Overhead Impacts on Long-Term Evolution Radio Networks Master of Science Thesis

Overhead Impacts on Long-Term Evolution Radio Networks Master of Science Thesis
Overhead Impacts on Long-Term
Evolution Radio Networks
PETTER EDSTRÖM
KTH Information and
Communication Technology
Master of Science Thesis
Stockholm, Sweden 2007
COS/CCS 2007-19
Overhead Impacts on Long-Term Evolution Radio Networks
Petter Edström
Master of Science Thesis in Information Technology
School of Information and Communications Technology (ICT)
Royal Institute of Technology (KTH)
Project conducted at
Wireless Access, Radio Network Features
Ericsson Research
Ericsson AB
May 31, 2007
Supervisor at Ericsson:
Supervisor and Examiner at KTH:
Arne Simonsson,
Senior Specialist, Radio Network Product Performance
Radio Network Features, Wireless Access
Ericsson Research, Ericsson AB
Luleå, Sweden
Gerald Q Maguire Jr.
Professor,
Department of Communication Systems (COS), ICT
Royal Institute of Technology (KTH)
Kista, Sweden
Overhead Impacts on Long-Term Evolution Radio Networks
Overhead Impacts on Long-Term Evolution Radio
Networks
Abstract
As a result of the constant efforts to improve mobile system performance and
spectral efficiency, the 3GPP standardization forum is currently defining new
architectural and functional requirements that hope to ensure long-term
evolution (specifically defined as the “Long-Term Evolution (LTE) concept”)
and general future competitiveness of the 2G and 3G radio access
technologies.
Previous discussions on LTE efficiency have been focused on general
assumptions on signaling overhead and overall system capacity, based on
experience from existing mobile systems. However, as 3GPP standardization
has become more mature (although not yet settled), there is a need to
investigate how different potential LTE services will be affected by the use of
available overhead information and basic scheduling algorithms.
This thesis investigates the lower protocol layers’ overhead impacts on the
downlink for different packet switched services, in an LTE radio access
network (RAN).
Results show that the use of RTP/TCP/IP header compression (ROHC) is the
single most important factor to reduce payload overhead, for packet sizes of
~1kB or smaller. However, for packets larger than ~1 kB, the use of ROHC
becomes insignificant.
Protocol headers – including the AMR frame header, RLC/MAC headers, and
CRC where applicable – remain the largest part of payload overhead
regardless of packet size and header compression (ROHC).
For VoIP over the UDP protocol (with ROHC), RLC/MAC headers constitute
the largest part of protocol headers.
For TCP/IP applications (without ROHC), TCP/IP headers are predominant.
Services that require packet sizes beyond ~1 kB will require about the same
power per payload bit regardless of percentage of payload overhead.
Keywords
Long-Term Evolution, LTE, Power Emission, Efficiency
2 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Sammanfattning
Som ett resultat av ständiga ansträngningar att förbättra såväl prestanda som
spektrumeffektivitet
för
mobila
system,
definierar
3GPPs
standardiseringsforum nya krav på arkitektur och funktionalitet. Dessa är
avsedda att säkerställa långsiktig utveckling (explicit definierat som konceptet
“Long-Term Evolution (LTE)”, samt framtida konkurrenskraft för både 2G och
3G som radioaccess-teknologier.
Tidigare diskussioner rörande effektivitet inom LTE har fokuserat på allmänna
antaganden vad gäller kontrolldata för signallering och övergripande
systemprestanda. Dessa har i sin tur baserats på erfarenheter från existerande
mobilsystem. När standardiseringen inom 3GPP mognar uppstår nu ett behov
av att undersöka hur olika tjänster inom LTE påverkas, av såväl hur man
använder den kontrollinformation som finns tillgänglig, som av basala
algoritmer for schemaläggning av resurser.
Denna rapport undersöker påverkan från lägre protokoll-lagers
kontrollinformation på nerlänken hos olika paket-kopplade tjänster inom ett
radioaccessnät för LTE.
Resultaten visar att användandet av ROHC (som packar kontrollinformation för
protokollen RTP/TCP/IP), är det ensamt viktigaste bidraget till minskad
kontrollinformation i relation till informationsbitar för paketstorlekar upp till c:a
1kB. För större paket är vinsten med ROHC dock försumbar.
Kontrollinformation för protokoll – inkluderat data avsett för AMR-tal-ramen,
RLC/MAC-protokollen, samt CRC – utgör för övrigt en stor del av
kontrollinformationen relativt informationsbitar, oavsett paketstorlek och
packning av kontrolldata.
Tjänster som kräver paketstorlekar på över c:a 1 kB kräver uppskattningsvis
samma mängd energi per informationsbit, oavsett andelen kontrollinformation.
3 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Acknowledgements
I extend my deepest gratitude to Anders Furuskär, Magnus Lindström, Stefan
Parkvall and Fredrik Persson at Ericsson Research, for much appreciated
guidance on the latest news on 3GPP discussions on channel configurations
and overhead data constellations, in spite of their tight schedules.
My advisor at Ericsson Research, Arne Simonsson has never ceased to
support me with insightful ideas and encouragement for the duration of this
thesis.
Discussions with Mårten Ericsson have been very useful when investigating
similarities and differences between the LTE and WCDMA/HSDPA access
networks.
I would also like to thank Karolina Bergman, Mikael Bertze, Mårten Sundberg
and Magnus Thurfjell for helpful Matlab support.
Professor Gerald Q Maguire Jr, has provided me with valuable hints during the
planning and writing of this thesis.
Last by not least I would like to thank my manager at Ericsson, Lennart Blixt,
for being very understanding and supportive as I have tried to coordinate my
studies
with
my
employment
duties
at
Ericsson.
4 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Table of Contents
1
Introduction ...................................................................................... 6
1.1
Background .......................................................................... 6
1.2
Standardization Efforts for Improved Performance .............. 8
1.3
Problem Statement............................................................... 8
1.4
Method ................................................................................. 9
1.5
Thesis Outline .................................................................... 10
2
Technical Background.................................................................... 12
2.1
Multi-Carrier Modulation and FDM ..................................... 12
2.2
Orthogonal Sub-Carriers in OFDM..................................... 14
2.3
The OFDM System Model.................................................. 14
2.4
Coding and Interleaving ..................................................... 14
2.5
Speech Encoding in LTE .................................................... 15
2.6
Discontinuous Transmission .............................................. 16
2.7
The Cyclic Prefix ................................................................ 16
2.8
OFDM Advantages............................................................. 18
2.9
OFDM Impairments............................................................ 19
2.10
LTE Network Architecture................................................... 24
2.11
Radio Interface Protocol Architecture................................. 25
2.12
Protocol Overhead Considerations .................................... 28
2.13
Multiplexing & Multiple Access ........................................... 29
2.14
LTE Resource Blocks & Resource Elements ..................... 29
2.15
Channels and Signals ........................................................ 32
2.16
Control Channel Transmit Diversity.................................... 39
2.17
MIMO ................................................................................. 39
3
Previous Studies ............................................................................ 40
3.1
Expectations of LTE ........................................................... 40
3.2
3GPP LTE Standardization ................................................ 46
3.3
Potential Standardization Improvements............................ 48
4
Radio Network Model: Assumptions & Parameters ....................... 50
4.1
Link Performance Model Definition .................................... 50
4.2
Protocol Overhead ............................................................. 51
4.3
Scheduling Overhead......................................................... 52
4.4
Summary of Control Information Overhead ....................... 54
4.5
Definition of Total Overhead............................................... 55
4.6
Overhead Data Impacts on Service Performance.............. 57
4.7
Energy Transmission Considering DTX ............................. 58
4.8
Environment Details ........................................................... 58
5
Results ........................................................................................... 61
5.1
Common Control Information Overhead ............................ 61
5.2
SID-frames, Scheduling, and HARQ ACK/NACK............... 63
5.3
Protocol Overhead ............................................................. 67
5.4
Total Overhead................................................................... 69
5.5
Required Energy for Service Provisioning ......................... 73
5.6
Boosted Reference Signal Power ...................................... 77
6
Discussion ..................................................................................... 79
7
Conclusions ................................................................................... 81
8
Further Studies .............................................................................. 82
Appendix A: Extended Cyclic Prefix Figures………………………. 90
Appendix B: Concepts………………………………………………... 94
Appendix C: Abbreviations……………………………………………98
5 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
1
Introduction
1.1
Background
Today there are in excess of 2 billion users of mobile systems for wireless
communication [1]. Alongside GSM, the most deployed mobile Radio Access
Network (RAN) technology by far, there are other technologies emerging, such
as Wideband Code-Division Multiple Access (WCDMA)-High Speed Packet
Access (HSPA) and Long-Term Evolution (LTE).
Enhanced modulation techniques such as Orthogonal Frequency Division
Multiplex (OFDM) – used with LTE – as well as numerous other services such
as wireless local area networks (WLANs), Asymmetric Digital Subscriber Line
(ADSL), and Very high-rate Digital Subscriber Line (VDSL), increases the endusers’ expectations on available services, their availability and performance.
This contributes to new requirements on efficient use of available spectrum.
New technologies provide higher peak data rates. This requires that the use of
power over a limited bandwidth and the subsequent introduction of network
interference are carefully considered.
When information is transferred over a radio interface in a wireless mobile
system such as LTE, a frequency carrier of a specified bandwidth is modulated
with information and coded at the transmitting end, conveying information that
can be interpreted by the receiver despite interference and signal degradation
along the transmission path. Designers of packet switched wireless systems
try to use the available frequency spectrum as efficiently as possible with
regards to both payload and control information given the busty nature of
packet-oriented end-user transmissions. High spectrum efficiency can provide
service to more users, or higher data-rate end-user services to fewer users
over a fixed allocation of spectrum. Improving service performance for some or
increasing service availability for many users while using the same amount of
bandwidth, are both examples of the improved spectral efficiency sought for
packet switched services.
In contemporary packet switched mobile systems a number of information bits
are coded into symbols that are modulated onto one single carrier as a single
stream of data. When transmission rates are increased in the ongoing quest
for improved spectral efficiency (or more specifically higher end-user data rates
while still using the same limited bandwidth), the time used to transmit each
symbol over a single carrier is decreased. A shorter symbol time renders the
system more susceptible to losses along the transmission path, noise (impulse
noise in particular), and interference on symbol or carrier level. The alternative
however – using a wider bandwidth to combat losses and interference
(normally required for achieving higher data rates over a single carrier) –
increases the risk of being subject to single strong interference sources that
use the same or adjacent resources in time and frequency.
6 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Using available techniques and an OFDM carrier constellation, an efficient
implementation that enables multiple, relatively robust (considering
interference), and narrowband carriers can be quite easily deployed in theory
with the use of Fourier transformations. However, in practical systems, system
impairments and inaccuracies caused by transceiver equipment and radio
propagation properties will reduce the stability and robustness of a system with
multiple carriers and a limited bandwidth. These imperfections need to be
accommodated in order to minimize unwanted interference and the
subsequent reduction in spectral efficiency.
Alongside the need for an efficient technique enabling increased spectrum
efficiency there are several access technologies attempting to co-exist in the
wireless market. An increasingly important factor in the planning of such multiaccess technology networks is the use of energy. Two main reasons stand out
when discussing why the use of energy is so important in complex mobile
networks;
1. Cost savings. Network operators’ costs for running mobile networks are
increasing. The power inefficiency – considering that transmitted power
has been subject to feeder losses, and that power is required for
cooling – heavily affects an operator’s expenditure. This is more and
more of a concern since adequate service performance needs to be
provided at all times, irrespective of the energy levels that need to be
transmitted. Different levels of network load provide an opportunity to
reduce energy consumption, if transmitted energy can be lowered to
match the need for the load in the network (given that the service level
can be maintained). As a consequence, lower end-user tariffs and
maintaining return-on-investment targets in modern mobile networks
suggest that operators should minimize their energy consumption while
providing a given service with acceptable coverage and capacity.
2. High power levels cause network interference. Emission of power from
a mobile network base station causes interference in the surrounding
area. This impacts service accessibility, retainability and overall
performance. In order to optimize performance it is of utmost
importance to reduce the intra-system and inter-system interference
when services are provided in co-existing networks with limited
spectrum. The information transmitted to and from a system should
ideally only use only as much energy as necessary to provide this
service, in order to maximize performance on a network level.
In order to save energy and minimize interference in the long run on the
network level, the service performance impacts of transmitted energy (in terms
of interference as well as payload and overhead data) need to be considered.
For deployment in different environments, constellations, or user scenarios, the
relative effectiveness of the transmitted energy needs to be explicitly
understood - specifically addressing control information when using different
types of services. Impacts from OFDM system impairments should ideally be
accounted for whenever applicable.
7 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Another aspect of optimizing service and system performance is the
coordination of the scheduling of available radio resources. The scheduling
procedures need to be adapted for best possible service performance from an
end-user perspective, given a certain interference situation. Current
standardization activities [28] are addressing many aspects of scheduling
coordination. For this reason, optimized and coordinated scheduling is left for
others to investigate in detail.
1.2
Standardization Efforts for Improved Performance
As a result of the constant effort to improve mobile system performance and
spectral efficiency, the 3GPP standardization forum is currently defining new
architectural and functional requirements that ensure long-term evolution
(specifically defined as the LTE concept) and general future competitiveness of
the 2G and 3G radio access technologies. The evolution from the existing
GSM and basic WCDMA access technologies is addressed [50], as well as
further enhancements through HSPA [51] and the LTE concept. The main foci
are increased spectrum efficiency, data rates, and coverage, as well as
reduced latency. Considering the downlink, the focus of this thesis, spectrum
efficiency as well as the mean and cell-edge user throughputs are proposed to
be increased to three times that of a basic WCDMA system (as defined in the
3GPP standards release 99). In addition to these foci, flexible spectrum
allocations, reduced costs, peak data rates above 100 Mbps, and a
significantly reduced latency for control information, are included as targets for
performance improvements. More comprehensive information on the evolution
of the LTE standard is available in [28] and [32].
1.3
Problem Statement
Changing the level of emitted power is the major factor which can alter levels
of interference in a network. This is one of the major aspects of potentially
improved resource utilization in addition to reduced operator expenditure for
electrical power. However, there is no linear relation between the reduction of
power and lower interference in a network. Nor is there a linear relationship
between the reduction of transmitted power and the electrical power
consumed. Therefore it is important to investigate how efficiently a modern
mobile system such as LTE can use the available power to provide adequate
services.
Since the resource structure in an OFDM based LTE system enables optimum
use of resources whenever orthogonality can be maintained, it would be
interesting to investigate the energy required to transfer various amounts of
payload data, given that different services likely have different overhead
information.
A few introductory concerns describe the basic issues that will be dealt with in
this thesis:
•
How much energy is required for different packet sizes in an LTE
network?
•
How much and what type of overhead data exists and what are the
basic overhead requirements for different types of services, numbers of
users, and potential downlink antenna configurations?
8 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
1.4
Method
The aim of this investigation is to describe the relative efficiency with which
available radio resources are deployed in LTE, using different end-user
services, in order to obtain generally higher data rates and lower latency than
what is achievable with conventional mobile systems, being they 2G or 3G,
deploying EDGE or HS(D)PA.
When investigating the performance impacts from the use of extensive control
information, the extreme cases of two very different services would be
interesting to compare; one service being sensitive to delays using very small
packets, and another service using large packets that can sustain larger
service and end-to-end delays. Examples of such services could be VoIP and
an FTP transfer over TCP (respectively). Previous investigations of LTE
system capacity have focused on absolute levels of capacity and on one
service only, mainly speech services over VoIP using AMR CODECs.
Theoretical capacity estimates at the system level have also assumed the use
of a predefined, constant level of overhead and the use of only one speech
CODEC, without considering the actual energy consumption per unit of
transferred information content in an LTE system.
Upcoming investigations will first of all try to clarify the energy needed for a
basic service, given different AMR CODECs, including DTX activation. All
types of overhead required to enable downlink connection establishments will
be investigated in detail. Overhead related to the actual payload transfer will be
examined separately, including the impact of using different types of resource
scheduling algorithms.
The results from this study will hopefully clarify previous issues of concern for
overhead and resource efficiency in LTE, especially for services with high
demands for low service delays. In addition, recent progress in the 3GPP
standardization of LTE (with regards to channel constellations and resource
mapping), hopefully will enable this thesis to contribute to a more
comprehensive understanding of the basic resource requirements related to
overhead information, possibly even without regard to the packet sizes used in
an LTE downlink transfer.
The necessary steps to achieve these goals include the following actions:
•
A literary study of available OFDM techniques and available channel
constellations when applied in a mobile system. Suggested frequency
planning strategies in LTE, potential OFDM impairments and some
reasoning behind packet switched resource management strategies are
included for orientation.
•
A transmitter energy calculation shall be done including all known
channels and signaling, but excluding transmitter losses and other
hardware energy consumption. Downlink (base station) energy usage
is estimated.
•
The considered channel constellations are modeled using a generally
accepted link performance model, based on Shannon’s theories
adopted to accommodate changes in radio characteristics over the
radio interface.
9 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
•
Typical end-user services (with different packet sizes) are investigated
with regards to the required power levels and chosen scheduling
techniques as well as performance, in terms of Ws/Mbits and its
inverse Mbits/Ws.
•
The signaling overhead needed for the above investigated scenarios, is
investigated specifically, both the overhead required on the system
level and the additional payload overhead per user (see section 4.4).
There are a number of performance figures that potentially can provide
answers, or at least be subject to further investigations based on the problem
statement in section 1.3 and initial LTE standardization activities referenced in
section 3.1. They can be summarized as:
•
What are the Ws/Mbit figures at different packet sizes, with constant
system latency?
•
How much and what type of overhead data is used, and what are the
basic overhead data requirements for different types of services, given
the Ws/Mbit figures above, and potential downlink antenna
configurations?
The energy cost for a packet-switched transmission in LTE can – given that the
propagation losses are accommodated – be approximated as the power used
during a specified time interval. Hence, the quantity watt seconds [Ws] will be
the relevant unit. Looking at various packet data sizes including necessary
overhead data this quantum can be normalized into watt seconds per megabit
[Ws/Mb], where the relative energy would be its inverses [Mb/Ws] or [Mbps/W].
This thesis focuses on the downlink of 3GPP LTE, since most services require
higher throughput and increased efficiency specifically for downlink
transmissions. Secondly, many of the channel content details related to the
uplink have yet to be standardized.
Potential system impacts on system characteristics from OFDM impairments
are discussed mainly for orientation. Predominant performance bottlenecks
due to impairments have been identified on the uplink during this thesis; hence
impairments will not be handled specifically, as the focus is on downlink
performance.
Current standard assumptions and state of the art algorithms and solutions are
assumed.
Unless stated otherwise, references to the status of 3GPP standards are up-todate as of December 2006.
1.5
Thesis Outline
An historical background to the evolution of the OFDM technique, the basic
properties of OFDM, the access technology deployed in LTE, and the intended
LTE network architecture and channel constellations are all investigated in
section 2. Concerns for LTE system performance are also handled in this
section (for orientation).
10 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Previous studies on energy emission and 3GPP LTE related radio resource
management techniques are examined in section 3.
The applied evaluation model, assumptions and parameters are described in
section 4.
Results are presented in section 5.
Section 0 discusses interpretations of these results.
Finally, sections 7 and 8 contain thesis conclusions and ideas for potential
further studies.
Discussions within this thesis, as well as some of the presented results and
conclusions conceptually consider different aspects of energy consumption in
an LTE network, in the context of:
•
Recently discussed 3GPP standardization content for LTE, focusing on
the use of overhead data, and
•
Potential improvements to reduce the energy consumption in general or
for specific services in particular.
It should be noted that the general introduction and the initial sections are
intentionally very basic with regards to energy and the use of spectrum. The
latter part of this thesis examines protocol details focusing on overhead of the
data link layer (layer 2) of the OSI-model, which requires a basic
understanding of the protocol structures of data communication.
11 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
2
Technical Background
2.1
Multi-Carrier Modulation and FDM
One modulation technique that has proven very useful given the requirements
for high interference and noise robustness, stable channel characteristics and
high data rates in current mobile system standardization efforts, is a multicarrier modulation scheme called OFDM [20]. However, before the advantages
and potential impairments of OFDM in mobile systems are discussed, one
needs to understand the basic concept of how the available bandwidth is
utilized and the techniques that OFDM has evolved from.
Multi-carrier modulation systems, of which OFDM is one example, were first
developed during the 1960’s for military applications. Keller and Hanzo [26]
and references therein provide further details on the historical details of these
pioneering applications of multi-carrier modulation. When the Discrete Fourier
Transform was proposed for modulation and demodulation some years later in
1971 [25], mathematical operations - using Fourier transforms - could be
applied to transform data between the time and frequency domains. One
example of an early implementation of OFDM with parallel carriers is the
Telebit Trialblazer Modem, using the packet ensemble protocol [53].
The fast Fourier transform (FFT) and related frequency domain calculations
made it possible to introduce OFDM carriers using a digital modem, since data
could be mapped onto orthogonal carriers.
Advances in hardware design have made it possible to use FFT (and its
inverse IFFT) handling OFDM channels on integrated circuits for commercial
applications, albeit not yet in commercial wide area mobile systems [21]. IEEE
802.11a/g and Hiperlan2 (WLAN/WiFi), Digital Audio Broadcasting (DAB), and
terrestrial Digital Video Broadcasting (DVB-T) are but a few examples of
current wireless OFDM applications. OFDM is also used for wired applications
such as ADSL – where it is often referred to as Discrete Multitone Modulation.
The available spectrum intended for high data rates in a multi-carrier
modulation system is divided into a large number of sub-channels of slower
rates. Parallel carriers can then be assigned simultaneously using frequency
division multiplexing technique (FDM). Having a number of parallel narrowband channels instead of one wideband channels drastically simplifies the
equalization process that operates upon a signal at the receiver. A channel that
is small enough to be considered narrow-band can also be considered to have
constant or flat fading frequency characteristics [23], [24]. Such a channel can
be interpreted by a much simpler equalizer than one designed for processing
wideband channels.
12 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
The sum of transmission rates of multiple channels in FDM equals the sum of
all lower transmission rates of the sub-carriers. Thus the symbol time can be
increased without decreasing the over all data rate. This renders the subcarriers less susceptible to time dispersion between symbols, transmission
loss, noise, and interference, which reduces the need for complex equalization
at the receivers even further. Instead of using the entire spectrum for
transmission of only one symbol during a certain time period, several symbols
are transmitted within the allocated bandwidth using all N sub-carriers. Here
we assumed that the sub-carrier spectra have equal amplitude and overlap to
some extent, although the main lobes of the sub-carriers do not. The spectra
of three such carriers are shown in Figure 1.
The use of several parallel carriers enables high transmission rates for a
specific service, even though each sub-carrier uses a much smaller bandwidth
than the total assigned bandwidth. However, the potential interference between
the sub-carriers needs to be minimized so that each sub-carrier can make the
most out of its available bandwidth, hopefully resulting in optimized
performance over the entire channel.
One way to remedy the interference between sub-carriers is to introduce guard
bands between the non-overlapping sub-carriers, as shown in Figure 1. This
technique is used in traditional FDM systems such as for NTSC television and
FM stereo transmissions [15]. However, this approach will reduce the system’s
spectral efficiency in comparison to a system using the same bandwidth but
only one carrier, since parts of the available spectrum are left unused, although
the sidelobe interference is suppressed.
amplitude
f
Figure 1: Three FDM carriers separated by guard bands
In a mobile system demanding high data rates, it is not acceptable to reduce
spectrum efficiency with FDM by using guard periods to minimize interference
between the sub-carriers and to simplify signal interpretation. Rather it is more
optimal to use the entire bandwidth of each sub-carrier, while ensuring no
interference at all between sub-carriers – this requires making the sub-carriers
orthogonal to each other.
13 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
2.2
Orthogonal Sub-Carriers in OFDM
The O in OFDM represents the ideal suppression of interference between the
narrowband sub-carriers in an FDM system. When sub-carriers are made
orthogonal to each other, the interference between them is eliminated. The
energy from any orthogonal sub-carrier is completely uncorrelated with that of
the other sub-carriers, and cannot be interpreted as useful energy by another
sub-carrier. This allows the spectra of the sub-carriers to overlap, transmitting
more information over the same total bandwidth without causing interference.
This also improves the spectral efficiency of the system. With overlapping subcarriers that are perfectly orthogonal, the equalization in the receiver is made
easy, and the number of sub-carrier that can be used over a specified
spectrum is doubled as compared to FDM using guard periods [4].
The technique used to generate orthogonal frequencies in OFDM is based
upon using the Inverse fast Fourier transform (IFFT) operations. Details of this
will be discussed in later sections. A basic system layout when deploying
OFDM is shown in Figure 2.
2.3
The OFDM System Model
A simple model of a communication system consists of a source and related
coding, a well defined channel, some signal processing, and a receiver at the
end of the transmission path. The receiver decodes the received signals and
processes the input to extract the desired signal from transmission
impairments and interference. Figure 2 displays such a simplified model. It
should be noted that the necessary conversions between serial and parallel
data streams are implicitly implemented and not shown in the figure.
Channel encoder
Input
data
bit
stream
Cyclic
Prefix
h1[n]
h2[n]
hN[n]
Modulation
Mapping
Transformation
of symbols
Frequency
Domain ->
Time Domain
using IFFT
Data mapped
onto N
orthogonal
subcarriers
Parallel signals
converted
Into serial
stream
Digital/Analog
conversion
Low pass
filtering
Frequency
Carrier
Modulation
Figure 2: A simplified model of a downlink OFDM transmission
2.4
Coding and Interleaving
In [15], Coded OFDM is referred to as a concept of closely connecting error
control coding and modulation in OFDM. Coding and interleaving prior to the
IFFT transformation from data in a continuous frequency domain into the
discrete time domain, is vital for deployment of OFDM in a mobile system.
14 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Interleaving spreads out bit errors so that the receiver more easily can interpret
the transmitted data and the now loosely spaced errors. This technique is
especially important in a mobile communication system with varying radio
characteristics when using packet switched services. The data is transmitted
upon request (and potentially over several transmission paths) which creates
bursts of information bits. If an entire burst in such a transmission were lost,
the receiver would have a very difficult task to re-create the transmitted data.
The interleaving (which performs a spreading in time) allows different error
correction bits to be applied to the different errors which occurred during a
narrow interval of time to the interleaved signal; if the error affects less than the
number of bits which can be corrected, then the entire message can be
received - despite the errors.
As coding and interleaving are such fundamental components of the
information processing over fading channels in a mobile network, these ideas
are not considered specifically but included implicitly on link level when
discussing packet transfers in a mobile system such as LTE (as depicted in
Figure 3).
Source
Encoding
Speech
Processing
Encryption
Channel Encoding
Block encoding (CRC)
FEC (Turbo Coding)
Differential Encoding
(to enable interpretation of a
modulated signal)
Multiplexing
Interleaving
Modulation
Figure 3: Basic elements of a transmission chain on the physical layer
2.5
Speech Encoding in LTE
The first step in digital speech transmission is coding of the speech itself,
shown as source coding in Figure 3. Most of the available Adaptive Multi Rate
(AMR) speech CODECs used in both GSM and WCDMA/HSPA systems have
been assumed for speech services in LTE as well, providing several different
CODEC rates depending on the coding needed in different radio environments
[43]. The basic CODECs assumed for LTE are AMR 4.75, 5.9, 7.4, 12.2 for
narrowband AMR, and AMR 6.60, 8.85, and 12.65 for wideband AMR
deployment [63].
Each of these speech CODECs classifies the speech into bits of different
grades of significance, class A through C (only class A and B are used for
Wideband AMR). Each class of bits is separately coded since class A specifies
the most important bits and class C the least important bits. Erroneous class A
bits typically result in a corrupted speech frame, which is why all class A bits
are always subject to a cyclic redundancy check (CRC) to detect bit errors.
The CRC is added as an extra bit in the AMR speech frame1. Erroneous class
B bits typically do not cause serious degradation in the perception of the
speech frame, and class C bits consequently are of even lower importance.
1
A speech frame is sent every 20 ms, corresponding to the time segmentation of speech transcoders
15 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
The AMR speech frame structures are described in [43]. In addition to the
speech payload each speech frame consists of a header and auxiliary
information (for mode adaptation and error detection).
The coding of the channel carrying speech frames depends on the chosen
modulation schemes for that channel, i.e. the PDSCH (described in an
upcoming section). Although several modulation schemes are possible,
16 QAM has been chosen as a likely general modulation for data channels in
LTE systems.
2.6
Discontinuous Transmission
To conserve energy and optimize use of bandwidth, silent periods in a
conversation can be detected both on the uplink and downlink and indicated
using silence indicator (SID) frames. DTX transmission can then adjust for the
active speech intervals and transmit less information as well as avoid coding
and decoding of empty speech frames. AMR packets use SID frames
containing 39 payload bits2 [49], including information on comfort noise3.
During DTX, these SID frames are sent once every eighth speech frame
(every 160 ms). The speech frame overhead discussed in section 4.2 is added
onto both SID frames and regular speech frames, albeit the minimum header
compression size is larger for SID frames. Hence; SID frames are not
considered overhead information in an LTE system, but payload sent during
periods of no speech.
Compared to the active AMR 12.2 speech frame payload of 244 bits [49] sent
every 20 ms during speech periods, the mean bit rate would be reduced by 98
39
= 243.75
% during DTX periods, since the SID frame bitrate is 160
[bits/ms]
which equals 244 bits/s or 2 % of the bitrate for a AMR packet coded for
12.2 kbps.
2.7
The Cyclic Prefix
As previously mentioned, Inter-Carrier Interference (ICI) is eliminated if the
sub-carriers can be kept perfectly orthogonal to each other. However, in a
mobile system with constantly changing radio conditions, this orthogonality can
not be maintained. This results in both inter-carrier and inter-symbol
interference (ISI). The latter is produced when a mobile system is subject to
multipath fading where signals travel over different paths.
2
Excluding associated overhead [43].
Noise characteristics are sent to provide the illusion of a constant voice data stream. Comfort noise
prevents the user from disconnecting based on the assumption that the connection is lost when the
speaker is silent.
3
16 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
By copying a number of samples at the end of a symbol and adding these to
the beginning of the same symbol, both ICI and ISI can be avoided (see Figure
4) once the added cyclic prefix is filtered out by the receiver. However, as
shown in the lower part of Figure 5, it is vital that the length of the cyclic prefix
is appropriate with regards to the maximum delay spread or multipath delay of
the channel. For instance, if the multipath delay of the channel is longer than
the cyclic prefix the orthogonality is lost at the receiver despite these counter
measures, and both ICI and ISI are introduced when trying to interpret the
signal. Note that the channel’s characteristics are left out of Figure 5, to
simplify the example.
Cyclic prefix, Tcp
Symbol length without the cyclic prefix, Ts
Total symbol length after inserting the cyclic prefix, Ttotal
Figure 4: Adding a cyclic prefix to an OFDM symbol
Adding extra bits to an OFDM symbol consumes additional power and will
affect the user-data bitrate since more time is spent on sending the same
amount of user-data but with more control information overhead. Looking at
this at one instant in time, one realizes that when using a longer symbol time, a
smaller percentage of bits is added to the symbol in the form of the cyclic
prefix.
Considering spectrum efficiency, one would like the symbol time to be as long
as possible, given that the data throughput can remain unchanged. However,
unless the channel fading can be considered constant during one sub-carrier,
a simplified receiver design can no longer be used, since the sub-carriers no
longer can be considered narrow-band and flat-fading. Hence the symbol time
needs to be shorter (preferably much shorter) then the coherence time –
during which the channel can be considered constant – in order to still benefit
from using the cyclic prefix [4].
Transmitted
signal
Received
Signal using
correct CP
length
CP
CP
CP
Signal reconstructed
without ICI or ISI
CP
Signal reconstructed
without ICI or ISI
t
Received
signal using a too
ICI and ISI added
to the signal
ICI and ISI added
to the signal
short CP
t
17 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Figure 5: Effects of using a cyclic prefix of different lengths
The effectiveness of the cyclic prefix given that extra overhead and power is
needed, has been questioned in several studies, for instance in [6] and [7].
Herein it is investigated how to increase system performance and potentially
also spectral efficiency without adding as much control overhead as is needed
for the cyclic prefix and the necessary channel estimation information. The
channel estimation information is used to estimate the channel’s impact on the
signal, in order to apply appropriate countermeasures at the receiver. Although
considerable amounts of control data is used when adding a cyclic prefix, the
complexity of using the type of OFDM modification suggested in [6] in a mobile
system, and existing OFDM simulation tools [4], still suggest the use of a cyclic
prefix. Hence, it will be used in this thesis unless specifically stated otherwise.
The use of a cyclic prefix is also recommended in the 3GPP standardization for
LTE, where two fixed prefix lengths are defined. These are intended to
compensate for different maximum delay spreads for different cell or
transmission properties [31].
Further details discussing the usability of the cyclic prefix are presented in
section 3.3.1.
2.8
OFDM Advantages
The OFDM technique offers a number of potential performance advantages
against a system using a single frequency carrier. This section presents a
summary of the main aspects of OFDM that make it desirable in a mobile
communication system [4], [12].
•
Robustness against Multipath Fading: OFDM makes use of several
parallel sub-carriers to transmit information. More time can then be
spent on transmitting each symbol, compared to when symbols are
transmitted over a single limited frequency carrier. With a long symbol
time, potential multipath delay would impact a smaller fraction of the
symbol time, making it easier for the equalizer in the receiver to
compensate for the multipath differences. As a consequence the
receiver design can be simplified.
•
Higher Spectral Efficiency: If parallel sub-carriers can be kept
orthogonal to each other, then several spectra can overlap. This
enables transmission of more data over a fixed bandwidth without
causing performance degrading interference.
•
Robustness against Frequency-selective Fading: The available
spectrum is divided into several narrow-band sub-carriers. Potential
frequency selective fading will affect each sub-carrier’s performance
respectively. However, since the bandwidth of each sub-carrier is small,
the performance loss of these sub-carriers can be accommodated with
efficient coding.
•
Modulation & Code Rates: One user can utilize several sub-carriers.
As each sub-carrier can use different modulation techniques and code
rates, the end-user performance can be optimized in comparison to
when using only one modulation technique and one or a few code
rates.
18 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
2.9
•
Spectrum scalability: If several of the narrow-band sub-carriers are
unused, they can be allocated to other services. However, due to
interference from sidelobes of neighbouring sub-carriers, effective
filtering needs to be applied and the excluded sub-carriers need to be
contiguous and of significant numbers to prevent interference between
sub-carriers. Due to complexity, practical implementations of this
remains to be evaluated [4].
•
Simplification for MIMO: Systems planned to use flat fading channels
can utilize OFDM properties, since narrow-band flat fading channels
are deployed.
OFDM Impairments
Although the OFDM technique presents numerous advantages and has high
potential for supporting demanding packet data services – as summarized in
section 2.8 – the implementation of OFDM in a practical system such as LTE
reveals stability concerns that need to be handled (optimally) to achieve the
performance goals of LTE.
Before discussing the details of such radio channels one must be familiar with
the basics properties of a radio channel subject to different types of fading. The
type of fading mostly depends on factors such as multi-path propagation, time
dispersion, and time variance (Doppler frequency shift), all affecting the radio
channel.
Significant OFDM impairments include
•
Frequency offsets,
•
time offsets,
•
phase offsets and
•
sampling rate changes, that all impact the orthogonality,
as well as
•
a high peak-to-average ratio that reduces the power efficiency,
and
•
performance degradation from added overhead during transmission.
19 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
When a radio channel varies over time, and its characteristics are fluctuating
during one OFDM symbol period, the desired orthogonality is lost. This
reduces a sub-carrier’s useful signal and introduces inter-carrier interference
(ICI) and inter-symbol interference (ISI) (as described in section 2.7). With a
reduced signal and increased interference, the effective signal-to-interferenceand-noise ratio (SINR) is reduced. Lower SINR implies a low tolerance for
interference and noise, which likely results in lower bitrates and worse overall
performance. In addition to this, symbols sent on channels with high
attenuation are more difficult to reconstruct at the receiver. For these reasons it
is imperative that the synchronization between the transmitter and the receiver
can be maintained in time as well as in frequency to subsequently suppress
interference by restoring orthogonality at the receiver.
However, introduction of ICI and subsequent reduction of SINR are but a few
of the impairments that can reduce a mobile communication system’s
performance. Many of the OFDM implementation considerations for
transmitting and receiving a modulated signal with several parallel sub-carriers
and potential system impairments are studied specifically in [15], and some of
them will be examined in more detail in later sections, based on previous
studies. The intention is to shed some light on how to quantify power usage
and signaling overhead, so that service performance and spectrum efficiency
can be maximized without using more energy than necessary.
2.9.1
Frequency Errors
In a common radio transmitter a local oscillator and mixer is used to impose
lower frequencies onto a high frequency carrier. The receiver then reverses the
same technique to extract lower frequency content from the received high
frequency carrier. If these local oscillators do not use the exact same
frequencies, the result will be an offset in frequency.
The created frequency shift (on all sub-carriers) renders the received subcarrier frequencies no longer orthogonal, causing energy from one sub-carrier
to interfere with that on other sub-carriers. In Fourier transformation theory, this
phenomenon is referred to as DFT leakage [15]. If only one carrier would
transmit energy and cause interference due to the local oscillator frequency
offset, the sub-carrier closest in frequency to the transmitting sub-carrier would
intuitively experience the most interference. However, in most OFDM systems,
the majority of sub-carriers are used to transmit at the same time. Assuming
that the sub-carriers transmit energy - or in this context interference – in a
random fashion, the central limit theorem provides us the conclusion that the
large number of sub-carriers causing interference to the desired signal can be
considered to be additive white Gaussian noise. In order to combat the
introduced loss of orthogonality, a correction signal could be used to
compensate for the offset in the original signal. However, if the correction
factor is not of the exact same size as the original frequency offset, the
problem of lost orthogonality and introduced interference would still remain.
[15]
20 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Another frequency property concern for the equalizer is the signal level at the
receiver. Upon receiving information, the equalizer needs to differentiate
between frequency components that are of importance and those that do not
contain important information. A low signal needs to be compensated for, but
by doing so, there is a risk that a frequency component that has been lost in
transmission is interpreted as being of extremely low signal strength. A
reliability factor is then used to enable the decoding processes to determine
whether an apparently strong received signal should be interpreted or filtered
out of the information context. [15]
The reference signal discussed in section 2.15.6 is used in the LTE system to
accommodate for such channel changes along the transmission path between
the transmitter and receiver.
The impact from frequency error impairments is under investigation in the
3GPP committees. Based on the length of the cyclic prefix, a correctly defined
timing assessment of the radio channel is crucial, in order to maintain
orthogonality. If the signal, due to frequency errors, is considered outside the
timing boundaries of the system, it will not be heard at all, and its energy will
be interpreted as interference by receivers.
Timing is critical in an LTE system, in much the same way as power control is
a limiting factor in an WCDMA system. At a certain power level in a code
division based multiple access system, the introduced interference will make it
impossible to decode scrambling codes and differentiate one user from the
other. If the timing adjustments in LTE are inadequate, signals will remain
undetected and their energy will be interpreted as interference.
However, as the carrier frequency spacing currently is defined as 15 kHz,
interference between sub-carriers has been considered manageable.
Secondly, the timing issue is mostly an issue for the uplink due to lower
sensitivity at the mobile transmitters and receivers. There are several
additional concerns with regards to timing, mostly for the uplink, such as
random access signaling upon connection establishment. However, these
aspects are not considered in this thesis, but will be handled in upcoming
supplementary simulation studies in the 3GPP standardization activities, later
this year.
2.9.2
Sampling Time offset
Sampling, the process of converting a continuous analogue carrier into timediscrete values, is used to capture digital information at discrete levels
transmitted over the frequency band of the carrier.
21 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
The carrier signal is sampled and subsequently quantified into digital values at
an ideally static sampling interval. A short sampling interval or a high sampling
rate generates many discrete values, which increases the likelihood of a
correct reconstruction of the analogue signal. The Nyquist-Shannon sampling
theorem – proven in 1949 by Shannon [34] based on Harry Nyquist’s
conclusions from 1928 [32] – is also known by many other names due to many
complementary discoveries, but is commonly referred to simply as the
sampling theorem. This theorem states that in order to reproduce an infinite
periodic analogue signal exactly using discrete values, the signal has to have a
limited bandwidth and the sampling rate must be at least twice that of the
bandwidth of the signal. Problems with having a too short or to long sampling
interval are examined further in section 2.9.4.
If the transmitter and receiver would be slightly out-of-synch, a sampling time
offset would emerge. The sampling of received signal would then take place at
a different time than expected, although at a constant rate. This would mean
that the samples taken at the receiver could not be perfectly matched to an
OFDM symbol.
The use of a cyclic prefix, described in section 2.7, makes it easier to
distinguish between OFDM symbol boundaries. As long as the OFDM symbol
boundaries are maintained, a sampling time offset is equivalent to a linear
phase shift, which in most cases can be handled by the receiver (see section
2.9.3). However, as even a sampling time offset of just one sample will cause
distortion, an optimal design of the cyclic prefix length is crucial in order to
avoid both inter-symbol and inter-carrier interference.
2.9.3
Phase Offset
In addition to errors in frequency, the changes in phase of a signal also cause
offsets and loss of orthogonality at the receiver. Small phase shifts could
normally be corrected by an equalizer, but larger errors could cause errors in
bit value interpretation, since the rotation could exceed the area used to decide
symbol values. [15] This causes ambiguous bit interpretations. Phase changes
are mainly introduced due to multipath fading over the radio interface.
2.9.4
Sampling Rate Error
A sampling rate error or an offset in the sampling frequency occurs when
sampling takes place more seldom or more often than expected. In an OFDM
system with many parallel sub-carriers, a sampling frequency offset on one
sub-carrier causes inter-(sub)- carrier interference in the time domain, since
one sampling interval overlaps that of another sub-carrier.
Sampling for instance at too long intervals would in practice cause the channel
to be subject to time dispersion, and introduce a risk of aliasing or spectrum
distortion, as predicted by Nyquist in [32].
Regardless of the type of sampling frequency offset, energy from one subcarrier is interfering with other sub-carriers just as with frequency errors
described in section 2.9.1.
22 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
2.9.5
A High Peak-to-Average Ratio
A processed OFDM signal (using Fourier analysis) can be approximated with a
large number of random components [12], resulting in a Guassian distribution
given the central-limit theorem. Components in a Guassian distribution can
individually have large peaks resulting in a large dynamic range and a high
peak-to-average ratio. This puts high performance requirements on the system
amplifiers as well as signal converters in the receiver [4], [12]. If the received
signal level is so high that receiver amplifiers or the digital-to-analogue
converter are saturated, the signal will be distorted. This distortion introduces
non-linearity into the OFDM symbol properties, which in turn causes the bit
error probability to increase, and introduces interference outside the intended
dynamic range of the receiver, due to added higher frequency harmonics [6],
[12].
To reduce implementation costs and system complexity it is therefore of utmost
importance to reduce the Peak-to-Average Ratio (PAR) when using advanced
modulation techniques to enable higher data rates and greater spectrum
efficiency.
To combat a high PAR on the downlink, sub-carriers that do not need to send
information can be left empty. Thus, no unnecessary energy is added to the
transmitted signal.
2.9.6
Reference Signals
The use of a cyclic prefix to combat ICI and ISI requires that additional bits
representing channel state information and the prefix itself are transmitted as
shown in Figure 4 on page 17. In addition to the cyclic prefix, channel
estimation information is also needed in order to estimate arrival times of
received symbols. For this purpose, so called pilot symbols (referred to as
reference signals later on in this thesis) are added using a pattern known both
to the transmitter and the receiver, thereby allowing the receiver to estimate
the channel impacts on both phase and frequency on the transmitted symbols.
However, as the pilot symbol positions cannot be used for payload information
there is increased overhead. As a consequence, the placing of these pilot
symbols becomes crucial for the throughput performance of the system.
Further details on necessary control information and its impacts on service
performance are discussed in section 2.15.6.
For FDM systems that cannot rely on time-shifts between users over one
channel (as is done in TDM systems), a feedback-loop is needed in order to
send channel information and parameter adjustment data to and from the
transmitter and receiver [21]. However, if the inherent delay in the system is so
long that a parameter change is made based on no longer valid channel data
(the fading characteristics have changed and the channel is longer flat fading),
this will result in sub-optimal parameter adjustments. This reasoning makes
an optimal use of channel information overhead data even more crucial for
efficient system utilization.
23 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
2.10
LTE Network Architecture
In an LTE system, the base station controller of a GSM system and the RNC of
a WCDMA/HSPA system have been omitted in the system architecture.
Instead much of the functionality, such as the handling of mobility, has been
moved to the base stations. However, this requires a new interface between
the base stations. Secondly, some of the previous controller functionality has
been transferred to higher layers in the architecture, specifically into the core
network. The standardization of the details of the LTE architecture in terms of
specific core network nodes and specific functional responsibilities is still
ongoing, but a simplified model of the LTE architecture can be seen in Figure
6.
Operator Specific Services,
Internet etc.
The LTE CN
The S1 interface
The LTE RAN
Control plane information
Base station
User plane information
Base station
The X2 interface
Figure 6: A simplified model of the LTE architecture
As for the GSM/EDGE and WCDMA/HSPA systems, the user related
information (including related flow control) and other control information have
been separated into separate user plane and control plane architectures.
At this point in time it still remains to be defined whether the core network
functionality will be split over two separate nodes, and whether user plane
information and control plane information (such as mobility management) will
be functionally separated. However, this is of no relevance in a simplified LTE
system model nor to this thesis. The results from LTE core network
architecture discussions can be monitored in [32].
24 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
2.11
Radio Interface Protocol Architecture
Figure 7 describes the protocol architecture of the radio interface between user
equipment and the LTE network. The physical layer handles the physical
transport of data and the communication with higher network layers. The
transport channels are defined by how a transfer is performed and the
characteristics of that specific transfer, while the layer 2 (Medium Access
Control (MAC)-layer) control and traffic channels are categorized by their
logical content. Existing channel types and their mapping relation will be
described in upcoming sub-sections and the access method and physical
channel constellations will be addressed separately to clarify the implications
on the radio interface performance.
Layer 3
Layer 1
Control / Measurements
Layer 2
Radio Resource Control (RRC)
Logical channels
Medium
Access
Control
Transport channels
Physical layer
Figure 7: Radio Interface Protocol Architecture around the physical Layer
[30]
Figure 7 indicates the basic structure of information transfer in a mobile
system. Specific user information and associated flow information is defined on
a logical user plane and common control information is defined on a logical
control plane.
2.11.1
User Plane Protocols over the radio interface
The user plane architecture illustrated in Figure 8 identifies the protocols used
for the transfer of user data, including related flow information. The Radio Link
Control (RLC) and MAC protocols applied over the physical channel define the
transfer between a mobile and the base station over the radio interface4.
Figure 8: User plane architecture for LTE [41]
4
In later stages of the (release 8) 3GPP standardization, the end-point of the PDCP protocol has been
moved to the eNodeB (the base station), affecting the distribution of L1/L2 signaling and related
overhead. However, this change is not accounted for in this thesis.
25 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
The functionality defined on layer 2 in the radio interface architecture includes
Hybrid Automatic Repeat ReQuest (HARQ), multiplexing, scheduling and
priority handling (handled by the MAC protocol); segmentation and ARQ
(handled by the Radio Link Control (RLC) protocol); as well as header
compression/decompression (using Robust Header Compression (ROHC))
and encryption using the Packet Data Convergence Protocol (PDCP) [42] (see
Figure 9:). The encryption of control plane signaling can be performed in either
the RAN or the core network depending on the type of control signaling.
Figure 9: The Structure for Layer 2 on the downlink [31]
For the downlink an asynchronous HARQ is currently assumed in ongoing
3GPP discussions, meaning that the downlink scheduler can select when to
transmit retransmissions without having to notify the receiver in which frame
the retransmission will occur. Consequently the RLC protocol will have to
reorder incoming packets whenever they arrive out of order. However, on MAC
level a one bit synchronous HARQ can be used as feedback as to whether the
previous transmission was successful or not. This assumption will be used
when discussing the control data overhead in upcoming sections.
On the RLC layer, one can choose to acknowledge transmissions or not
depending on the type of service (radio bearer) and the assumed reliability of
the link. Services with requirements for low delays such as VoIP could be
transmitted without acknowledgements, while a more delay insensitive service
using larger packets (such as TCP traffic) could use acknowledgements for
each packet. The RLC layer performs segmentation, and if necessary
concatenations on packets from higher layers, creating RLC Packet Data Units
with specific sequence numbers. If the radio environment should worsen
considerably, further segmentation of the RLC PDUs is possible on the RLC
layer. This process including the header compression stage is schematically
described in Figure 10 below.
26 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Prior to the standardization discussions on LTE, the RLC PDU, was defined as
having a fixed length. However, to enable high service reliability and low delay
without MIMO or very high modulation schemes such as 64 QAM, a flexible
RLC PDU size is needed. In order to avoid reduced link adaptation caused by
stalling RLC windows when higher data rates are applied, the PDU sizes
should be increased, but the extra overhead in terms of excessive padding
needed to match the increased PDU sizes can be avoided by using flexible
PDU sizes [51]. Ericsson, Nokia, and Samsung suggest in [58] an
improvement that secures RLC header optimization even though RLC
performs concatenation of RLC PDUs of flexible sizes.
Higher Layer PDU
Radio Bearer 1
header
PDCP
(Header Compression
& Ciphering)
PDCP
header
RLC
(segmentation &
concatenation)
Higher Layer Payload
header
H
Higher Layer Payload
header
PDCP
header
PDCP SDU
RLC SDU
PDCP SDU
PDCP
header
RLC SDU
MAC
header
PDCP SDU
RLC SDU
RLC
header
RLC PDU
RLC PDU
RLC
header
RLC PDU
MAC
header
MAC SDU
MAC PDU
PHY
Higher Layer Payload
H
H
RLC
header
MAC
(multiplexing)
Higher Layer PDU
Radio Bearer 2
Higher Layer PDU
Radio Bearer 1
MAC SDU
MAC PDU
Transport Block
CRC
Transport Block
CRC
Figure 10: The flow of user data in a downlink transmission [41], [31]
The RLC layer is also responsible for adjusting for errors in the synchronous
one-bit HARQ acknowledgements on the MAC layer. Acknowledgements could
be switched (a NACK instead of an ACK), or they could be misaligned in time
(thus misinterpreted), causing retransmissions of unacknowledged RLC PDUs.
If a NACK would be interpreted as an ACK there would on the other hand be
too few retransmissions, resulting in an erroneous packet.
The RLC PDU can be multiplexed in time on the MAC layer for one radio
bearer for parallel transmissions or if a retransmission is needed. This means
that several radio bearers with separate RLC PDUs can be multiplexed using
the same MAC header. The sequence number for each RLC PDU can be used
to make sure that a correct reassembly takes place at the receiver. The size of
the transport block on the physical layer is flexible. Its boundaries have yet to
be set in the 3GPP standards, but a flexible size accommodating at least a
small VoIP-packet and an Ethernet frame of roughly 1500 bytes payload is
assumed.
27 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
As the modulation and coding can be altered for different types of channels
and services (or data streams if MIMO is deployed), the transfer and mapping
of information between the L2 packet data units and the resource blocks on the
physical layer need to be investigated. L1 overhead, channel coding (using
turbo codes first descibed in [52]), and HARQ processing is applied prior to the
mapping onto numerous assigned resource blocks, done by the resource
scheduler. The scheduler subsequently decides on a common modulation
scheme. The process will be examined further in an upcoming section (see
Figure 12 on page 32). Following this procedure the data is subject to
processing required for TX diversity, beam forming, and potential MIMO
configurations. OFDM modulation is then applied to each of streams to be sent
to the transmitting antennas.
2.12
Protocol Overhead Considerations
Figure 10 described how a transmitted IP packet is structured prior to
transmission via the physical layer. Each protocol stage adds header
information that is included in the transport block being subject to a cyclic
redundancy check.
For speech services that require low delays, the RTP protocol is used along
with UDP [38], [59]. In this thesis, RTP header information, the fixed UDP
header, and the IP header are all together assumed to occupy 40 bytes in
uncompressed mode.
PDCP enables compression of headers down to three bytes for most
continuous flows, although the header compression in this thesis is assumed
to be 5 bytes during DTX periods [44]. These three bytes are then assumed to
include RTP, UDP and IP headers. Adding the PDCP protocol overhead for
being octet aligned, adds another two bytes to the PDCP layer. However, use
of fixed header sizes in PDCP is still being discussed in 3GPP for specific
bearer services, such as a potential header size reduction for delay sensitive
services such as VoIP.
Considering that several higher-layer resources can be concatenated and the
potential re-arrangements needed for packets received out-of-order, the RLCoverhead can in general be assumed to consist of identification, resegmentation, and reassembly information. For simplicity, the RLC layer is in
this thesis assumed to handle only one block from the PDCP-layer, although
multiplex of several blocks would be possible given Figure 10 and ongoing
3GPP discussions.
Recalling the discussions from [38] of necessary basic information contained in
most protocol headers for identification, function, sequence number, and
potential flags for out-of-order reassembly, the RLC header in this case could
be assumed to be slightly more than 2 bytes long. As octet alignment is
assumed on RLC/MAC level (and the MAC header later is defined in multiples
of bytes), the exact number of RLC header bits is disregarded, and we will
assume three bytes.
Since RLC PDUs from different bearer services can be multiplexed on MAClevel, the protocol overhead added is based upon their length and the level of
multiplexing. Consequently, to accommodate this, another two bytes are
assumed, leading to an overhead of five bytes.
28 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
2.13
Multiplexing & Multiple Access
The definitions of future systems in the 3GPP standards make it possible to
deploy LTE over various parts of the spectrum, as well as using variable
bandwidths in adjacent cells or for different mobiles depending on spectrum
allocations and service needs. The available spectrum bandwidths range
between 1,25 MHz and 30,70 MHz ([4], [35]) at different locations of the
available spectrum. The more bandwidth allocated, the more sub-carriers can
be deployed for downlink transmissions. However, currently the minimum
bandwidth requirement for capability in the LTE mobiles is 20 MHz [31].
In the 3GPP LTE implementation of OFDM, a paired spectrum is supported by
applying Frequency Division Duplex (FDD), enabling a separate frequency
band to be used for multi-carrier downlink and single-carrier uplink
transmissions.
The physical Layer in LTE uses a multiple access technique for the downlink
based on OFDM called OFDMA, along with the use of a cyclic prefix. As the
resource structure in Figure 11 on page 31 suggests, the frequencies are
reused over time implying that time division multiplexing also is applied.
Hence, resources are divided and shared in frequency as well as over time.
This is the same concept as the existing GSM system where separate
frequency spectra for uplink and downlink are allocated to multiple users over
the same instances in time.
2.14
LTE Resource Blocks & Resource Elements
The standardized generic radio frame in an LTE downlink transmission is 10
ms in duration. Each radio frame is further divided into 20 slots, with a
duration of 0.5 ms each. However, handling such small units on each subcarrier of a bandwidth of 15 kHz would lead to considerable control data
overhead in a transmission, which is why the concept of resource blocks was
introduced [35]. A Resource block is a sub-band, a group of sub-carriers
(defined to be 12 in [30]) used by one user for the duration of the slot. The
maximum number of resource blocks over a given spectrum is referred to as
NRU below. The signaling structure for LTE, discussed in later sections, is
based on the slot duration of 0.5 ms (as shown in Figure 11), although the
smallest TTI in recent 3GPP discussions has been determined to 1 ms, as a
sub-frame consisting of two slots. This means that at least two resource blocks
need to be used by each user during one TTI. For uplink transmissions these
two resource blocks can be shifted in frequency so that frequency hopping can
be applied. On the downlink the resource blocks need to be consecutive in
both time and frequency [35] .
29 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
The maximum number of sub-carriers equals NRU*12+1, divided equally
among NRU resource blocks. The extra sub-carrier represents a fictive DC
sub-carrier in the middle of the used spectrum. This sub-carrier is neither used
nor transmitted, but used as a reference for some of the control signaling in the
frequency domain. The resource blocks are assumed to be of constant size in
current standardization discussions, which results in half of the resource blocks
being distributed on either side of the DC carrier. If the number of resource
blocks would be odd, this would result in the DC carrier splitting one resource
block. The feasibility of such a solution where you would not always know the
frequency position of a resource block, has yet to be fully investigated and
decided upon in 3GPP standardizations. Even numbers of resource blocks will
be assumed for the remainder of this thesis.
With spectrum allocations of different sizes, resource blocks at each end of the
spectrum could be truncated. In order to use the entire spectrum and maximize
the spectrum efficiency, one option would be to allow resource blocks of
different sizes. However, since the 3GPP standards have yet to decide on the
issue (as of December 2006), this document will assume resource blocks of a
single fixed size unless specifically stated otherwise.
An additional radio frame structure different from frequency division
multiplexing and multiple access exists for time-division multiplexing systems,
where un-paired spectra are used for the up- and downlink transmission.
However, that structure will not be referred to further within this thesis, nor will
the related changes to the length of the cyclic prefix in a TDD system. Instead
the FDD technique using paired spectra for the different links will be assumed.
Each transmitted FDD signal consists of one or several sub-carriers Nsc of 15
KHz bandwidth each, and a number Nsymbol of OFDM symbols. Each
separate symbol interval on each sub-carrier is referred to as a Resource
element in the 3GPP specifications [32], [35]. The relation between resource
blocks, sub-carriers, OFDM symbols and resource elements is described in
Figure 11.
30 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Cyclic Prefix
f
Resource block
Resource element:
1 symbol
on
1 sub-carrier
NSC sub-carriers
7 OFDM symbols
t
0.5 ms
slot
Figure 11: Downlink Resources (Normal Cyclic Prefix Length)
The mapping above displays seven OFDM symbols in a resource block. The
3GPP standards stipulate the use of either six or seven OFDM symbols per
resource block depending on the length of the cyclic prefix. The two fixed
prefix lengths are configured by higher link layers depending on the multipath
delay spread of each OFDM symbol. The extended cyclic prefix (applied within
resource blocks with six OFDM symbols), is used in geographically larger cells
with a larger delay spread or for the purpose of multi-cell broadcasting [4].
Noticeable for smaller delay spreads applying the normal cyclic prefix, is that
the cyclic prefix length can have two predefined values, within the same
resource block. [35].
As a result of the resource structure displayed above there will be 14*12
resource elements available for use during each TTI of 1 ms when the normal
cyclic prefix length is used. These resource elements can and will carry
payload data along with physical signaling information and the signaling
required for resource allocation.
2.14.1
Transport Block to Modulated Signal
In case of a MIMO configuration where there are several streams of data in
transmission, the process below of mapping transport blocks onto resource
elements will be applied in parallel. Currently it is possible to deploy no more
than two parallel processes as shown below, regardless of whether two or
more parallel data streams are transmitted.
31 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Transport block
CRC
Channel Coding
HARQ
&
Adaptive
Coding
Rate
Generation of complex values
(scrambling)
Physical
Channel
Mapping
onto
Resource
Units
….
Selection of a common modulation
Resource unit assignments
Figure 12: Basic Modulation and coding of Resource blocks [27]
2.15
Channels and Signals
The LTE network and protocol architectures were discussed in
previous sections, along with the basic resource structure of LTE. The
upcoming investigation of different types of channels and signals in
LTE will examine the details of information distribution, given these
structures.
2.15.1
Logical Channels
Control channels are used for to describe logical control information content on
the MAC-layer above the physical layer. The control channels are:
•
Broadcast Control Channel (BCCH), a logical downlink channel for
broadcasting system control information.
•
Paging and notification Control Channel (P(N)CCH), a logical downlink
channel transferring paging information or MBMS notifications when a
mobile is to be located.
•
Common Control Channel (CCCH), a channel used by mobiles having
no radio resource connection with the network. At this point in time it
has yet to be decided whether the network access mechanism should
be handled on the CCCH or via L1 signaling. CCCH will be used for
logical cell access content if the random access channel is defined as
an uplink transport channel.
•
Multicast Control Channel (MCCH), transmitting downlink scheduling
and control information of to several users of MBMS.
32 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
•
2.15.2
Dedicated Control Channel (DCCH), a channel used in both uplink and
downlink for dedicated control information once a radio resource
connection between a mobile and the network has been established.
Traffic Channels
Traffic channels are used for transferring payload information. The traffic
channels offered by layer 2 are:
2.15.3
•
Dedicated Traffic Channel (DTCH), used to transfer user information to
and from a mobile in dedicated mode.
•
Multicast Traffic Channel (MTCH) used to transfer user information to
several mobiles using MBMS.
Transport Channels
The logical content of a physical channel is defined on a higher layer as logical
channels, and should not be confused with transport channels. The
characteristics of a transfer and how it is performed define a transport channel.
The physical transport channels used for downlink transmissions and some of
their characteristics as defined in [31], are presented below;
•
The Broadcast channel (BCH). This channel has pre-defined transport
format that cannot be altered, and is required to be used over the entire
area covered by cell in LTE.
•
The Downlink Shared Channel (DL-SCH). Supports several Forward
Error Correction techniques and dynamic link adaptation, since the
modulation coding and transmitted power can be altered. Different
types of resource allocations are also supported. These are discussed
in a later section. This transport channel and its related control
information are mapped onto the physical channel Physical Downlink
Shared Channel (PDSCH).
•
The Paging Channel (PCH). This channel is mapped to physical
resources that can be used dynamically for other control channels or
for traffic channels.
•
The Multicast Channel (MCH). Used when transmitting information to
several simultaneous users, creating the illusion that the same data is
transferred from several cells, improving the reception conditions. [30]
Recalling Figure 12 it should be noticed that creation of complex values based
on the symbols can be applied differently depending on the type of the
transport channel. The broadcast and paging channels will have cell specific
scrambling since they are intended for the entire cell. The downlink shared
channel can have either a specific scrambling applied related to one single or
a group of users. The MCH channel used for MBMS purposes will use
scrambling that is specific to a cell or a group of cells intended for the MBMS
transmission.
33 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
2.15.4
Mapping between logical channels and transport channels
Figure 13 depicts the planned mapping between logical and transport channels
(the shaded arrows are suggested mappings that have yet to be decided in the
3GPP standardization fora):
Figure 13: Suggested mapping between logical channels and transport
channels [35]
The mapping onto transport channels has been introduced to distinguish
between the type of channel content and the characteristics of the channel
itself (with regards to modulation etc).
However, for this thesis, it is mainly the logical information content and the
usage of the physical channel that are of interest, conveying the payload and
related control information.
2.15.5
Physical Channels and Signals
The physical layer in an LTE system can provide transfer services to higher
layers and related services. There are currently three types of physical
channels defined for downlink transmission [35];
•
Physical Downlink Shared Channel (PDSCH). This channel is used for
data transmission and paging. It can utilize several modulation
schemes (QPSK, 16 QAM, 32 QAM, or 64 QAM).
•
Physical Downlink Control Channel (PDCCH)5. This channel will handle
L1/L2 control information such as downlink scheduling information and
uplink scheduling grants, as shown in Table 1 and Table 2.
•
Common Control Physical Channel (CCPCH). This channel will carry
broadcast information and can currently be implemented using only the
modulation scheme QPSK. The channel is mapped onto resource
elements over 72 sub-carriers, centered around the DC sub-carrier,
once every radio frame (not depicted in Figure 11).
The physical channels are created to convey payload and control information
user data.
5
Transmitted on up to the three first OFDM symbols in every subframe (those not used by the
reference signal [35]).
34 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
A potentially additional channel to those presented thus far, conveying HARQ
ACK/NACK responses, has been under discussion in 3GPP. The exact details
of the implementation of physical channels and signals has yet to be defined in
the 3GPP standards [35].
As was discussed in section 2.15.4 for logical and transport channels, the
number and exact definition of channels defined is not necessarily relevant
when investigating the usage of physical channels, as long as a channel’s or
signal’s relative energy consumption and impact on service performance can
be quantified.
Not all resource elements in a resource block can be used for information
transmission, since broadcast information and the two physical signals
reference signal and synchronization signal (comprised of a primary and
secondary signal) have to be allocated specific resource elements in a
resource block. These signals are not mapped onto any physical channel, but
should be handled separately. Further discussions of the physical channel
mapping have been omitted in this thesis to concentrate on the signals
transmitted and the different modulation techniques applied to these signals.
For simplicity, the physical channel notation will however still be used,
assuming four physical channels including one for HARQ ACK/NACK
feedback. These semantics should not be interpreted as standardized channel
concepts, unless stated otherwise.
To save energy, resource elements that are unused (neither carrying a physical
channel nor a physical signal) contain no energy [35] .
A physical downlink transmission is prepared in four steps, as partly shown in
Figure 2; after scrambling bits are modulated so that symbols of complex value
are generated [35]. These complex symbols are then mapped onto resource
elements, and finally the complex-valued OFDM signal is generated for each
physical antenna.
2.15.6
Control Information on the Downlink
As previously indicated, the necessary overhead on the downlink in LTE can in
general be summarized as;
•
broadcast information on the broadcast channel,
•
primary and secondary synchronization signals,
•
control information on layer one and layer two
•
downlink scheduling information, and uplink grants
and
•
one reference signal per transmitting antenna.
35 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
For each of the potentially allocated OFDM spectra, there is a guard-band
applied at each end to defend against adjacent spectra with interfering out-ofband or adjacent frequencies. For a spectrum allocation of 5 MHz a guard
band of 10 % leaves 4.5 MHz or 300 sub-carriers of 15 kHz bandwidth each, to
be used by sub-carriers transmitting payload and control information each 1
ms.
The broadcast channel, uses 72 sub-carriers centered around the DC carrier
for 14 symbols once every radio frame of 10 ms. Remembering the resource
element and resource block structure from Figure 11 on page 31, this results in
168 resource elements per resource block being used for every 10 ms. This
also means that 24 percent of the bandwidth is transmitting broadcast
information during ten percent of the time (2.4 % of the total bandwidth is used)
when 4.5 MHz is allocated for sub-carriers excluding the guard spectrum.
The same constellation with 72 sub-carriers centered around the DC-carrier is
utilized for the synchronization channel, consisting of two sub-signals; the
primary and secondary synchronization signal. These synchronization signals
however span over two symbols for two TTIs during one radio frame, instead
of over all symbols as the broadcast channel. The primary signal is used to find
the timing of the synchronization channel, but as the receiver could shift
slightly in frequency, the second secondary signal is used to remedy this
potential error as well as to provide some initial cell information.
The signals are transmitted over two symbols at two different TTIs during one
radio frame [35]. This results in ~0.69 % of the bandwidth when 4.5 MHz is
allocated for the sub-carriers (see below).
⎛ SynchSymbol perframe × 2 ⎞
BWSC × 72
⎟
×⎜
BWtotal − Guardband ⎜⎝ Symbols perTTI × 10 ⎟⎠
(eq. 2)
L1/L2 control channel information is transmitted in the beginning of each TTI
(in resource elements not used by reference signalling) and provides
information on who is to receive the next resource block and when and what
kind of modulation is applied to that specific resource block. In the downlink,
the L1/L2 control channels are also used to provide resource block information
related to the uplink. Thus, the control information on layer one and layer two
altogether convey downlink allocation details, information on the size of the
transport blocks, modulation information, and scheduling of downlink
transmissions. It also conveys grants for and acknowledgements of uplink
transmissions [27]. Given the structure of resource elements the L1/L2 control
channel overhead depends upon the number of scheduled users using the
downlink as well as the uplink.
The amount of transmitted control information needed for scheduling depends
on the number of users in the network as well as how often transmission as
scheduled. Which scheduling procedure to chose is a very complex issue that
could require a thesis in itself, which is why the scheduling options discussed
in this paper primarily will focus on dynamic scheduling and persistent
scheduling, although there are numerous variants thereof.
36 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Dynamic scheduling allocates resources at every transmission interval, every
TTI, depending on channel requirements. When using small packets and
allocating resources for every TTI whenever there is data available, the
overhead will be considerable keeping in mind the short TTIs of 1 ms used for
LTE. However, if DTX is deployed for a speech service, resources would not
need to be scheduled for the periods in-between speech bursts, excluding
silence indicator frames (SID-frames). This could significantly reduce the
scheduling overhead.
Persistent scheduling - as the name implies - is a more static resource
allocation method, allocating resources as long as there are bits to send,
reducing the need for control signaling information. However, in the most
extreme case of persistent scheduling, the scheduling information is sent once
and resources are fixed during a pre-defined time interval; hence they cannot
be changed based on altered channel conditions and coding rates. Potential
retransmissions have to be accounted for during scheduling, which could result
in that some of the scheduled resources never or very seldom will be used.
The scheduling signaling described in Table 1 and Table 2 (resource
allocations, grants, and HARQ acknowledgements) will be required for each
transmission when dynamic scheduling is applied.
During persistent
scheduling, HARQ acknowledgements are still transmitted every TTI, while the
additional scheduling signaling can be transmitted less frequently.
Dynamic and persistent scheduling can be considered two extreme scheduling
algorithms in terms of use of scheduling resources, both having advantages
and disadvantages. The most likely scheduling scenario is presumably
somewhere in-between, although outside the scope of this thesis. The
deployment of DTX can be used to illustrate this statement; dynamic
scheduling schedules resources every TTI, but when DTX is activated there is
no need to allocate resource during DTX intervals of no speech activity (not
considering SID frames). If dynamic scheduling is assumed, DTX periods of no
speech would significantly reduce the necessary percentage of scheduling
information, as discussed in section 4.3. The impact on total overhead of using
either of the two scheduling algorithms, will be presented.
There are discussions ongoing in 3GPP to try to solve the obvious drawbacks
of both dynamic and persistent scheduling methods (for instance to verify
retransmissions before allocating resources with persistent scheduling), but as
mentioned details on the subject are outside the scope of this thesis.
The reference signal provides a reference in order to obtain coherent
demodulation at the mobile receiver. The signal is also used for channel quality
estimation to improve link adaptation and scheduling that is dependent on the
quality of the channel. It is transmitted once every 6th resource element in the
frequency domain and on two out of seven OFDM symbols in the time domain
(the first and third last OFDM symbol per each slot interval of 0.5 ms with an
offset from the first reference symbol in the slot of three sub-carriers) [35]. This
results in one resource element out of 21 being used for the reference signal in
a configuration using the normal cyclic prefix without downlink diversity. Thus
leading to an overhead of about 4.8 percent.
37 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
One reference signal is used for each downlink transmit antenna, and 3GPP
standards allow 1, 2, or 4 downlink transmit antennas. For two downlink
antennas this results in a 2/21 (~9.5) reference signal percentage. From [35] it
can also be concluded that half of these amounts are added when using three
of four antennas (5/42 (~11.9) and 3/21 (~14.3 %) of overhead for the
reference signals, see Table 3) .
Discussions in the 3GPP standards are currently evaluating whether to
suggest a power boost on reference signal information whenever applicable.
Stronger reference signals in the downlink would enable better coverage or
lower sensitivity requirements on mobile receivers.
Increasing the power of the reference signals by for instance 3 dB would mean
that twice as much power is spent on conveying control information using 1/21
of system resources in a single antenna configuration (see Table 1).
(1 − commonOH ) *
Let
Power
speechframe
(eq. 3)
represent the power that is used to convey payload and speech frame
overhead. CommonOH represents energy spent on reference signals,
synchronization signals and broadcast information.
refsignal
Secondly, let commonOH (eq. 4) represent system overhead power for
reference signals. A 3 dB increase in reference signal power (twice the energy
level) would then mean that the relative energy spent on commonOH would
increase by said factor. The power left for payload and related overhead would
then decrease by a factor
⎛ ⎛
⎞
refsignal ⎞
⎜1 − ⎜1 +
⎟ * commonOH ⎟
⎝ ⎝ commonOH ⎠
⎠
1 − commonOH
(eq. 5)
if the available power resources remain constant (see Figure 15).
The signal/-s is/are mapped onto the resource elements as shown in Figure
11, depending on the length of the cyclic prefix. Since the extended cyclic
prefix uses a different number of OFDM symbols per TTI, the placing of the
reference symbols in the time domain is adjusted accordingly, for instance
when the system is designed to handle larger cells and more extensive
multipath delays.
In a one-antenna transmit system every 6th or 5th resource element is used
for the reference signal, when applying a normal or extended cyclic prefix
respectively. For systems using two or more transmit antennas, the mapping of
the reference signal onto resource elements and its impact on system
performance is more complex as described in [35].
38 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Downlink paging has not yet been standardized; hence this has been omitted
from this thesis in order to simplify our analysis. However, one could assume
that the structure used for transferring paging messages is similar to that of
broadcast information, although this does not mean that the paging load will be
the same. Paging load is heavily dependent on traffic load and the core
network and base station capacities to convey paging commands.
2.16
Control Channel Transmit Diversity
Potential downlink diversity gains for control channels are addressed
specifically in [27], since diversity improvements for control channels
specifically, through retransmissions and link adaptation, are deemed difficult.
Although a number of open-loop diversity techniques are suggested for further
evaluation, cyclic delay diversity is assumed for L1/L2 control information using
either two or four downlink transmit antennas.
2.17
MIMO
MIMO enables the use of the use of multiple streams and improved diversity.
In addition to a basic MIMO configuration of two antennas at the transmitter
and one or several at the receiver respectively, four Tx antennas deployed
using either two or four RX antennas should also be considered as plausible
MIMO configurations to obtain higher diversity gains [27].
39 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
3
Previous Studies
3.1
Expectations of LTE
A number of techniques to achieve the proposed performance figures using
the LTE concept are proposed and evaluated in [18]. This report also examines
new radio transmission technologies intended for deployment beyond LTE.
The latter can hopefully simplify the suggested technical solutions for LTE even
in future radio access networks.
Although the suggestions mainly are focused on LTE, the performance of
some WCDMA/HSPA improvements is shown to be equal to LTE. However,
the LTE improvements for layer one, including broadcasting and spectrum
flexibility, are still advantageous [27], [30].
For a basic description of suggested Long-Term Evolution improvements, see
section 1.2 and [19]. Further details and historical background on the ongoing
LTE standardization can be found in [28], [20], [31], and [32].
3.1.1
Frequency Reuse
Although the use of a specific frequency has been the focus of many studies
on frequency reuse designs historically, frequency reuse is not directly related
to how a channel can be allocated in an LTE mobile system. LTE is targeting
packet data users that can use different amounts of the available bandwidth
among the users competing for the same resources. Constant levels of
interference and fixed allocations of specific resources, as with circuit-switched
voice calls in a GSM system, are no longer applicable.
There are a number of different frequency reuse patterns that can be used in
an LTE radio access network, and the usefulness of these frequency allocation
schemes all depend on the amount of available spectrum, the allocation or
scheduling strategies (how efficiently the spectrum is utilized), the use of
power, and the system’s ability to handle or suppress the interference
introduced. In [11], a number of variants of classical reuse patterns, adjusted
for an LTE RAN with packet-switched services, are discussed and evaluated.
A frequency reuse of 1, i.e. using the same frequencies in all cells, is shown to
be most effective among reuse patterns considering only downlink
performance. However, interference between the cells in such a scenario is an
issue that needs to be dealt with. Inter-cell interference will affect resources
that can be scheduled successfully. Hence, the useful bandwidth or the
maximum output power is reduced.
The total system throughput is estimated for two use-cases in [11]; one where
the users use different amounts of data and one where the users consume
exactly the same amount of data but during an extended time period. For the
first case, the system throughput will mostly be used by users experiencing
good radio conditions. The latter case corresponds to a service where files of a
specific size are downloaded.
40 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
The estimated link bitrate is measured from two perspectives; when the
system is used for a constant period of time, and when a specific amount of
data is downloaded over different time periods. However, in the case where
users are evenly distributed over a larger area, the fixed-data scenario is not
particularly representative. As users experiencing bad radio conditions need to
stay longer in the system to complete a transfer, even more users will be
gathered in areas with poor radio conditions and hence suffer more from
downlink interference.
Two different antenna configurations are evaluated, using a single antenna
and MIMO (2*2). Transmission diversity was used on the downlink, and upon
reception of the signals, the combining techniques MRC and IRC have been
used.
3.1.2
Peak-to-Average Reduction (PAR)
In [12], solutions on how to reduce the high inherent peak-to-average ratios in
an OFDM system are investigated. As highlighted in both [12] and [6], the
definition of the PAR is the maximum of the absolute value squared, divided by
the expectation of the absolute value squared:
2
PAR{ y (t )} =
max{ y (t ) }
(eq. 6).
2
E{ y (t ) }
From the OFDM signal representation during one general symbol interval n as
defined in [6]:
y (t − NT ) =
1
N −1
∑c
T
m =0
m
(n)e j 2πf m ( t − nT ) ,
(eq. 7)
where t is the time instant, N the number of sub-carriers, Cm the symbol, and T
the period,
2
and the signal expectation E{ y (t ) }
1
T
N −1 N −1
∑∑ E{c
M =0 l =0
m
1
(n)c1 (n)}e j 2πj ( f m − f1 ) t = N ,
T
(eq. 8)
it can be concluded that the level of the PAR is directly proportional to the
number of sub-carriers N. The more sub-carriers used in a transmission, the
larger the PAR will be at the receiver.
However, this conclusion relies on the fact that information symbols are
perfectly uncorrelated in time and that the symbol energy is of constant
amplitude but varying phase [6].
41 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
3.1.3
Downlink Power Consumption
In [3], different approaches are investigated to describe the energy transmitted
in the downlink. The first technique defines an interference area based on the
Okumura-Hata propagation model in a GSM system. The relative size of the
interference area is used to indicate the introduced interference in downlink.
Another measure defines the transmitted energy based on how much of the
available channel quantity is used.
M. Ericson, et al. describes an equation that defines the total power used in a
downlink in a WCDMA system [36], and S. Wänstedt et al., concludes that
services using very small packets of data such as VoIP, suffers from the
extensive overhead information in a WCDMA/HSDPA system [37].
Retransmissions are considered with regards to their impacts on capacity and
the carrier-to-interference ratio. However, it remains to be investigated whether
the restrictions of transferring small data packets also are applicable to LTE,
given the structure of the overhead of data transmissions.
Recalling the discussion on channel structures from section 2.11, and
discussions on resource handling within a resource block, the total power
transmitted can be summarized as power on a physical channel used to
transmit payload data along with power used to transmit control information.
To describe the total transmitted downlink energy, the transmitted power is
calculated as a function of the number of resource blocks in a cell, given all
traffic and control channels used during a TTI. It is assumed that all base
stations use the same total power in the downlink and that the system is
interference limited. The latter implies that relatively small cells are used which
renders noise interference insignificant. A similar calculation is performed in
[36] to describe the distribution of downlink power in a mixed services WCDMA
network.
As the total transmitted energy is the sum of all transmissions on all subcarriers in an OFDM resource block, the total transmitted energy over a cell
coverage area from one base station bs can be described as
Ptot ,bs DL =∑ PbsRU +
N RU
∑
∑ (P ) + P
sched
bch
+ Pref + Psch + Ppaging
NUSERS SCHED N LINKS SCHED
= PPDSCH + PPDCCH + PCCPCH
(eq. 9)
where NRU is the number of resource blocks defined in a cell.
Noticeable for the paging channel in LTE is that the paging procedure is similar
to that of GSM/EDGE with defined paging groups where the handsets listen for
paging requests at regular intervals and save power during the time in
between these occasions. The downlink notification deployed in
WCDMA/HSDPA systems is not used, since the time when the paging group is
scanned for information by reading control information is very short due to the
sub-frame structure in an LTE network.
An estimated SINR target at the receiver, for the resource block RU could then
be described as follows, given that the power transmitted from a base station
bs is PbsRU:
42 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
PbsRU i G bsRU i
ΓDLi =
α RU (Ptot ,bs
i
3.1.4
DL
)G
bsRU i
+ I otherRUs,i + N 0
(eq. 10).
Scheduling
Apart from dynamic and persistent scheduling, another common scheduling
principle is the Proportionally Fair in Time and Frequency (PFTF) technique
investigated in [39]. This scheduling technique not only considers the time
used, but also the frequency allocation.
Optimized scheduling principles have yet to be agreed on for the 3GPP
standards. The advantages and disadvantages of persistent scheduling are
discussed in [54] [55], and [62]. Semi-dynamic scheduling is examined in [56],
and blind detection without impacting L1/L2 signaling is described in [57],
along with an overview of various mechanism for downlink scheduling as of
December 12 2006.
Discussions as of late February 2007 are focusing on semi-dynamic
scheduling and blind detection, with the main purpose to prevent L1/L2
signaling from restricting capacity. More references on the subject of
scheduling can be found in [61].
3.1.4.1
SINR Requirements
When assuming round-robin scheduling, the arrival probability of a packet to
be scheduled for an AMR speech frame can be described as in [36] (this also
describes the subsequent reasoning concerning SINR requirements per user
subject to scheduling):
λ =λTTI =
TTI
1ms
=
AMRFrameLength 20ms
This only describes the likelihood of a packet arriving at the scheduling queue.
The actual transmission of the packets depends on the chosen scheduling
algorithm and the load in the system in terms of multiplexed users and users in
one cell.
Given that an assumed 10 % guard band is applied to the available spectrum
and that 12 sub-carriers of 15 kHz bandwidth are allocated to each resource
block, the spectrum available for scheduling during each TTI is BW × 0.90
(BWMHz × 0.90)
MHz divided among M users where M equals
one resource block per user.
15kHz
12 , assuming
The time between each complete transmission of a packet depends on the
number of users M and the multiplexing factor mux. Thus the number packets
aggregated (when M is large) and the average sent to one user can be
described (based on [36]) as
43 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
E (n ) = λ
M
mux .
(eq. 11)
Assuming that a packet of an RLC PDU frame size requires a specific SINR,
one can assume that a packet size equal or close to multiples of an RLC PDU
will require multiples of the required SINR. The total SINR requirement can
then be estimated to be
ΓDL tot = nΓDL m = λ
M
ΓDL m
mux
(eq. 12)
as is done in [36].
3.1.5
Retransmissions
As noted in [37], it is necessary to establish a fast feedback link on layer 2
since the channel estimations made from the transmitted downlink information
will be offset by a number of TTIs.
In addition to retransmissions due to erroneous (too slow) feedback on the
channel quality, these could also occur if a mobile experiences a low SINR or
high inter-cell interference. In those cases the power available as shown in the
previous section can only cater for a fraction of the n packets that can be
transmitted using the power available. Alternatively packets could be
retransmitted a number of times and decoded by the receiver using Chase
Combining. In both [36] and [37] these two retransmission aspects are
considered equally common and the increased number of packets has been
defined as
ntotal = λ
M
(retrans feedback _ delay + βretranschase _ comb )
mux
(eq. 13)
βretrans
chase _ comb
where
denotes the fraction β of all the users that experience
bad coverage multiplied by the number of total transmissions (including the
first one) needed for a successful reception, given chase combining.
The number of possible retransmissions depends on the performance
requirements of a given service. The longer delay a service can accept, the
more retransmissions can be allowed. However, the maximum number of such
retransmissions without exceeding delay requirements are dependent on the
interval between them, which in turn is dependent upon the number of users in
the system, from equation (6) above [37]. If the delay threshold is defined in
terms of number of complete packets of a predefined size, and related to the
total number of packet transmissions ntotal the maximum number of
retransmissions (including the first one) can be described as in [36] and [37]:
retranschase _ comb =
Maximum _ delay packets
ntotal
44 (98)
(eq. 14)
Overhead Impacts on Long-Term Evolution Radio Networks
As discussed in [37], the transmission of multiple packets of a certain size
without retransmissions will require a SINR multiplied by the SINR required for
the transmission of one such packet. Considering retransmissions, we now
know that each transmission before successful decoding is defined as
1/retranschase_comb. The total SINR including retransmissions can then be
defined as the packets transmitted until successfully received multiplied by the
SINR of each transmission
ΓDLtot =
ntotalΓDLm
retranschase_ comb
=
ntotalΓDLm
=
Maximum_ delaypackets
ntotal
ntotal ΓDLm
2
Maximum_ delaypackets
(eq. 15)
From this formula it can be seen that when considering retransmissions, the
required SINR is proportional to the square of transmitted packets. It can also
be seen that additional retransmissions reduce the required SINR (left part of
the formula), as well as increase the need for SINR as more packets are
transferred (the left part of the formula). However, simulations referred to in
[36] have shown that retransmissions as defined in [36] and [37] overall
decrease the need for SINR.
The inherent delays in a mobile system have resulted in that the modulation
and coding scheme selection initially is based on an estimation of the average
link quality, rather than direct feedback from CQI information contained in
transfer and reception buffers.
3.1.6
LTE Downlink Performance
Given previous discussion during the 3GPP standardization activities, the
uplink can in most cases be considered as the limiting link, and the control
channels are potential bottlenecks in terms of LTE coverage at maximum
traffic load.
In [13] the basic LTE downlink throughput for data services using OFDM is
compared to the throughput of 3GPP release-6 HSDPA. The introduction of
more complex antenna solutions (MIMO) are also examined. Simplified
models are used for the physical and MAC layers. Potential performance
improvements such as shorter round-trip-times, based on higher layer protocol
improvements, were not accounted for.
The conclusions are that the main contributors to increased throughput for LTE
are the introduction of several data streams deployed through MIMO
configurations. In this comparative study, scheduling, link adaptation, and
power control in the frequency domain have not been included, which all could
improve the relative performance of LTE. Nor has the potentially wider
spectrum for LTE (using a larger number of sub-carriers) been considered.
The spectrum used in the study was 3.84 MHz.
45 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
3.2
3GPP LTE Standardization
3.2.1
Scheduling Overhead
As the 3GPP standardization is continuously ongoing until the third quarter of
2007, [32] should be referenced for more recent updates. However, in [27]
suggested bit structures for scheduling and feedback information are
presented, dividing the dedicated bit allocation into three different categories of
downlink scheduling (Table 1), and into resource assignment information and
transport format data for uplink scheduling grant information (see Table 2). The
synchronous acknowledgement feedback on the downlink is assumed to be
one bit per uplink transport block when using one data flow in the uplink per
user [27]. As the details of the control channel properties have yet to be
finalized in the 3GPP standardization, the following figures are only a
presentation of ongoing discussions. Further details are presented in [27].
Based on [45], the first parts of the message, including the identification of a
mobile and the transport format (see Table 1), can be assumed to be about 40
bits. Depending on the chosen scheduling mechanism, these information bits
can be sent for every transmission, for every session, or scheduled in many
other ways. The HARQ indicates the status of the earlier transmission using 6
bits and is needed for every transmission. The scheduling grant content which
is based on uplink transmission, is discussed in [41] and [46] and assumed
accordingly in these discussions, even though the final decision on scheduling
content for LTE has yet to be taken. Thus about 36 bits cater for resource
assignment of scheduling grants and additional two bits for synchronous uplink
HARQ sequence numbers. Presented bit configurations are assumed to reflect
maximum system scheduling capacity. However, the bit size fields have been
omitted in Table 1, Table 2, since the tables only present suggested formats,
leaving the details of implementation for finalized 3GPP standards.
46 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Table 1: Downlink Scheduling Information [27]
Category
Field
Resource
Indication
ID
Resource assignment
Assignment duration
Transport
Format
Multi-antenna info.
Modulation scheme
Payload size
Hybrid
ARQ
Asynch.
HARQ
HARQ
no.
process
Redundancy
version
New
indicator
Synch.
HARQ
data
Retransmission
sequence no.
Table 2: Uplink Scheduling Grant [27]
Category
Field
Resource
Assignment
ID
Resource
assignment
Assignment duration
Transport Format
Transmission
parameters
47 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
3.2.1.1
Transmission Schemes
In [16] two types of LTE downlink resource allocation schemes, block-wise and
scattered transmission, are handled based on previous related discussions in
the 3GPP standardization body. A block is here defined as the number of subcarriers dedicated to one particular user. The discussion is focused on how to
optimize the use of consecutive sub-carriers and OFDM symbols for each
user, the design of diversity transmission, multiplexing, and how to maximize
the efficiency of signaling overhead. If more users are sharing the available
sub-carriers, the CQI overhead increases. This paper also concludes that
neither channel dependent scheduling nor link adaptation is suitable for very
fast channel variations, e.g. when mobile handsets are used in cars going on a
highway. The data rate performance is reduced when more high-speed users
are introduced into the system. The user with the highest SNR simply cannot
be selected if the channel varies too fast to be estimated, disabling diversity
gains over several users at high speeds. Trying to remedy this situation, a
transmission scheme, aiming at maximum time- and frequency diversity for a
single packet transmission, is presented. Block-wise and scattered
transmissions are compared, and simulation results show that diversity
transmissions per sub-carrier, provides better performance than using blockwise diversity for high-speed users. For low-speed users the results are
reversed, showing high throughput gains for block-wise transmission.
However, an interesting observation is that the signaling overhead indicating
the resources to use for transmission remains the same.
Although these two methods (in simulation) show conclusive results for these
transmission techniques with users at different speeds, the former, also known
as localized transmission has been agreed on in the 3GPP standardization
due to extensive time dispersion when using scattered transmission.
3.3
Potential Standardization Improvements
3.3.1
Modulation Improvements
In [4] the OFDM modification CP-OFDM is used. This technique, further
described in section 2.7, potentially removes all inter-symbol and inter-carrier
interference (ISI and ICI), minimizing subsequent system performance
degradation. However, the use of a cyclic prefix requires a prefix size in tune
with channel properties and requires extensive overhead data transmission.
Both pilot symbols (used to estimate the channel in question), and the cyclic
prefix itself need to be transmitted.
The concerns for the properties of the cyclic prefix have resulted in studies on
modifications of the OFDM technique. There are a number of studies striving
to improve the frame time and frequency offset estimation needed for OFDM.
In [5] the use of number of time-domain pilot signals (using dedicated binary
synchronization patterns) is presented as a synchronization alternative to the
cyclic prefix. The synchronization patterns are alternated between OFDM
symbols. However, the gain from omitting the cyclic prefix has not been
evaluated using faded channels.
48 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
The PRP-OFDM technique first presented in [7] uses a postfix instead of a
cyclic prefix. The procedure, further evaluated in [6], still uses some overhead
information, but drastically reduces the need for pilot symbols. PRP-OFDM can
potentially increase spectral efficiency at the same as the system performance
is increased or at least remains the same as for a system using CP-OFDM.
However, the major drawbacks of using PRP-OFDM are the increased system
complexity and the introduction of both ISI and ICI. This technique has mainly
tested and evaluated a WLAN implementation from a performance
improvement perspective, i.e. increasing data throughput with improved
mobility and performance. Its complexity has to be drastically reduced in order
to achieve an implementation for use in a mobile cellular system. [6]
3.3.2
Frequency Error Sensitivity6
A technique called Polynomial cancellation coding (PCC) is proposed in [8] to
reduce the sensitivity to frequency errors in an OFDM system. Effects on
fading characteristics and the error performance are investigated, although
only for a two-path fading channel. PCC is said to improve error performance
in OFDM with lower sensitivity to large delay spreads, at the expense of lower
spectral efficiency. However, when combining PCC effects with those of
channel coding and the cyclic prefix, the spectral efficiency is said to be
comparable to that of OFDM. This technique remains to be investigated in a
multi-user multi-path fading system.
6
As detailed evaluation of OFDM impairments are outside the scope of this thesis, this
section should be considered orientation on potential further standardization.
49 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
4
Radio Network Model: Assumptions & Parameters
4.1
Link Performance Model Definition
In all packet oriented radio access technologies, the service bitrate is adapted
to the quality of the radio link through link adaptation. The radio link quality will
thus affect the transmission time for data packets of a given size. Not
considering potential delays, the system capacity will thus be restricted by
channel utilization, which will be defined by transmission time for a number of
parallel users over a given bandwidth.
The normalized measure discussed to investigate energy consumption in
section 1.4, [Ws/Mbit], would be an appropriate assumption for the cost of
energy, if power is equal to the cost of radio resources. This is generally
applicable to systems where intra-cell interference dominates (such as
WCDMA, deploying code division multiplexing). However, for OFDM systems
where orthogonality is dominant within a cell, propagation losses will reduce
the interference (in this case mostly inter-cell interference), and consequently
propagation loss will dominate the cost of power. A loss factor should be
added when modeling the link performance. As the working assumption in the
3GPP standards prerequisites for improved link performance for LTE, bit-rates
close to the Shannon link performance model can be modeled, given a loss
factor. It should be noted that a different model than Shannon could be more
appropriate to accommodate delays, if link delays are large relative to the
service delay. Such a situation would require extensive system simulations.
The model applied in this thesis assumes relatively small link delays, but
instead accommodates retransmissions. The energy required is thus
calculated with regards to the worst-case scenario in terms of resource
consumption during a transfer.
As frequency reuse considerations for LTE in many aspects are similar to
those for the existing GSM system, basic assumptions on the use of channel
resources can use those valid for a GSM system when applying tight
frequency reuse, given that the higher frequency range and specific resource
structure in LTE is accounted for. As LTE also applies an FDMA as well as an
TDMA structure for radio transmissions, the same reasoning could be used for
radio issues in an interference limited LTE system with reasonably high load
where the frequency reuse pattern could be assumed to be similar, if not
actually the same [11].
The link performance model assumed in this thesis has been determined in a
LTE system simulator with 100 % traffic load and a frequency reuse of one (all
frequencies used in all cells). Figure 14 shows the cumulative distribution
function of this model with a loss of ~20 % using 16 QAM as modulation
technique in a system with a site-to-site distance of ~1 km.
50 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
1
1*1
2*2 MIMO
CDF of Fully Loaded 1/1 System with 16 QAM [%]
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
1
2
3
4
Bits per Symbol
5
6
7
8
Figure 14: CDFs of the LTE System Model (Normal Cyclic Prefix)
4.2
Protocol Overhead
Considering transmission of an AMR speech frame over a packet switched
network and the data flow architecture shown in Figure 10, the additional
protocol overhead also has to be considered. Assumptions on included
protocol overhead have been discussed previously in section 2.12. Potential
additional information transmitted during a speech frame interval would likely
consist of the following protocol header information [63], [64]7:
•
AMR CODEC frame overhead [43]. Assumed to be 10 bits + padding
bits 0. The Coded CRC has been omitted.
•
RTP and UDP headers of 12 + 8 bytes
•
ROHC, IP-level Header compression, and PDCP protocol overhead of
minimum 24 + 16 bits. It should be noted that ROHC is assumed to be
5 bytes when SID frames are transmitted.
•
RLC and MAC protocol overhead of >2 bytes + 16 bits resulting in 40
octet aligned bits.
•
CRC on layer one, a multiple of 8 bits. Here 24 bits are assumed.
7
The Header sizes are assuming PDCP protocol termination in the Core Network. When PDCP is
moved to eNodeB (the base station) in a later stage of the (release 8) standardization [35], the header
sizes could be altered. However, related impacts on header sizes are not included in this thesis.
51 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
For a voice service that requires low delay, it is advisable to deploy the RTP
protocol [38], along with the UDP protocol instead of the TCP protocol that
requires more complex acknowledgement procedures [59].
The compression of packets can differ depending on the type of protocols
used, and packet errors that require the decompression context to be updated.
However, a constant minimum IP-level compression overhead of 3 bytes is
assumed for the speech services considered in this thesis. These 3 bytes
include IP-header, RTP, and UDP headers.
For a voice service it is assumed that discontinuous transmission is deployed
in a mobile system such that 40 % of the speech frames do not contain user
payload (the user is silent), i.e. the voice activity factor (VAF) is 100% 40 % = 60 %.
For packets of different sizes the transport block size will be different. For a
TCP/IP transfer using a total packet size of 1500 bytes, of which 1460 bytes
are payload, the additional header information would consist of an additional
20 byte TCP header, and 20 bytes for the uncompressed IP header (assuming
no header options). In some implementations with large packets it could be
advisable to avoid IP header compression in order to avoid unnecessary
system complexity since the gain of using header compression is smaller when
using larger packet sizes. A successful decoding of ROHC compressed IP
headers could require information from previously transmitted bits, which could
result in packets being lost during subsequent decoding of the compressed
header information.
Assuming no header compression for a TCP/IP transfer, the protocol overhead
would be 20 bytes for the IP header and 20 bytes for the TCP header, apart
from the PDCP, RLC/MAC, and CRC headers already mentioned. Thus, the
addition to the transport block with a compressed IP header of 3 bytes, would
be an additional 20 bytes for the TCP header, in total 37 bytes (or 57 bytes if
20 bytes of TCP header options are assumed).
4.3
Scheduling Overhead
Assuming that all control signaling for scheduling have the same modulation
rate as the PDCCH (QPSK), and turbo coding with channel rate 1/3, one can
investigate the control signaling overhead for downlink transmissions as in
Table 3.
The downlink resource assignment information and the uplink scheduling
grants transferred on the PDCCH can for simplicity reasons initially be
assumed to be around 40 bits as previously discussed8. As defined in [35], the
PDCCH is transmitted on up to the three first symbols during a TTI except on
the resources elements used for reference signaling9. The maximum number
of PDCCHs per TTI can then be calculated as
8
This size of 40 bits was mentioned in [48], although it is still under discussion in the 3GPP
standardization. In this thesis 46 and 38 bits are assumed for downlink assignments and uplink
scheduling grants respectively.
9
The reference signal is transferred as described in section 2.15.6. Every sixth frequency is used on
two symbols out of every 7 symbols.
52 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
⎛ (PDCCH symbols _ TTI − REFsymbols _ collide _ TTI )× Re sourceblocks BWeff
floor ⎜⎜
Symbolcontent PDCCH
⎝
⎞
⎟⎟ (eq. 16)
⎠
where PDCCH symbols _ TTI equals the 3 first symbols on all the (12) frequencies
in a resource block, and
REFsymbols _ collide _ TTI equals the reference symbols
transmitted in the first slot (7 OFDM symbols for a normal cyclic prefix) of a
resource block10. PDCCH symbols are not sent when there are reference
symbols in this slot; hence; these symbols are excluded. Re sourceblocksBWeff
equals the number of resource blocks over the available bandwidth,
disregarding the guard spectrum.
Symbolcontent PDCCH is different depending on the chosen modulation and
coding scheme (MCS), however, assuming 1/3 FEC coding and QPSK
modulation (two bits per symbol), 2 information bits would result in 6 coded bits
and subsequently 3 symbols. Applying FEC and modulation as per the
formula above results in that a maximum of 13 PDCCHs are available per TTI
with a 10 % guard period (15 PDCCHs over a 5 MHz spectrum with no
downlink transmission diversity. This conclusion assumes that all reference
elements are used consecutively so that PDCCH information is transmitted
without delays using the resources available. Further on it is assumed that the
PDCCH symbols are mapped primarily in the frequency domain (three
symbols intended for PDCCH are mapped onto three resource elements,
before resource elements on another carrier are used), the PDCCH
information would be divided between downlink resource assignments and
uplink scheduling grants.
Recalling the discussing of scheduling content from Table 1 and Table 2, they
result in a scheduling transmission of on average 42 bits per 20 ms, using
QPSK modulation this is equivalent to 63 symbols. As the resource element
structure with the normal cyclic prefix contains 84 000 resource elements
during 20 ms with a normal cyclic prefix length, these 63 symbols used for
PDCCH would occupy 63/84000 0.075 % PDCCH overhead per user11 (see
Table 3). As the HARQ acknowledgements use only one bit for every LTE
radio block interval (10 ms) regardless of the scheduling mechanism, the data
needs to be protected to a higher degree than for other downlink
transmissions.
10
The reference signal is transferred as described in section 2.15.6. Every sixth frequency is used on
two symbols out of every 7 symbols.
11
Given that the PDCCH overhead does not exceed the maximum of 850/84000 ≈10.1 % (13 PDCCHs
per TTI).
53 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
A more robust 1/6 FEC coding rate is assumed instead of 1/3, resulting in
~0.036 % overhead per user for HARQ acknowledgements related to UL
transmissions. If the scheduling algorithm would be changed to persistent
scheduling instead, the percentage of resources needed for scheduling
information would be reduced drastically, depending on how often users are
scheduled. With persistent scheduling the PDCCH data is scheduled less
frequently and preferably only once per average user session. However, as
discussed earlier, persistent scheduling requires that all resources needed
during a session are scheduled beforehand, including the potential need for
retransmissions. In order to cater for this potential need for extra resources, it
could happen that more resources than actually required are reserved, based
on pre-determined assumptions.
4.4
Summary of Control Information Overhead
Recalling the description of control channel constellations in LTE from section
2.15.6, the common control information transmitted to all users in a cell is
transmitted over a predefined number of resource elements every time
interval. This information content consists of the reference signal information
(where the system load is dependent on antenna configurations as shown
below), the synchronization signals, and broadcast information. How much of
the system resources that are used for necessary scheduling information and
HARQ ACK/NACK messaging depends on the number of users in the system,
and therefore this is handled separately, as in the previous section.
Table 3 displays the percentage of system resources used for control
information by one user in a 5 MHz spectrum, where 12 frequencies of
15 kHz each are dedicated to each user. For dynamic scheduling, the
PDCCH and HARQ ACK/NACK percentages will be multiplied by the
number of users (below the maximum PDCCH capacity). However, for
persistent scheduling where resources typically are reserved for a
service session, the length of the session for one user will dictate how
often scheduling of resources takes place over the PDCCH. For an
average session of 30 seconds, the PDCCH usage would amount to
only ~0.07 % percent of PDCCH usage with dynamic scheduling,
without considering potential retransmissions. Using long sessions
with little need for retransmission will thus result in a much lower
percentage of resources used for control information. However, as
previously discussed, not knowing how often retransmissions could
take place forces the scheduler to pre-allocate additional resources for
the entire session, considering the worst-case-scenario including
potential retransmissions (even though they might not be needed).
This could result in a lower degree of system utilization.
54 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Table 3: Control information in a downlink transmission over a 5
MHz spectrum using dynamic scheduling, excluding protocol
overhead
4.4.1
Signaling
Overhead
over 5 MHz
spectrum with 10% guard
band [% per user, except the
reference signal]
Reference signal
1/21
Synchronization
signal (s)
0.6912
CCPCH
2.4 /1.713
PDCCH
0.075
Hybrid ARQ
ACK/NACK
0.036
2/21
3/21
3GPP Standardization Status
The 3GPP standardization of LTE is (as previously mentioned in section 3.2)
not planned to conclude until September 2007 [31]. Thus, the assumptions and
prerequisites related to LTE as defined by 3GPP are based on the current
state of the standards. These details, used as a basis for assumptions and
basic calculations, as well as for ideas on the need for further studies, are
subject to change. Such changes could subsequently impact the results, the
discussion, and the conclusions, which should be considered accordingly.
4.5
Definition of Total Overhead
The resource element structure presented in section 2.14 and Figure 11
describes the cyclic prefix, the basic overhead needed to avoid inter-symbol
and inter-carrier interference.
On average this overhead remains the same for all resource elements within a
resource block, albeit the first symbol will actually have a smaller cyclic prefix
then subsequent symbols on the same sub-carrier in a resource block, when
deploying the normal cyclic prefix length [35].
The overhead in terms of reference signals, synchronization information and
broadcast information is defined as occupied resource elements relative to
available resource elements. This overhead is shared by all users at any given
time; hence the reference to “common OH” in upcoming presentations.
12
Assuming the use of 72 sub-carriers centered around the DC sub-carrier and 14 OFDM symbols per
subframe using 90 % of the frequencies of a 5 MHz spectrum.
13
Assuming the use of 72 sub-carriers centered around the DC sub-carrier and 10 OFDM symbols per
subframe as in [46], using 90 % of the frequencies of a 5 MHz spectrum.
55 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
The overhead for the cyclic prefix along with reference signals, synchronization
information and broadcast information constitute the basic necessary overhead
regardless of what is to be transmitted. However, the common overhead is
shared among all users, and the overhead introduced by the cyclic prefix
remains constant over all coded symbols regardless of the number of users. In
addition to this control information overhead, control overhead is required for
the transfer of payload among transmitters and receivers. One can thus
describe the total required overhead as;
1.
Common overhead; system resources required to enable OFDM
network access for each user. As this overhead is shared by all users,
fewer resources will be required per user for common overhead if the
number of users in a system increases.
2.
The cyclic prefix, being a duplicate of some of the symbol content on
each sub-carrier (see Figure 11).
3.
Overhead required in addition to payload data to enable transmission
of a data packet.
Figure 15 depicts an example of the relation between the different types of
overhead information in LTE, with two dynamically scheduled speech users in
the system. The 12.2 AMR CODEC and a normal cyclic prefix has been used
(the exact relation will be clarified in an upcoming section).
As mentioned before, the relative percentage of common overhead decreases
as the number of users increases. However, since all users require scheduling,
the system resources required for scheduling will increase as more users enter
into the system. Each users scheduling overhead will however remain constant
with regards to the payload, as a constant number of symbols are used to
schedule each user. This is valid for HARQ ACK/NACK and the protocol
overhead as well, as shown in the figure below.
Common OH
Payload data
Dynamic
Scheduling
OH
HARQ A/N
Cyclic
Prefix
OH
HARQ A/N
Payload data
Dynamic
Scheduling
OH
Protocol OH
Protocol OH
Figure 15: A Schematic Overhead Constellation in LTE for an AMR 12.2 CODEC
using the Normal Cyclic Prefix
The system resources available to each user to transfer payload and related
overhead hence can be formalized as
56 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
(1 − CyclicPrefixOH ) *
resource _ elements _ per _ user _ interval * (1 − commonOH )
number _ of _ users _ per _ user _ interval
(eq.17)
where the user_interval is dependent on type of service. For Speech users the
20 ms speech frame is used, but for other service the TTI could be
appropriate.
As the payload related overhead is constant in relation the speech CODEC,
the additional overhead can be described as
HARQ_ UL _ ACKNACKbits + scheduling _ bits + protOHbits
payloadbits
(eq. 18)
where
Pr otOHbits = AMRframeheader + PDCPheader + RTP, UDP, IPheaders
(compressed if ROHC is applied) + RLC & MAC headers + CRC
(eq.19)
If TCP packets are transferred, the TCP header should be added regardless of
header compression, while RTP and UDP headers, should be discarded;
Pr otOHbits = PDCPheader + TCPheader + IPheader
(compressed of ROHC is applied) + RLC & MAC headers + CRC
4.6
(eq. 20)
Overhead Data Impacts on Service Performance
The initial investigations will try to quantify the level of necessary overhead for
the transfer of a specific service. Different services require different levels of
overhead information. As a consequence, the power needed to transfer a
given amount of payload data will differ as the percentage of transferred
overhead increases or decreases depending upon the packet size.
The results are expected to show that the level of overhead will become more
and more insignificant as the size of transferred packets grow. This will
indicate the efficiency of a specific service at different packet sizes. As the
overhead percentage is reduced with larger packet sizes, the required power
level per payload bit will then even out and eventually remain constant. If such
a break-point can be identified, the minimum required power level per payload
has been defined. Considering the size of the packets with the minimum
energy requirement per payload, suitable services can be identified, albeit
without the delay requirements for those services having been considered.
57 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Depending on the size of packets, specific types of overhead information could
impact the need for energy more than others. By separating the overhead
content, individual overhead contributors are singled out and quantified relative
to the payload. Results will hopefully suggest potential improvements on when
and where to exclude or apply more compressed overhead information, and
when to adjust the service expectations to required levels of overhead.
The transmission of overhead information can potentially degrade the service
performance when small data packets are transferred. To investigate this, the
transmitted downlink power needs to be quantified, where the major
contributors of overhead data are pin-pointed and discussed specifically. The
coding strategy used for different types of data packets also needs to be
clarified.
4.7
Energy Transmission Considering DTX
In order to transmit the SID frames over the speech channel during DTX
periods, in total 51 bits14 are transmitted every 160 ms, requiring 42 resource
elements, or 42/672000 (0.00625 %) of the available resource elements per
user with 16 QAM and a code rate of 1/3, during DTX periods.
The voice activity factor stipulates how many speech frames are transmitting
actual speech). Remaining speech frames are transmitting SID frames and
related overhead. The total number of speech payload symbols needed per
user can thus be described as a weighted average of DTX and speech
periods:
⎛ Symbols SID
⎜
⎛ Symbols AMR ⎞
Elements AMRframe
⎜
⎟
VAF
+ (1 − VAF )⎜
8
⎜ Elements
⎟
⎜
AMRframe ⎠
⎝
⎜
⎝
⎞
⎟
⎟
⎟
⎟
⎠
(eq. 21)
where Symbols AMR and SymbolsSID represent symbols used for speech
payload (A/B/(C) bits), and symbols required for SIDframes, including the AMR
CODEC speech frame overhead (see section 4.2).
4.8
Environment Details
Resource elements not used to transfer control or payload information are
assumed empty in accordance with [35], enabling a reduction in power
whenever resource elements do not need to transfer information.
A resource block is defined as resource elements during 1 TTI spanning over
12 frequencies (as in [35]). One user is mapped to one resource block unless
otherwise stated.
Normal cyclic prefix length is assumed unless otherwise stated, resulting in a
structure per resource element of 7 consecutive symbols per slot and
frequency, and 14 consecutive symbols per a 1 ms TTI.
14
39 bits payload bits + 10 bits + 7 padding bits assumed as AMR frame overhead during DTX[43], 0.
58 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
MIMO configurations mentioned in this thesis are all defined as 2*2, using two
parallel streams on the downlink intended for the same block of resource
elements, i.e. for the same user.
Results will be presented assuming a 5 MHz spectrum with 10 % guard band,
with 20 W (43 dBm) full power unless otherwise stated. It is assumed that
larger spectra (such as 10, 15, or 20 MHz) will deploy multiples of 20 W in
proportion to the bandwidth. The size of the guard band is initially assumed to
adapt accordingly.
The services chosen to be evaluated are packet-switched speech using AMR
CODECs (those discussed for LTE deployment in the 3GPP standards for
narrow-band and wide-band AMR), and a TCP-transfer of different packet
sizes, this latter transfer is assumed to not be sensitive to delay. The packet
sizes have been chosen to reflect VoIP services and likely TCP transfers
ranging from relatively small (byte-aligned) packets, through Ethernet-sized
packets to packet sizes close to the maximum receiver-window size of 64 kB
including overhead.
Dynamic scheduling (every 1ms TTI) is assumed for TCP packet transfers.
Byte-alignment (added padding bits to even byte packets) is applied to payload
overhead, but not to the payload itself.
The maximum allowed number of scheduling bits (L1/L2 signaling) is assumed
at all times (see section 2.6).
Potential impacts on downlink capacity from presence information sent on the
uplink have not been considered in this thesis, although such discussions are
ongoing for the 3GPP standardization.
The environment details are summarized in Table 4 below.
Table 4: Environment Details
Frame Structure
Generic [35]
Division Duplex Method
FDD
Cyclic Prefix Length
Normal (14 symbols/1ms TTI)
Channel Separation
15 kHz
Subcarriers per Resource Block
12
Frequency Spectrum
5 MHz
Guard Period [of spectrum]
10 %
Payload Modulation
16 QAM
Signaling Modulation
QPSK
Forward Error Correction
1/3, 1/6 (HARQ ACK/NACK)
59 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Power
20 W per 5 MHz
DTX DL Factor
40 %
SID Frame Content
39 bits/160 ms
MIMO Configuration
2*2
Link Performance Model
Fully Loaded System,
1/1 reuse, site-to-site distance
1 km
Evaluated Services
VoIP, TCP
Scheduling Methods
Dynamic & Persistent
Scheduling15
TCP Header size
20 Bytes (no options)
RTP, UDP and IP Header Sizes
12+8+20 = 40 Bytes
ROHC Size
3 bytes for speech periods,
5 bytes for DTX periods,
including RTP, UDP & IP
Headers
PDCP Header Size (excl.
ROHC)
2 Bytes
RLC/MAC Header Size
5 Bytes
CRC
3 Bytes
Narrowband AMR Speech
CODECs [kbps]
4.75, 5.9, 7.4, 12,2
Wideband AMR Speech
CODECs [kbps]
6.60, 8.85, 12.65
TCP Payload Packet Sizes
[Bytes]
25, 75, 300, 1000, 3000, 4000,
64000
Transmisson Link
Downlink
15
Dynamic Sceduling every TTI of 1 ms, and Persistent Scheduling once every assumed session of
30 s. Dynamic Scheduling has been assumed for TCP services services
60 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
5
Results
5.1
Common Control Information Overhead
All users of LTE are subject to a specific percentage of common control
information; pilot symbols or reference signals, synchronization signals, and
broadcast information. This control information is common for all users,
regardless of the number of users in the system.
The discussions in section 2.7, explain why a number of duplicated symbols the cyclic prefix – should be added to the signal to inhibit ICI and ISI, and
maintain orthogonality by accommodating changes in multipath delay. When
the cyclic prefix is extended, the control information overhead will increase.
However, the number of symbols transferred per resource block is reduced (as
described in section 2.14). As a consequence, larger overhead is added onto
each symbol (see Figure 16), and more system resources are required for
transmitting cyclic prefix information that is filtered out by the receiver. This will
likely reduce the capacity for simultaneous users accordingly; hence, the
energy requirement per bit of payload is of interest.
.
20
Symbol Percentage used by the Cyclic Prefix [%]
18
16
14
12
10
8
6
4
2
0
Normal Prefix length
Extended Prefix Length
Figure 16: Cyclic Prefix Overhead per Symbol
Figure 17 describes the percentage overhead of system resources in terms of
signals that are common to all users (reference signals, synchronization
signals, and broadcast channel information). The bars display the overhead
percentage using the normal and extended cyclic prefix, for a single-antenna
configuration and for a 2*2 MIMO configuration.
61 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
.
System Resources needed for Common OH [%]
15
ref. symbols
synch. symbols
CCPCH (Broadcast Channel)
10
5
0
Single Antenna, n. CP
MIMO 2*2, n. CP
Single Antenna, ext. CP
MIMO 2*2, ext. CP
Figure 17: Common Control Information Overhead for Normal and Extended
Cyclic Prefix Length
Results show that the overhead shared by all users amount to ~7.8 % for a
single-antenna configuration with a normal cyclic prefix length, and ~12.6 % in
a 2*2 MIMO configuration. With the extended cyclic prefix, the resources
required for common control information increases to ~8.8 % and 14.3 %
respectively.
This increase of system resources used by common overhead is due to an
increase in transmitted reference symbols for MIMO configurations [35].
Excluding the system overhead, the remaining resources are examined
in Table 5 below, for different configurations.
As discussed in section 2.15.6, the percentage of reference symbols are
dependent upon the antenna configuration, but not upon the system
bandwidth. The synchronization signal overhead and broadcast information are
however dependent upon the bandwidth. Given this, these will affect the
common overhead whenever the system bandwidth is altered.
Table 5 examines common overhead percentage and remaining system
resources for a 5 MHz spectrum, as well as remaining resources system for
several other spectra.
62 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Table 5: System Overhead [%]
Cyclic Prefix Length
Normal
CP
Extended
CP
Antenna Configuration
1*1
1*1
2*2
2*2
[Downlink*Uplink antennas]
Cyclic Prefix
6.61 6.61 20
20
Common OH
7.85 12.6 8.76
14.3
Remaining Resources @ 5 MHz
86.0 81.6 73.0
68.6
77.4 72.9 65.3
60.9
87.4 83.0 74.3
69.8
88.2 83.7 74.9
70.5
[1-OH = (1-CP)*(1-CommonOH)]
Remaining Resources @ 1.25 MHz
[1-OH = (1-CP)*(1-CommonOH)]
Remaining Resources @ 10 MHz
[1-OH = (1-CP)*(1-CommonOH)]
Remaining Resources @ 20 MHz
[1-OH = (1-CP)*(1-CommonOH)]
As discussed in previous sections, the extended cyclic prefix would
predominantly be deployed for large cells with large delay spreads or for multicell broadcasting [4]. A question that arises based on these findings is whether
a changed cyclic prefix length will affect the required energy given that all the
overhead is accounted for.
Figure 16 and Table 5, along with related figures in the appendix corroborate
the assumption that an increase in cyclic prefix length and common overhead
reduce the user capacity and increase the required energy per transmitted
payload accordingly, given that the changes to the common overhead in Figure
17 are considered.
5.2
SID-frames, Scheduling, and HARQ ACK/NACK
In periods of DTX very little information is transmitted, since the user on one
end is silent. However, SID frames need to be sent to inform the receiver of
changes in the voice activity (see section 2.6). As the 39 payload bits (and its
related AMR frame overhead) are used for SID-frames sent at 160 ms intervals
[49], the speech frame size during DTX periods related to its size during
speech periods can be described as in Figure 18.
63 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
6
SID Frame payload
Speech Frame Payload during DTX [%]
5
4
3
2
1
0
N4.75
N5.9
N7.4
N12.2
W6.60
AMR Speech CODECs
W8.85
W12.65
Figure 18: AMR Payload transmitted during DTX periods
Results show that resources used by SID frames in speech packets can range
from ~5.1 to ~2.0 % during DTX periods depending upon the chosen speech
CODEC, where ~2.0 % reflects the least robust CODECs, appropriate for good
radio conditions.
The resources occupied to transmit HARQ acknowledgements as responses
to uplink transmissions are shown in Figure 19. For larger packets the
overhead becomes negligible (~0.1 % and ~ 0.2 % for a 25 byte TCP byte
packet in single antenna and MIMO 2*2 configurations respectively), and is
therefore not considered.
64 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
4.5
HARQ ACK/NACK on UL transmissions: Single Antenna
HARQ ACK/NACK on UL transmissions: MIMO 2*2
Speech Frame Percentage used by HARQ [%]
4
3.5
3
2.5
2
1.5
1
0.5
0
N4.75
N5.9
N7.4
N12.2
W6.60
AMR Speech CODECs
W8.85
W12.65
Figure 19: HARQ ACK/NACK Overhead
Figure 20 and Figure 21 display the occupancy of scheduling resources
transmitted on the downlink (downlink resource assignment and uplink
scheduling grants) in relation to the size of the payload, when dynamic
scheduling is deployed every TTI for different packet sizes. The persistent
scheduling overhead – in this thesis sent once every 30 seconds – is negligible
in comparison to dynamic scheduling overhead (or “messages”) sent 1500
times more often (every TTI). Hence; persistent scheduling is not shown in
either figure.
65 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
90
DL Assignment DYN Sched OH
UL Grants in DL, DYN Sched OH
Scheduling Overhead per Speech Frame [%]
80
70
60
50
40
30
20
10
0
N4.75
N5.9
N7.4
N12.2
W6.60
AMR Speech CODECs
W8.85
W12.65
Figure 20: Scheduling Overhead for Speech Frames
25
DL Assignment DYN Sched OH
UL Grants in DL, DYN Sched OH
Scheduling OH per TCP packet [%]
20
15
10
5
0
25
75
300
1000
1500
TCP Payload packet size [Bytes]
4000
64000
Figure 21: Scheduling Overhead for TCP packets
The scheduling graphs assume an even distribution of downlink resource
scheduling and uplink scheduling grants for speech services, when all
available scheduling resources are allocated.
66 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
As Figure 21 reflects, the uplink/downlink transfer ratio is changed for TCP
transfers in comparison to Figure 20. 80 % of the scheduling traffic is now
assumed for downlink resource assignments and 20 % for the uplink
scheduling grants. Even though either downlink or uplink scheduling could be
predominant depending on the service, the total scheduling overhead would
however remain the same for a given packet size, unless the scheduling
strategy itself is changed as all available scheduling resources are assumed
allocated.
With regards to scheduling overhead discussed above it should be noted that
dynamic scheduling can limit the total number of simultaneous users in an LTE
system, since the PDCCH channel only can cater for a limited number of users
per TTI (as discussed in section 4.3). However, if scheduling information can
be sent at longer intervals, the PDCCH capacity will allow more users, given
that the services provided can tolerate the delays potentially introduced when
resources are scheduled less frequently.
5.3
Protocol Overhead
Based on the assumptions for protocol overhead discussed in sections 2.12
and 4.1, it would be interesting to see whether any of these protocol headers
occupy a larger part of a transmitted packet than others when packet sizes
change. The percentage of protocol header overhead will decrease as the
packet size increases, but it remains to be determined how this will affect the
energy that needs to be transmitted to convey a given amount of payload data.
Figure 22 displays the level of protocol headers in an AMR packet, relative to
the size of the payload, while Figure 23 shows protocol overhead relative to
the size of the payload of a TCP packet. The different header types are
summarized in the two lowest bars for all packet sizes, with and without the
use of ROHC compression.
67 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Protocol OH in total without ROHC
Protocol OH in total with ROHC
L1 CRC
RLC & MAC Headers
ROHC [RTP/UDP/IP]
PDCP Header
RTP/UDP/IP Headers [no ROHC]
AMR Frame Header
W12.65
AMR Speech CODECs
W8.85
W6.60
N12.2
N7.4
N5.9
N4.75
0
50
100
150
200
250
300
350
Speech Frame Payload Overhead Required by Protocol Headers [%]
400
Figure 22: Protocol overhead for AMR packets
From Figure 22 it is apparent that the payload overhead is quite large for a
small packet service such as speech, ranging from ~139 % of the payload size
down to ~51.3 % with dynamic scheduling and header compression (ROHC),
depending on the speech CODEC. Without header compression, the overhead
will range from ~445 % down to ~168 % of the payload for speech packets.
68 (98)
450
Overhead Impacts on Long-Term Evolution Radio Networks
Protocol OH in total without ROHC
Protocol OH in total with ROHC
L1 CRC
RLC & MAC Headers
ROHC [IP]
PDCP Header
IP Header [no ROHC]
TCP Header
64000
TCP Payload packet size [Bytes]
4000
1500
1000
300
75
25
0
20
40
60
80
100
120
140
160
Packet Payload Overhead Required by Protocol Headers [%]
180
Figure 23: Protocol overhead for TCP packets
The graphs suggest that header compression can drastically reduce the
relative protocol overhead, at least for a packet size equal to or smaller than
~1000 bytes.
An interesting observation in Figure 22 and Figure 23 (recalling previously
shown overhead figures) is that protocol header overhead seems to constitute
a large part of the overhead, regardless of the packet size. This is discussed
further in section 0.
5.4
Total Overhead
When summarizing the previous findings the overhead percentage can be
described as shown in Figure 14 where the total overhead consists of system
resources shared by all users, and payload overhead that is additional to the
payload of each user.
Figure 24 shows the required system overhead for up to 260 users –
increasing with the number of users when dynamic scheduling and ROHC is
applied. The maximum of 260 users have been specifically chosen to reflect
the presumed scheduling capacity limits of dynamic scheduling of speech
users in a single antenna configuration (given a certain message size and
MSC), as previously discussed.
69 (98)
200
Common OH, (Dyn.) Scheduling, HARQ ACK/NACK & Protocol Headers [%]
Overhead Impacts on Long-Term Evolution Radio Networks
35
Using Persistent Scheduling once every 30 s
30
25
20
15
10
5
0
1-260 users, Single Antenna
Scheduled Users
1-260 users, MIMO 2*2
Figure 24: System Overhead
Figure 25, Figure 26, and Figure 27 depict the total summarized required
payload overhead, depending upon the scheduling algorithm and packet size.
1*1 ROHC
2*2 ROHC
1*1 no ROHC
2*2 no ROHC
W12.65
AMR Speech CODECs
W8.85
W6.60
N12.2
N7.4
N5.9
N4.75
0
100
200
300
400
500
Total AMR Payload Overhead [HARQ ACK/NACK, DYN Sched, Protocol OH] [%]
Figure 25: Summarized Payload Overhead per AMR Speech Frame (Dynamic
Scheduling)
70 (98)
600
Overhead Impacts on Long-Term Evolution Radio Networks
1*1 ROHC
2*2 ROHC
1*1 no ROHC
2*2 no ROHC
W12.65
AMR Speech CODECs
W8.85
W6.60
N12.2
N7.4
N5.9
N4.75
0
50
100
150
200
250
300
350
400
450
Total AMR Payload Overhead [HARQ ACK/NACK, PER Sched, Protocol OH] [%]
500
Figure 26: Total Payload Overhead per AMR Speech Frame (Persistent
Scheduling)
1*1 ROHC
2*2 ROHC
1*1 no ROHC
2*2 no ROHC
TCP Payload Packet Size [Bytes]
64000
4000
1500
1000
300
75
25
0
50
100
150
200
Total TCP Overhead (HARQ ACK/NACK, Sched, Protocol OH [%]
250
Figure 27: Total Payload Overhead per TCP Paket
From Figure 25, Figure 26, and Figure 27 the apparently predominant factor to
address in order to reduce packet overhead is the use of ROHC.
Despite ROHC, the protocol overhead remains a dominant part of the total
overhead.
71 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
It should also be noted that the percentage of total overhead continuously
decrease, and the impact on required energy will diminish as the packet size
increases beyond 1kB.
As previously discussed, the use of different scheduling algorithms can lower
the required payload overhead. However, in order to quantify the impact of
changing algorithms, the predominant overhead contributors have to be
examined. Figure 28 shows how much of the total overhead that is used for
scheduling and protocol headers for two common AMR speech CODECs. It
now remains to investigate how much the total overhead is reduced if the
scheduling overhead from Figure 28 is reduced drastically.
Single Antenna, With ROHC
150
Dyn Scheduling OH
Protocol Headers [AMR Frame Header, PDCP, ROHC [RTP, UDP, IP], RLC, MAC, CRC]
Payload OH [%]
100
50
0
NAMR 7.4
AMR Speech CODECs
NAMR 12.2
Figure 28: Predominant Payload Overhead for AMR CODECs
Figure 29 describes the reduction in total overhead that could be obtained if
using persistent scheduling every 30 seconds instead of dynamic scheduling
every 1ms TTI. A reduction of ~35 % and ~16 % in overhead is achieved with
persistent instead of dynamic scheduling with or without ROHC, for two typical
AMR speech CODECs.
72 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
ROHC
no ROHC
AMR Speech CODECs
NAMR 12.2
NAMR 7.4
0
5
10
15
20
25
30
Lowered AMR Payload Overhead for PER vs. DYN sched [%]
35
40
Figure 29: Reduced Total Overhead at Worst-Case Scheduling: Persistent
Scheduling vs. Dynamic Scheduling
5.5
Required Energy for Service Provisioning
Along with the overhead observations presented in the previous section, the
necessary energy level per transmitted payload bit can be examined, given a
specified power level shared by a number of users at different payload packet
sizes (in this case 43 dBm or ~20 W is used over 5 MHz spectrum excluding
guard bands). Consequently the relative energy requirement per bit of payload
could be determined for different packet sizes.
73 (98)
Control Channel Capacity in Downlink (Ideal LA) [Users/20ms]
Overhead Impacts on Long-Term Evolution Radio Networks
1500
PER sched, Single Antenna
PER sched, MIMO 2*2
Dynamic Scheduling, Single Antenna
Dynamic Scheduling, MIMO 2*2
1000
500
0
N4.75
N5.9
N7.4
N12.2
W6.60
AMR Speech CODECs
W8.85
W12.65
Figure 30: Estimated Control Channel Capacity
The figure above does not include delays, or detailed scheduling. Perfect
(ideal) link adaptation is assumed, which results in a very rough estimate on
the control channel capacity. The results are based on the number of bits that
will be transmitted, which is directly related to previous overhead discussions.
However, the inherent loss in the link performance model (described in section
4.1) enables numerous retransmissions.
The two rightmost-bars indicate the presumed capacity limit of the scheduling
channel for this configuration (given a specific message size and MCS) when
using dynamic scheduling (see section 4.3). Thus the figure reflects the control
channel capacity considering the number of scheduling messages, which is
closely connected to how often these messages are sent for one user. Ideal
link adaptation (LA) is assumed for both the scheduling algorithms (see
comments in section 0).
Figure 31 displays the energy needed for every speech user at 20 ms
intervals, while Figure 32 depicts the required energy level for one TCP packet
of increasing size.
74 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Energy per user
-3
2.5
x 10
PER Sched, Single Antenna
PER Sched, MIMO 2*2
Dynamic Shed, Single Antenna
Dynamic Shed, MIMO 2*2
Ws/speech frame
2
1.5
1
0.5
0
N4.75
N5.9
N7.4
N12.2
W6.60
AMR Speech CODECs
W8.85
W12.65
Figure 31: Energy per Speech User
3.4
Dynamic Shed, Single Antenna
Dynamic Shed, MIMO 2*2
1.6
Energy per TCP Packet [Ws]
1.4
1.2
1
0.8
0.6
0.4
0.2
0
25
75
300
1000
1500
TCP Payload packet size [Bytes]
Figure 32: Energy per TCP Packet
75 (98)
4000
64000
Overhead Impacts on Long-Term Evolution Radio Networks
If the energy needed to transfer a TCP packet is distributed over the payload
for that packet, a relative measure of the required level of energy to provide a
specific packet service is obtained. Figure 33 presents this, while Figure 34
reflects the inverse by depicting the amount of payload bits transferred per
energy level.
16
Single Antenna
MIMO 2*2
14
Energy per payload [Ws/Mbit]
12
10
8
6
4
2
0
25
75
300
1000
1500
TCP Payload packet size [Bytes]
Figure 33: Energy per TCP Payload
76 (98)
4000
64000
Overhead Impacts on Long-Term Evolution Radio Networks
0.35
Single Antenna
MIMO 2*2
Payload per Energy Quanta [Mbit/Ws]
0.3
0.25
0.2
0.15
0.1
0.05
0
25
75
300
1000
1500
TCP Payload packet size [Bytes]
4000
64000
Figure 34: TCP Payload per Ws
5.6
Boosted Reference Signal Power
As discussed in section 2.15.6, increasing the power on reference signals will
increase required system overhead. Figure 35 shows the absolute levels of
required system control overhead when the reference signal power is
increased in steps of one dB.
77 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
100
Single Antenna
MIMO 2*2
90
80
CommonOH [%]
70
60
50
40
30
20
10
0
0
1
2
3
4
5
6
Increased Ref. Signal Power [dB]
7
8
9
10
Figure 35: System Level Control Overhead Impacts from Increased Reference
Signal Power
Results show that an increase of reference signal power of 3 dB gives the
same system overhead as the MIMO configuration presented in this thesis;
hence the impacts on increased energy levels from such an increase in power
are the same.
78 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
6
Discussion
For smaller packets, such as packet-oriented speech packets, the protocol
overhead will – depending on the CODEC – require roughly 1.4 up to ~4.5
times the resources of the payload (see Figure 25 and Figure 26). Evidently
(as expected), the smaller the packet, the more useful IP header compression
(ROHC) seems, in terms of added overhead relative to the size of the payload.
The high percentage of overhead for smaller packets (relative to the payload
size) suggests that a number of users should be multiplexed in order to make
the most effective use of the available spectrum.
For packets with payload sizes beyond ~1kB, the required increase in receiver
processor complexity and likely increase in retransmissions due to an increase
in faulty decoding, might not justify the use of ROHC, unless the system
strategy is to simplify the overall design by deploying only one mapping
algorithm, regardless of the service or packet size.
Despite ROHC header compression, protocol overhead seems to be a major
contributor to payload overhead. However in most cases the transferred
packets will be a maximum transmission unit (MTU) in size. Hence the protocol
overhead will be very small. For file transfers, small packets – with a larger
percentage overhead – will likely occur about once per file, while for AMR or
other packets with a minimal size header, the overhead is more extensive.
While non-multiplexed AMR packets would need a change in scheduling when
there is a silence period to active period transition (and the same for the
reverse), multiplexed AMR packets would only need a change in scheduling
when there is a substantial change in resource requirements. Thus an efficient
multiplexing algorithm would not require changed scheduling very often, as the
scheduling would be based on the statistics of the multiplexed AMR stream
rather than an individual AMR stream.
Comparing two extreme cases of scheduling – dynamic scheduling every TTI,
and persistent scheduling once every average session of 30 seconds –
indicate a reduction in total overhead of either ~16 % or ~35 %, where the
latter applies when ROHC is used. What it means capacity-wise to schedule
resources more seldom depends on details of, e.g., the chosen algorithm and
environment. As previously mentioned, this is outside the scope of this thesis.
If considering the dynamic and persistent scheduling investigated in this thesis
as two extreme cases of scheduling algorithms, while assuming that the gain
in capacity is proportional to the gain in control channel capacity, one can
estimate the capacity gain to ~1.7-2.7 times that of dynamic scheduling in a
single antenna configuration, and ~3.6-5.9 times using MIMO 2*2, depending
upon the chosen CODEC (see Figure 30). Such a comparison is based on
ideal link adaptation and that the system capacity with dynamic scheduling is
limited by the capacity of the scheduling channel. However, even the most
optimistic persistent scheduling would have to cater for users on the cell
edges, resulting in more overhead for users on a better channel. Thus, the
results with persistent scheduling are far too optimistic, depending upon the
link adaptation and the modulation and coding scheme.
79 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
The power required for a transfer decreases as the overhead is reduced.
However, the power per transferred payload content evens out as the packet
sizes grow beyond ~1kB. This suggests that services that require for instance
Ethernet transfers (~1.5 kB packets) up to the maximum receiver window size
(~64 kB), will require about the same power per payload bit regardless of the
overhead.
Given that packet sizes above ~1kB requires a constant energy per
transmitted bit of payload, and the multiplexing of small packets (mentioned
earlier in this section), one could speculate on the optimum multiplexing of a
number of small packet (AMR speech) users. In order to reach a payload of
~300 byte – where the required energy per bit of payload start to even out –
one would need to multiplex about 10 users given the AMR 12.2 codec and the
DTX factor assumed in this thesis. However, this assumes that the protocol
overhead currently required per user can be shared among several users. If
multiplexing several small packets services one would additionally have to
accommodate different protocol headers.
Due to increased control information overhead when the extended cyclic prefix
is used, more power is needed per transferred bit of payload. The increase in
required power per payload bit corresponds to the added control information
overhead and the additional system resources required by the extended cyclic
prefix.
80 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
7
Conclusions
•
The use of TCP/IP header compression (ROHC) is the single most
important factor to reduce payload overhead, for packet sizes of ~1kB
or smaller. However, the gain of using ROHC for larger packet sizes is
negligible (see Figure 27).
•
Protocol headers – including the AMR frame header, RLC/MAC
headers, and CRC where applicable – remain the largest part of
payload overhead regardless of packet size and header compression
(ROHC).
o
For VoIP over the UDP protocol (with ROHC), RLC/MAC
headers constitute the largest part of protocol headers (see
Figure 22).
o
For TCP/IP applications (without ROHC), TCP/IP headers are
predominant for small payloads (see Figure 23).
•
Services that require packet sizes beyond ~1 kB will require about the
same power per payload bit regardless of the percentage of payload
overhead (see Figure 33).
•
Comparing two extreme cases of scheduling – dynamic scheduling
every 1 ms TTI, and persistent scheduling once per session of 30
seconds – the payload overhead can be reduced by ~35 % whenever
ROHC is applicable. Without ROHC, the payload overhead reduction
by using persistent scheduling is ~16 % (see Figure 29).
•
Considering the dynamic and persistent scheduling investigated in this
thesis as two extreme cases of scheduling algorithms, one can
conclude that the overhead is reduced drastically with persistent
scheduling (see Figure 30). However, the capacity gain will not be
equivalent, due to link adaptation and adjustments for users on the cell
edges.
81 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
8
Further Studies
Some possible extensions are:
•
Additional studies and system simulations to verify the indications
found in this thesis, especially the usefulness of MIMO for larger
packets, the overhead impacts from higher layer protocols, and the
scheduling efficiency impact on transmitted energy.
•
Additional antenna solutions; 4x4 MIMO etc.
•
Comparison of the emission and signaling investigation results and
similar scenarios using GSM EDGE and or WCDMA/HSDPA, as
suitable simulation tools are being developed.
•
For what type of services are the two CP lengths chosen by GPP for
different delay spreads optimal?
82 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
References
[1] Internet resource: The GSM Association website, GSM/3G Fast
Facts, Accessed on Sept. 23 2006, from URL:
http://www.gsacom.com/news/statistics.php4?PHPSESSID=f7a18672
22cb6b0d07260d0a89d079d5
[2] S. He, Equalization Techniques for Co-channel Interference
Suppression, Chalmers University of Technology, Department
of Signals and Systems, Master Thesis, EX003/2005, 2005
[3] A. Simonsson, M. Almgren, M. Thurfjell, A Power Control and
Scheduling Concept for EGPRS, proc. 52nd Vehicular
Technology Conference, IEEE, Sept. 2000, pp 2219-2224
[4] J. Heyman, Intercell Interference Management in an OFDMbased Downlink, Linköping University, Department of Electrical
Engineering, Master Thesis, EAB-06:028543 Uen, June 2006
[5] J.D. Bakker, Eliminating the OFDM Cyclic Prefix, 13th IEEE
International Symposium on Personal Indoor and Mobile
Communications, September 2002, pp 2219-2224.
[6] M. Sundberg, Psedo Random Postfix OFDM,
University, Master Thesis, Oct. 2005
Uppsala
[7] M. Muck et al., A Pseudo Random Postfix OFDM modulator
and inherent channel estimation techniques, Globecom 2003.
[8] R. Kusha et al., The Performance of PCC-OFDM in Multipath
Fading Channels, Submitted to Australian Communication
Theory Workshop, 2004
[9] International telecommunication union (ITU) Press Release
ITU/99-22, Nov. 5, 1999
[10] International telecommunication union (ITU) Press Release
ITU/99-24, Dec. 10, 1999
[11] Internet resource: International telecommunication union (ITU)
website, What is IMT-2000?, Accessed on Sept. 23 2006, from
URL: http://www.itu.int/osg/imt-project/docs/What_is_IMT2000-2.pdf
[12] A. Simonsson, Frequency Reuse and Intercell Interference
Co-ordination in E-UTRA. proc. 65th Vehicular Technology
Conference, IEEE, 22-25 April 2007, pp 3091-3095
[13] K. Haider, Peak to average ratio reduction in Wireless OFDM
communication systems, Master Thesis, Blekinge Institute of
Technology, Department of Telecommunications and signal
processing, Dec. 2005
83 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
[14] Internet Resource: 3GPP TSG-RAN WG1,
R1-071956,
Ericsson: E-UTRA Performance Checkpoint: Downlink Initial
Super 3G downlink performance assessment, Ericsson, Tech.
Report EAB-05:033452 Uen, Nov 2005, Revision Accessed on
May 12 2007 from URL:
http://www.3gpp.org/ftp/tsg_ran/WG1_RL1/TSGR1_AH/LTE_P
erformance_evaluation_April-07/Docs/R1-071956.zip
[15] Internet Resource: The RF Design website, L. Litwin, M.
Pugel, The Principles of OFDM, Jan. 2001, Accessed on Sept.
30 2006 from URL:
http://rfdesign.com/mag/radio_principles_ofdm/index.html
[16] 3GPP TSG-RAN WG1, R1-051348, Samsung: Downlink
Resource Allocation and Multiplexing, Accessed on May 7
2007 from URL:
http://www.3gpp.org/ftp/tsg_ran/WG1_RL1/TSGR1_43/Docs/R
1-051348.zip
[17] R. G. Gallager, Information Theory
Communication, New York, Wiley, 1968
and
Reliable
[18] E. Dahlman, et al., The 3G Long-Term Evolution – Radio
Interface Concepts and Performance Evaluation, proc. 63rd
Vehicular Technology Conference, IEEE, 2006, pp 137-141
[19] H. Ekström et al., Technical Solutions for the 3G Long-term
Evolution, IEEE Communications Magazine, March 2006, pp
38-45
[20] Internet resource: 3GPP, Requirements for Evolved UTRA (EUTRA) and Evolved UTRAN (E-UTRAN), Tech. Report
25.913, version 7.3.0, Accessed on October 13 2006 from
URL: http://www.3gpp.org/ftp/Specs/html-info/25913.htm.
[21] M. A. Aceña, Power and Bit Allocation for Multicarrier
Modulation in Multi-user Environments, Master Thesis,
Chalmers University of Technology, 2005
[22] Ericsson AB, WCDMA Evolved, Ericsson White Paper, 284
23-2958 Uen Rev A, 2004.
[23] L. Ahlin, C. Frank, and J. Zander, Mobil Radiokommunikation,
Studentlitteratur, 1995.
[24] B.
Sklar, Rayleigh fading channels in mobile digital
communications, IEEE Communications Magazine, July 1997,
pp. 90 –100
[25] S.B. Weinstein and P. M. Ebert, Data transmission by
frequency-division multiplexing using the discrete fourier
transform,
IEEE
Transactions
on
Communications
Technology. Volume: COM-19, No. 5 , October 1971, pp. 628
– 634
84 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
[26] T. Keller and L. Hanzo, “Adaptive multicarrier modulation: a
convenient framework for timefrequency processing in
wireless communications”, IEEE Proceedings of the IEEE,
2000, Volume: 88 No. 5, May 2000, pp. 611 – 640
[27] Internet resource: 3GPP, Physical Layer Aspects for Evolved
Universal Terrestrial Radio Access (E-UTRA), Tech. Report
25.814, version 7.1.0, Accessed on November 1 2006 from
URL: http://www.3gpp.org/ftp/Specs/html-info/25814.htm.
[28] Internet resource: 3GPP, Long Term Evolution of the 3GPP
radio technology, Accessed on November 1 2006 from URL:
http://www.3gpp.org/Highlights/LTE/LTE.htm.
[29] Internet resource: NASA, Goddard Space Flight Centre,
Dopppler Shift, Accessed on November 3, 2006 from URL:
http://imagine.gsfc.nasa.gov/YBA/M31-velocity/Doppler-shift2.html
[30] Internet resource: 3GPP, LTE Physical Layer - General
Description, 36.201, version 0.1.1, Accessed on November 8
2006 from URL:
http://www.3gpp.org/ftp/Specs/html-info/36201.htm
[31] Internet resource: 3GPP, Evolved Universal Terrestrial Radio
Access (E-UTRA) and Evolved Universal Terrestrial Radio
Access (E-UTRAN); Overall description; Stage 2, 36.300,
version 0.3.1, Accessed on January 2 2007 from URL:
http://www.3gpp.org/ftp/Specs/html-info/36300.htm
[32] Internet resource: 3GPP, Evolved UTRA Aspects,
Specification series 36, Accessed on November 8 2006 from
URL: http://www.3gpp.org/ftp/Specs/html-info/36-series.htm
[33] H. Nyquist, Certain Topics in Telegraph Transmission Theory,
Transactions of the AIEE (American Institute of Electrical
Engineers) vol. 47, Feb. 1928, pp 617-644
[34] C. Shannon, Communication in the presence of noise, Proc. of
the IRE, vol. 37, no.1, Jan. 1949, pp. 10–21
[35] Internet resource: 3GPP, Evolved Universal Terrestrial Radio
Access (E-UTRA); Physical Channels and Modulation, 36.211,
version 0.2.1, Accessed on December 18 2006 from URL:
http://www.3gpp.org/ftp/Specs/html-info/36211.htm
[36] M. Ericsson, et. al., Effects of simultaneous circuit and packet
voice traffic on total capacity, proc. 63rd Vehicular Technology
Conference, IEEE, Spring 2006, pp. 538-542, Accessed on
May 3 2007 from URL:
http://www.ericsson.com/technology/research_papers/wireless
_access/doc/Effects%20of%20simultaneous%20circuit%20an
d%20packet%20switched.pdf
85 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
[37] S. Wänstedt, et. al., The Effect of F-DPCH on VoIP over
HSDPA Capacity, proc. 63rd Vehicular Technology
Conference, Spring 2006, pp. 410-414, Accessed on May 3
2007 from URL:
http://www.ericsson.com/technology/research_papers/wireless
_access/doc/The%20effect%20of%20FDPCH%20on%20VoIP%20over%20HSDPA.pdf
[38] B. A. Forouzan, TCP/IP Protocol Suite, 3rd Edition, ISBN 0-07111583-8, McGraw-Hill, 2006
[39] Internet Resource: 3GPP TSG-RAN WG1, R1-061382,
Ericsson: E-UTRA Downlink User Throughput and Spectrum
Efficiency Evaluation, Accessed on January 9 2007 from
URL:http://www.3gpp.org/ftp/tsg_ran/WG1_RL1/TSGR1_45/D
ocs/R1-061382.zip
[40] Internet resource: 3GPP, Technical Specification Group Radio
Access Network, Physical layer aspects of UTRA High Speed
Downlink Packet Access, 25.848, version 4.0.0, Accessed on
December
20
2006
from
URL:
http://www.3gpp.org/ftp/Specs/html-info/25848.htm
[41] Internet resource: 3GPP, Technical Specification Group Radio
Access Network, Evolved Universal Terrestrial Radio Access
(E-UTRA) and Evolved Universal Terrestrial Radio Access
Network (E-UTRAN); Radio interface protocol aspects, 25.813,
version 7.1.0, Accessed on January 3 2007 from URL:
http://www.3gpp.org/ftp/Specs/html-info/25813.htm
[42] Internet resource: 3GPP, Technical Specification Group Radio
Access Network, Packet Data Convergence Protocol (PDCP)
specification, 25.323, version 7.3.0, Accessed on January 3
2007 from URL:
http://www.3gpp.org/ftp/Specs/html-info/25323.htm
[43] Internet resource: 3GPP, Technical Specification Group
Services and System Aspects, Adaptive Multi-Rate (AMR)
speech CODEC frame structure, 26.101, version 6.0.0,
Accessed on January 3 2007 from URL:
http://www.3gpp.org/ftp/Specs/html-info/26101.htm
[44] Internet Resource: The Internet Engineering Task Force,
Request For Comments, RFC 3095 - RObust Header
Compression (ROHC): Framework and four profiles: RTP,
UDP, ESP, and uncompressed, Accessed on January 6 2007
from URL:http://www.ietf.org/rfc/rfc3095.txt?number=3095
[45] Internet Resource: 3GPP TSG-RAN WG2, R2-063215,
Ericsson: Scheduling Request in E-UTRAN, Accessed on
January 9 2007 from URL:
http://www.3gpp.org/ftp/tsg_ran/WG2_RL2/TSGR2_56/Docum
ents/R2-063215.zip
86 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
[46] Internet Resource: 3GPP TSG-RAN WG2,
R2-062263,
Samsung: First quantification on downlink control overhead,
Accessed on January 9 2007 from URL:
http://www.3gpp.org/ftp/tsg_ran/WG2_RL2/TSGR2_54/Docum
ents/R2-062263.zip
[47] Internet Resource: 3GPP TSG-RAN WG1,
R1-063148,
Ericsson: L1/L2 Signaling for PICH, Accessed on January 9
2007 from URL:
http://www.3gpp.org/ftp/tsg_ran/WG1_RL1/TSGR1_47/Docs/R
1-063148.zip
[48] Internet Resource: 3GPP TSG-RAN WG1,
R1-063147,
Ericsson: Downlink L1/L2 control signaling, Accessed on
January 9 2007 from URL:
http://www.3gpp.org/ftp/tsg_ran/WG1_RL1/TSGR1_47/Docs/R
1-063147.zip
[49] Internet Resource: 3GPP, Technical Specification Group
Services and System Aspects, Mandatory speech CODEC
speech processing functions, AMR speech CODEC, General
description, TS 26.071, version 6.0.0, Accessed on January 18
2007 from URL:
http://www.3gpp.org/ftp/Specs/html-info/26071.htm
[50] P. Björkén et al., Ericsson White Paper, The Evolution of
EDGE, Feb. 2007, 285 23-3107, Uen, Rev.A. Accessed on
Feb. 15, 2007 from URL:
http://www.ericsson.com/technology/whitepapers/3107_The_e
volution_of_EDGE_A.pdf
[51] Internet Resoure: 3GPP, Technical Specification Group Radio
Access Network, HSPA Evolution (FDD), TR 25.999,
Accessed on Feb 19, 2007, from URL:
http://www.3gpp.org/ftp/Specs/archive/25_series/25.999/2599
9-100.zip
[52] Internet resource: C. Berrou et al, Near Shannon-limit ErrorCorrecting Coding and Decoding: Turbo Codes, IEEE proc.
ICC 1993, pp. 1064-1070. Accessed on February 19, 2007
from URL:
http://www-elec.enstbretagne.fr/equipe/berrou/Near%20Shannon%20Limit%20Erro
r.pdf
[53] Internet Resource: Telebit - PEP and the Trailblazer,
Accessed on February 27, 2007 from URL:
http://en.wikipedia.org/wiki/Telebit
[54] Internet Resource: 3GPP TSG-RAN WG2,
R2-070041,
Ericsson: Problems of Persistent Scheduling, Accessed on
March 2 2007 from URL:
http://www.3gpp.org/ftp/tsg_ran/WG2_RL2/TSGR2_56bis/Doc
uments/R2-070041.zip
87 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
[55] Internet Resource: 3GPP TSG-RAN WG2, R2-0662859,
Ericsson: Semi-Persistent Scheduling, Accessed on March 2
2007 from URL:
http://www.3gpp.org/ftp/tsg_ran/WG2_RL2/TSGR2_55/Docum
ents/R2-062859.zip
[56] Internet Resource: 3GPP TSG-RAN WG2,
R2-070055,
Ericsson: Scheduling for maximizing VoIP capacity, Accessed
on March 2 2007 from URL:
http://www.3gpp.org/ftp/tsg_ran/WG2_RL2/TSGR2_56bis/Doc
uments/R2-070055.zip
[57] Internet Resource: 3GPP TSG-RAN WG2, R2-063684, Email
agreement: Downlink scheduling: L1/L2 signalling overhead
reduction, Accessed on March 2 2007 from URL:
http://www.3gpp.org/ftp/tsg_ran/WG2_RL2/TSGR2_56/Docum
ents/R2-063684.zip
[58] Internet Resource: 3GPP TSG-RAN WG2,
R2-070396,
Ericsson, Nokia, Samsung, RLC PDU format for enhanced L2,
Accessed on March 2 2007 from URL:
http://www.3gpp.org/ftp/tsg_ran/WG2_RL2/TSGR2_56bis/Doc
uments/R2-070396.zip
[59] J. Sjöberg et al, Real-Time Transport Protocol (RTP) Payload
Format and File Storage Format for the Adaptive Multi-Rate
(AMR) and Adaptive Multi-Rate Wideband (AMR-WB) Audio
Codecs, RFC 3267, Accessed on April 4 2007 from URL:
http://www.rfc-archive.org/getrfc.php?rfc=3267
[60] Internet Resource: 3GPP TSG-RAN WG4, R4-0608573,
Ericsson: LTE UE Power Class, Accessed on May 3, 2007
from URL:
http://www.3gpp.org/ftp/tsg_ran/WG4_Radio/TSGR4_40/Docs/
R4-060857.zip
[61] Internet Resource: 3GPP TSG-RAN WG2,
Accessed on April 20 2007 from URL:
http://www.3gpp.org/ftp/tsg_ran/WG2_RL2/
Meetings.
[62] Internet Resource: 3GPP TSG-RAN WG2,
R2-063217,
Ericsson: Scheduling for maximizing VoIP capacity, Accessed
on April 24 2007 from URL:
http://www.3gpp.org/ftp/tsg_ran/WG2_RL2/TSGR2_56/Docum
ents/R2-063217.zip
[63] Internet Resource: 3GPP, Technical Specification Group
Services and System Aspects, IP Multimedia Subsystem (IMS)
Multimedia Telephony, Media handling and interaction, TS
26.114, version 7.0.0, Accessed on April 24 2007 from URL:
http://www.3gpp.org/ftp/Specs/html-info/26114.htm
[64] Internet Resource: 3GPP TSG-RAN WG2,
R2-070062,
Ericsson: RLC-MAC Header Formats, Accessed on April 24
2007 from URL:
88 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
http://www.3gpp.org/ftp/tsg_ran/WG2_RL2/TSGR2_56bis/Doc
uments/R2-070062.zip
[65] Internet Resource: The Internet Engineering Task Force,
Request For Comments, RFC 4867 - RTP Payload Format
and File Storage Format for the Adaptive Multi-Rate (AMR)
and Adaptive Multi-Rate Wideband (AMR-WB) Audio Codecs,
Accessed on May 7 2007 from URL:
http://www.rfc-editor.org/rfc/rfc4867.txt
89 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Appendix A: Extended Cyclic Prefix Figures
Figure 16 reflects the increase in overhead that an extended CP-length
requires. The figures below corroborate the assumption that an increase in
energy per payload bit corresponds to the increase in overhead when using
the extended cyclic prefix.
1
1*1
2*2 MIMO
CDF of Fully Loaded 1/1 System with 16 QAM [%]
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
0.5
1
1.5
Bits per Symbol
2
2.5
Figure 36: CDFs of the LTE System Model (Extended Cyclic Prefix)
90 (98)
3
Control Channel Capacity in Downlink (Ideal LA) [Users/20ms]
Overhead Impacts on Long-Term Evolution Radio Networks
500
PER sched, Single Antenna
PER sched, MIMO 2*2
Dynamic Scheduling, Single Antenna
Dynamic Scheduling, MIMO 2*2
450
400
350
300
250
200
150
100
50
0
N4.75
N5.9
N7.4
N12.2
W6.60
AMR Speech CODECs
W8.85
W12.65
Figure 37: Estimated Control Channel Capacity (extended CP)
7
6
x 10
Energy per user
-3
PER Sched, Single Antenna
PER Sched, MIMO 2*2
Dynamic Shed, Single Antenna
Dynamic Shed, MIMO 2*2
Ws/speech frame
5
4
3
2
1
0
N4.75
N5.9
N7.4
N12.2
W6.60
AMR Speech CODECs
Figure 38: Energy per Speech User (extended CP)
91 (98)
W8.85
W12.65
Overhead Impacts on Long-Term Evolution Radio Networks
8.8
4.2
Dynamic Shed, Single Antenna
Dynamic Shed, MIMO 2*2
1.6
Energy per TCP Packet [Ws]
1.4
1.2
1
0.8
0.6
0.4
0.2
0
25
75
300
1000
1500
TCP Payload packet size [Bytes]
4000
64000
Figure 39: Energy per TCP Packet (extended CP)
45
Single Antenna
MIMO 2*2
40
Energy per payload [Ws/Mbit]
35
30
25
20
15
10
5
0
25
75
300
1000
1500
TCP Payload packet size [Bytes]
Figure 40: Energy per TCP Payload (extended CP)
92 (98)
4000
64000
Overhead Impacts on Long-Term Evolution Radio Networks
0.14
Single Antenna
MIMO 2*2
Payload per Energy Quanta [Mbit/Ws]
0.12
0.1
0.08
0.06
0.04
0.02
0
25
75
300
1000
1500
TCP Payload packet size [Bytes]
Figure 41: TCP Payload per Ws (extended CP)
93 (98)
4000
64000
Overhead Impacts on Long-Term Evolution Radio Networks
Appendix B: Concepts
3G-network
A 3GPP system using a UTRAN Radio Access
Network.
3GPP
3rd
Generation
Partnership
Project
(for
Standardization). Formed to continue the technical
specifications of the UMTS standard, started by the
European Telecommunications Standards Institute
(ETSI).
aGW
Access Gateway. Entity in the LTE core network
handling the interaction with the LTE RAN. The aGW
is logically split into two entities (Mobility Mangement
Entity (MME) and User Plane Entity (UPE)) handling
mobility management and user plane transfers
respectively. A potential physical split of these entities
is being discussed in 3GPP (see Figure 6).
Central-Limit
Theorem
Chase Combining
Coherence
Bandwidth
Coherence Time
The sampling distribution of a mean value
approximates Gaussian or normal distribution
regardless of the distribution of the original variable, if
the number of variables is large.
A combining technique based on sending a number of
copies of each coded data packet. The decoder then
combines multiple received copies of the coded
packet weighted by the signal-to-noise-ratio (SNR)
prior to decoding. Also known as Hybrid ARQ- type III
with one redundancy version [40].
The channel frequency separation where the
correlation is considered zero or below an acceptable
threshold [6]. The coherence bandwidth can be
approximated as inversely proportional to the
maximum delay spread, based on Fourier analysis
theory.
Distance between two time instances where the
frequency function can be considered constant. Can
for time-invariant cases be considered inversely
proportional to the Doppler Spread.[6]
94 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Doppler Shift
(general)
The apparent change in wavelength of sound (or light)
caused by the motion of the source, observer or both.
Waves emitted by a moving object (as received by an
observer) will be compressed if approaching and
elongated if receding. The frequency thus is lowered
or increased. How much the frequency will change
depends on how fast the object is moving toward or
away from the receiver. This phenomenon occurs for
both sound and light. [29]
Doppler Spread
Describes how much a pure frequency is spread in
the frequency domain when transferred through a
channel. The spread is delimited by the maximum
Doppler frequencies when moving directly towards or
away from an approaching wave. Doppler spread can
be considered inversely proportional to the
Coherence Time in time-invariant cases. [6]
Flat Fading
If the maximum delay spread is very much smaller
than the symbol time, or the symbol bandwidth is very
much smaller than the coherence bandwidth [6]. For
small frequency separation, transferred symbols will
be subject to similar attenuation and a linear phase
shift over the channel.
Frequency Selective
Fading
If the symbol time is of the same order as the
maximum delay spread, or if the symbol bandwidth is
larger than the coherence bandwidth, there will be
frequency selective fading. [6]
Loading Algorithms
Bit- and Power loading algorithms define the
assignment of power and the scheduling of bits onto a
(sub-)carrier.
Equalization
Signal processing techniques used at the receiver to
combat interference in dispersive channels in the
presence of additive noise [2].
EUL/HSUPA
Enhanced Uplink/High Speed Uplink Packet Access.
Improvement techniques used in WCDMA to achieve
higher data rates on the uplink.
Gap Approximation
The modulation scheme performance (for instance for
16 QAM) is related to the theoretical channel capacity
and the achievable data rates is calculated using
existing capacity formulas.[21]
HSDPA
High Speed Downlink Packet Access. Improvement
techniques used in WCDMA to achieve higher data
rates on the downlink. Also referred to as WCDMA
Evolved, the first step [22].
95 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
IMT-2000
Maximum Delay
Spread
The generic name for the five terrestrial standards for
third generation (3G) wireless communication. [9],
[10], [11].
Time dispersion over a multipath channel below which
energy leakage and subsequent inter-symbol
interference relatively easily can be avoided. The
Maximum Delay Spread can be approximated as
reciprocal to the coherence bandwidth, based on
Fourier analysis theory.
Multipath Fading
Frequency distortion when a signal is composed of a
sum of phase shifted signals with different delays and
attenuation, being the result of reflections over
several transmission paths. In a mobile system there
is both large-scale fading (shadow fading) caused by
large objects, and small-scale fading (Rayleigh
fading) caused by more frequent reflections. Multipath
fading in a mobile system is in simplified terms a
result of Rayleigh-distributed small-scale fading
superimposed on large-scale fading. A signal can also
include a component transmitted directly from a
transmitter to a receiver. If a Line-of-Site component
is included the resulting signal follows a Rice
distribution.
PAR Reduction
The sum of many independent (orthogonal) signals
results, in a near-Gaussian distribution with a large
difference between the peak and average signal
levels, given the central-limit theorem. The reduction
of the Peak-to-Average ratio (PAR) is paramount to
ease receiver resolution requirements, reduce the
need for power, and simplify the general receiver
complexity.
Protocol Overhead
Header information required per packet to
accommodate RTP, PDCP, UDP, IP, RLC and MAC
headers, and cyclic redundancy check (CRC) for VoIP
packets. For TCP packets, the TCP protocol header is
assumed instead of the RTP and UDP headers.
S3G
The study Item in 3GPP, Evolved UTRA and UTRAN,
also known as 3GPP Long-Term Evolution or LTE,
has historically been referred to as Super-3G or S3G.
However the term will not be used considering
ongoing standardization activities. The term referred
to in this thesis will be the more appropriate LTE.
Spectral/Spectrum
Efficiency
The usage of available bandwidth in terms of
available resources, considering for instance service
provisioning and obtainable throughput.
96 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
UE
User Equipment. Defines the end-user handset or
device used in a mobile system of the 3rd standard or
beyond. In this thesis, the terms mobile or handset
will be used interchangeably.
UMTS
A mobile technology for 3rd Generation mobile
systems using either UTRA FDD (WCDMA) or UTRA
TDD (TD-SCDMA at higher chip rates) as modulation
technique. Comprises two of the five standards
defined as parts of the IMT-2000 program [10].
Water-pouring/
water-filling
Theorem
States that more power and higher order modulation
should be allocated where the channel attenuation is
low in order to maximize system capacity [17]. The
naming of the theorem originates from the analogy
with pouring enough water into an irregularly shaped
bowl so that the water level evens out. This theorem
applies to the OFDM applications where power is
distributed over different sub-carriers in accordance
with the respective signal-to-interference-and-noise
ratio. The use of this theorem to maximize the
average symbol energy per OFDM sub-carrier and
the total channel capacity, is further described in [21].
97 (98)
Overhead Impacts on Long-Term Evolution Radio Networks
Appendix C: Abbreviations
aGW
AGWN
ARQ
BER
BLER
CN
CP
CQI
DTX
EDGE
EIRP
FDD
FDM
FEC
GSM
HARQ
HSPA
HSDPA
HSUPA
IBI
ICI
IRC
ISI
LA
LTE
Mbps
MCS
MIMO
MME
MRC
MTU
OFDM
OFDMA
PAR
RAN
RRM
SISO
SNR
SINR
TDD
TTI
UMTS
UPE
UTRAN
VoIP
VAF
WCDMA
WLAN
Access Gateway
Additive Gaussian White Noise
Automatic Repeat Request
Bit Error Rate
Block Error Rate
Core Network
Cyclic Prefix
Channel Quality Indication
Discontinuous Transmission
Enhanced Data Rates for GSM Evolution
Equivalent Isotropic Radiated Power
Frequency Division Duplex
Frequency Division Multiplex
Forward Error Correction (Coding)
Global System for Mobile Communication
Hybrid Automatic Repeat Request
High Speed (Downlink and Uplink) Packet Access
High Speed Downlink Packet Access
High Speed Uplink Packet Access
Inter-Block Interference
Inter-Carrier Interference
Interference Rejection Combining
Inter-Symbol Interference
Link Adaptation
Long-Term Evolution
Megabits per Second
Modulation and Coding Scheme
Multiple Input Multiple Output
Mobility Management Entity
Maximum Ratio Combining
Maximum Transmission Unit
Orthogonal Frequency Division Multiplexing
Orthogonal Frequency Division Multiple Access
Peak-to-Average Ratio
Radio Access Network
Radio Resource Management
Single Input Single Output
Signal-to-Noise Ratio
Signal-to-Interference-and-Noise Ratio
Time Division Duplex
Transmission Time Interval
Universal Mobile Telecommunications System
User Plane Entity
UMTS Terrestrial Radio Access Network
Voice-over-IP (Internet Protocol)
Voice Activity Factor
Wideband Code Division Multiple Access
Wireless Local Area Network
98 (98)
COS/CCS 2007-19
www.kth.se
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