advances in Satellite and Space Communications, IARIA, SPACOMM
2011
[105] I. Guvenc and Z. Sahinoglu, ―Threshold based TOA estimation for impulse radio
UWB systems,‖ in Proc. IEEE Int. Conf. UWB (ICU), Zurich, Switzerland, pp. 420425, September 2005.
.
202
[106] A. Maali, A. Mesloub, M. Djeddou, G. Baudoin, H. Mimoun, and A. Ouldali,
―Adaptive CA-CFAR threshold for non coherent IR-UWB energy detector receivers,‖
IEEE Communications Letters, Vol. 13, No. 12, pp. 1-3, December 2009.
[107] G. Schiavone, P. Wahid, E. V. Doorn, R. Palaniappan, J. Tracy "Target detection and
tracking using a UWB sensor web," Antennas and Propagation Society International
Symposium, IEEE , vol.2, no., pp. 1287- 1290 Vol.2, 20-25, June 2004.
[108] G. Schiavone, P. Wahid, R. Palaniappan ―Study of Ultra Wide band Signal
Propagation in Urban Environments‖, Antennas and Propagation Society International
Symposium, 2002. IEEE, Volume 3, pp 346- 349, June 2002.
[109] G. Schiavone, P. Wahid, R. Palaniappan, J. Tracy, T. Dere, ―Analysis
of Ultrawide band signal propagation in an indoor environment‖, Microwave and
Optical Technology letters, Volume 36, Issue I, pp 13-15, January 2003.
[110] C. Bishop, ―Effects of averaging to reject unwanted signals in Digital Sampling
Oscilloscopes‖, AUTOTESTCON, IEEE, 2010.
[111] R.M. Morey,‖ Ground penetrating radar for evaluating subsurface conditions for
transportation facilities‖. NCHRP Synthesis 255, Transportation Research Board,
National Research Council, pg. 7, Washington, USA, 1998.
[112] G. Stephenson ― Advanced Mathematical Methods for Engineering and Science
Students‖, Cambridge University Press 1990.
[113] R. K. Jain, ―The Art of Computer Systems Performance Analysis‖, Wiley, 1991.
[114] D. Crutchley, ―Wireless Sensor Networks in Hostile RF Environments‖, PhD
thesis,The University of Manchester, 2012
[115] C. Yang, S. Bagchi, W. Chappell,‖ Location Tracking with Directional Antennas in
Wireless Sensor Networks‖, IEEE, 2005.
[116] C. Zhang, A. E. Fathy, “Performance Enhancement of a Sub-Sampling Circuit for
Ultra-Wideband Signal Processing‖, IEEE Microwave and wireless computer letters,
Vol. 17, NO. 12, December 2007.
[117] M. Chen and R. Brodersen, ―A subsampling UWB impulse radio architecture utilizing
analytic signaling," IEICE Trans. Electron., vol. E88-C, no. 6, pp. 1114-1121, 2005.
[118] J. A. Paradiso, ―Energy Scavenging for Mobile and Wireless Electronics‖, IEEE, 2005.
[119] Zhang, M. Kuhn, B. Merkl, M. Mahfouz, and A. E. Fathy, ―Development of an
UWB indoor 3D positioning radar with millimeter accuracy,‖ IEEE MTT-S Int. Dig,
pp. 106–109, June 2006.
[120] H. Chen, K. Sezaki, P. Deng, H. C. So, ―An Improved DV-Hop Localization
Algorithm for Wireless Sensor Networks‖, IEEE,2008.
.
203
[121] X. Wang, J. Lin,‖ Research and Improvement of DV-Hop Localization Algorithm For
Wireless Sensor Network‖, Applied Mechanics and Materials Vols. 44-47 pp 40284032, Trans Tech Publications, Switzerland,2011
[122] M. Keshtgary, M. Fasihy, Z. Ronaghi, ―Performance Evaluation of Hop-Based RangeFree Localization Methods inWireless Sensor Networks‖, International Scholarly
Research Network ISRN Communications and Networking Volume, Article ID
485486, 2011
[123] M. G. Pottinger, ―Localisation of Underwater Sensor Nodes in Confined Spaces‖, PhD
thesis, The University of Manchester, 2012
[124] K. Yedavalli, B. Krishnamachari, S. Ravula, B. Srinivasan, ―Ecolocation: A Sequence
Based Technique for RF Localization in Wireless Sensor Networks‖, IEEE, 2005
[125] P.N. Green, Private Communication, July, 2012
[126] E. Torres, G. A. Rincon-Mora, ―Harvesting ambient energy will make embedded
devices autonomous‖, Embedded System Design Magazine, 2005
[127] Storage Silos, available: http://en.wikipedia.org/wiki/Silo
[128] M.Antoniou, P.N. Green, ―Adaptive leading edge detection for WSN localisation inside
industrial processes‖, IET, WSS2012, June 2012
[129] Hewlett Packard ―HP 8510B Introductory Users Guide‖, July 1998
.
204
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertising