Modeling and Performance Improvement of TCP over LTE Handover MATTEO PACIFICO

Modeling and Performance Improvement of TCP over LTE Handover MATTEO PACIFICO
Modeling and Performance Improvement
of TCP over LTE Handover
MATTEO PACIFICO
Master’s Degree Project
Stockholm, Sweden 2009
CHAOO
CHAOO
CHAOO
“con tanto amore mi avete cresciuto,
con uno sguardo mi avete insegnato a vivere,
con un sorriso mi avete reso felice,
con coraggio avete accettato e sostenuto le mie scelte,
con tanta pazienza mi avete permesso di diventare quel che sono ora.
Siete per sempre nel mio cuore!”
A Mamma e Papà
Abstract
The Long Term Evolution (LTE) of the UMTS Terrestrial Radio Access and
Radio Access Network is a new communication standard aimed for commercial
deployment in 2010. Goals for LTE include support for improved system capacity
and coverage, high peak data rates, low latency, reduced operating costs, multiantenna support, flexible bandwidth operations and seamless integration with
existing systems.
The aim of this thesis project is to study the impacts on the end-user and system performance when users with high bit rates tcp services are moving through
the network. These impacts affect the reduced end-user or system throughput,
e.g., due to congestion in the transport network, leading to poor utilization of
the transport and radio resources available. To reach such an aim, it has been
necessary (1) to create a new simulator with the ns-2 and perform simulation in
different network settings and (2) formulate a mathematical model able to capture the principal dynamics of the real system. Possible solutions to mitigate the
impacts are investigated by comparing the simulations results of tcp performance
in the radio and transport network, with the mathematical model of it.
The project has been performed at the Automatic Control Lab at KTH in
Stockholm and in collaboration with Ericsson Research Lab. The work is part of a
joint effort with Davide Pacifico. Further results are available in the master thesis
“Analysis and Performance Improvement of tcp during handover of LTE ”[1].
iii
Preface
LTE, short for Long Term Evolution, is the result of ongoing work by the 3rd
Generation Partnership Project (3GPP), a collaborative group of international
standards organizations and mobile-technology companies. 3GPP set out in 1998
to define the key technologies for the third generation of GSM-based mobile networks (3G), and its work has continued to define the ongoing evolution of these
networks. Near the end of 2004, discussions on the longer-term evolution of 3G
networks began, and a set of high-level requirements for LTE was defined: the
networks must transmit data at a reduced cost per bit compared to 3G; they must
be able to offer more services at lower transmission cost with better user experience; LTE must have the flexibility to operate in a wide number of frequency
bands; it should utilize open interfaces and offer a simplified architecture; and
it must have reasonable power demands on mobile terminals. Standardization
work on LTE is continuing with some operators projected to deploy the first LTE
networks in 2009.
LTE (“Turbo 3G”), is a wireless broadband technology designed to support
roaming Internet access via cell phones and handheld devices. Because LTE offers
significant improvements over older cellular communication standards, some refer
to it as a 4G (fourth generation) technology along with WiMax.
LTE defines new radio connections for mobile networks, and will utilize Orthogonal Frequency Division Multiplexing (OFDM), a widely used modulation
v
Preface
technique that is the basis for Wi-Fi, WiMAX, and the Digital Video Broadcasting (DVB) and Digital Audio Broadcasting (DAB) technologies. The targets for
LTE indicate bandwidth increases as high as 100 Mbps on the downlink, and up
to 50 Mbps on the uplink. However, this potential increase in bandwidth is just
a small part of the overall improvement LTE aims to provide. LTE is optimized
for data traffic, and it will not feature a separate, circuit-switched voice network,
as in 2G GSM and 3G UMTS networks.
LTE is the successor to the current generation of UMTS 3G technology, which
is based upon WCDMA (3G), HSDPA, HSUPA, and HSPA. LTE is not a replacement for UMTS in the way that UMTS was a replacement for GSM, but rather
an update to the UMTS technology that will enable it to provide significantly
faster data rates for both uploading and downloading.
With its architecture centered on Internet Protocol (IP), Long Term Evolution
promises to have excellent support for browsing Web sites, VoIP and other IPbased services. LTE can theoretically support downloads at 100 Megabits per
second (Mbps) or more based on experimental trials.
Another important feature of LTE is the amount of flexibility it allows operators in determining the spectrum in which it will be deployed. Not only will LTE
have the ability to operate in a number of different frequency bands (meaning operators will be able to deploy it at lower frequencies with better propagation characteristics), but it also features scalable bandwidth. Whereas WCDMA/HSPA
uses fixed 5 MHz channels, the amount of bandwidth in an LTE system can be
scaled from 1.25 to 20 MHz. This means networks can be launched with a small
amount of spectrum, alongside existing services, and adding more spectrum as
users switch over. It also allows operators to tailor their network deployment
strategies to fit their available spectrum resources, and not have to make their
spectrum fit a particular technology.
vi
Preface
The Transmission Control Protocol (tcp) is one of the core protocols of the
Internet Protocol Suite. TCP is so central that the entire suite is often referred
to as tcp/ip. tcp has been optimized for wired networks. Any packet loss is
considered to be the result of congestion and the congestion window size is reduced
dramatically as a precaution. However, wireless links are known to experience
sporadic and usually temporary losses due to fading, shadowing, handoff, and
other radio effects, that cannot be considered congestion. After the (erroneous)
back-off of the congestion window size, due to wireless packet loss, there can be
a congestion avoidance phase with a conservative decrease in window size. This
causes the radio link to be under utilized. Extensive research has been done
on the subject of how to combat these harmful effects. Suggested solutions can
be categorized as end-to-end solutions (which require modifications at the client
and/or server), link layer solutions, or proxy based solutions, which require some
changes in the network without modifying end nodes.
In cellular telecommunications, the term handoff refers to the process of transferring an ongoing call or data session from one channel connected to the core
network to another. The British English term for transferring a cellular call is
handover, which is the terminology standardized by 3GPP within such European
originated technologies as GSM and UMTS.
The aim of this thesis is to study the performances of the tcp protocol in the
handover procedure in LTE. This study proposes a mathematical model of that
features a compromise between simplicity and accuracy in the representation of
the dynamics of the real system.
The handover of an end-user who moves in a train or a bus or walks in a
crowded street while is surfing on internet by his terminal could be an abrupt
service interruptions. It is therefore important that the operator control properly
the handover procedure to avoid these problems.
vii
Preface
For these reasons is essential put the attention to mitigate the problems of
TCP during the handover and find a way to keep it transparent to the end-users.
To reach this aim, our analytical model is compared with a set of simulations of
the real network scenario performed by an original LTE simulation implemented
the ns-2 environment.
The thesis at a glance
This thesis work is organized in five chapters.
Chapter 1 is an overview of the LTE system and its main new features, such as
the system architecture, the Hybrid Automatic Repeat reQuest and the Evolved
RAN. A section is dedicated to LTE handover in which the procedure is explained
in detail.
In Chapter 2 the tcp protocol is described and its combination with LTE. In
particular we focus our attention on the tcp Reno.
In Chapter 3 we investigate problems arising during the handover procedure
in LTE network and we will show how to solve or mitigate it. We introduce
also an ns-2 network simulator and how it is implemented to realize our LTE
simulations.
In Chapter 5 we develop a model of TCP during handover, which is based
on the Joint link model proposed by Möller in 2008 [2]. This model shows the
behavior of the TCP congestion window of a mobile that does the LTE handover
procedure.
Finally in Chapter 6, we discuss the results and propose some developments
for future studies.
viii
Contents
Abstract
iii
Preface
v
Contents
xi
List of Figures
xiv
List of Tables
xv
Abbreviations
xvii
1 Background
1.1
3
LTE (Long Term Evolution) . . . . . . . . . . . . . . . . . . . . .
3
1.1.1
Requirements and performance goals for LTE . . . . . . .
4
1.1.2
System architecture description . . . . . . . . . . . . . . .
8
1.1.3
Key Features . . . . . . . . . . . . . . . . . . . . . . . . .
10
1.1.4
Network Sharing . . . . . . . . . . . . . . . . . . . . . . .
11
1.2
Mobility management . . . . . . . . . . . . . . . . . . . . . . . . .
17
1.3
LTE Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . .
18
1.3.1
OFDM . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
18
1.3.2
Multiple antenna techniques . . . . . . . . . . . . . . . . .
19
1.3.3
Scheduling . . . . . . . . . . . . . . . . . . . . . . . . . . .
24
1.4
Handover . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
24
1.5
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
28
ix
Contents
2 TCP and LTE
2.1
Importance of TCP . . . . . . . . . . . . . . . . . . . . . . . . . .
30
2.2
TCP key features . . . . . . . . . . . . . . . . . . . . . . . . . . .
31
2.2.1
Flow control . . . . . . . . . . . . . . . . . . . . . . . . . .
31
TCP Applicability . . . . . . . . . . . . . . . . . . . . . . . . . .
33
2.3.1
Congestion control . . . . . . . . . . . . . . . . . . . . . .
35
2.3.2
Explicit Congestion Notification (ECN) . . . . . . . . . . .
39
2.4
Evaluation of TCP Performance . . . . . . . . . . . . . . . . . . .
40
2.5
Bottleneck in the core network . . . . . . . . . . . . . . . . . . . .
42
2.5.1
GPRS Tunneling Protocol (GTP) . . . . . . . . . . . . . .
42
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
45
2.3
2.6
3 Management of LTE handover
3.1
47
Forwarding procedure . . . . . . . . . . . . . . . . . . . . . . . . .
47
3.1.1
Optimal case . . . . . . . . . . . . . . . . . . . . . . . . .
49
3.1.2
Window halving case . . . . . . . . . . . . . . . . . . . . .
49
3.1.3
Timeout occurrence case . . . . . . . . . . . . . . . . . . .
50
Two improving solutions . . . . . . . . . . . . . . . . . . . . . . .
54
3.2.1
Advancing the path switch command . . . . . . . . . . . .
54
3.2.2
Predicting the handover occurrence . . . . . . . . . . . . .
55
3.3
Validation of the solutions . . . . . . . . . . . . . . . . . . . . . .
56
3.4
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
59
3.2
4 LTE simulator: an overview
x
29
61
4.1
LTE simulator through ns-2 and nsMiracle . . . . . . . . . . . . .
61
4.2
Network topology . . . . . . . . . . . . . . . . . . . . . . . . . . .
64
4.3
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
65
Contents
5 Mathematical model of TCP over LTE
67
5.1
Existing models . . . . . . . . . . . . . . . . . . . . . . . . . . . .
68
5.2
Model adaptation . . . . . . . . . . . . . . . . . . . . . . . . . . .
70
5.2.1
Single flow and multiple flow model . . . . . . . . . . . . .
72
Application scenario . . . . . . . . . . . . . . . . . . . . . . . . .
73
5.3.1
Handover preparation . . . . . . . . . . . . . . . . . . . . .
73
5.3.2
Handover actuation . . . . . . . . . . . . . . . . . . . . . .
74
5.3.3
Handover completion . . . . . . . . . . . . . . . . . . . . .
74
5.4
Simulations and Results . . . . . . . . . . . . . . . . . . . . . . .
75
5.5
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
78
5.3
6 Conclusion and future work
79
6.1
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
79
6.2
Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
79
Bibliography
81
xi
List of Figures
1.1.1 FDD and TDD operating bands.
. . . . . . . . . . . . . . . . . .
8
1.1.2 LTE architecture. . . . . . . . . . . . . . . . . . . . . . . . . . . .
9
1.1.3 Layer 2 structure for Downlink. . . . . . . . . . . . . . . . . . . .
13
1.1.4 Example of mapping of logical channels to transport channels. . .
16
1.2.1 Mobility states of the UE in LTE. . . . . . . . . . . . . . . . . . .
17
1.3.1 Downlink reference signal structure - normal cyclic prefix, two
transmit antennas . . . . . . . . . . . . . . . . . . . . . . . . . . .
20
1.3.2 Radio channel access modes . . . . . . . . . . . . . . . . . . . . .
20
1.3.3 Channel-dependent scheduling. . . . . . . . . . . . . . . . . . . .
24
1.4.1 Message chart of the LTE handover procedure. . . . . . . . . . . .
26
2.2.1 Sliding window mechanism. . . . . . . . . . . . . . . . . . . . . .
32
2.3.1 tcp slow start and congestion avoidance. . . . . . . . . . . . . . .
37
2.3.2 tcp fast retransmit and fast recovery. . . . . . . . . . . . . . . . .
39
2.3.3 The TOS field in ip header and ECN signaling. . . . . . . . . . .
40
2.3.4 Bytes 13 and 14 of the tcp header. . . . . . . . . . . . . . . . . .
40
2.5.1 GTP-U Protocol Stack. . . . . . . . . . . . . . . . . . . . . . . . .
43
2.5.2 GTPv1 header (default and optional fields). . . . . . . . . . . . .
44
3.1.1 rtt behavior during the LTE handover . . . . . . . . . . . . . . .
47
xiii
List of Figures
3.1.2 cwnd during handover in case of optimal case, window halving and
timeout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
48
3.1.3 Difference between optimal case and window halving . . . . . . .
50
3.1.4 Difference between optimal case and timeout verification . . . . .
51
3.1.5 Normalized throughput during the time interval (rtt1 , rtt2 ) considering optimal case, window halving and timeout for different
values of cwndMAX . . . . . . . . . . . . . . . . . . . . . . . . . . .
52
3.1.6 Different averages of the normalized throughput on varying the
cwndMAX parameter . . . . . . . . . . . . . . . . . . . . . . . . . .
53
3.2.1 Advancing the path switch command in the handover message chart 54
3.2.2 Modification to the handover message chart to predict the handover. 55
3.3.1 cwnd and rtt of the reference ME during the handover.
xiv
. . . . .
57
3.3.2 Reference ME received sequence number. . . . . . . . . . . . . . .
58
4.1.1 One node of nsMIracle. . . . . . . . . . . . . . . . . . . . . . . . .
62
4.2.1 Reference simulation scenario. . . . . . . . . . . . . . . . . . . . .
64
5.2.1 LTE handover scenario. . . . . . . . . . . . . . . . . . . . . . . . .
71
5.4.1 Bit rate comparison. . . . . . . . . . . . . . . . . . . . . . . . . .
76
5.4.2 cwnd comparison. . . . . . . . . . . . . . . . . . . . . . . . . . . .
76
5.4.3 Buffer status comparison.
. . . . . . . . . . . . . . . . . . . . . .
77
5.4.4 rtt comparison. . . . . . . . . . . . . . . . . . . . . . . . . . . .
78
List of Tables
1.1.1 LTE performance goal . . . . . . . . . . . . . . . . . . . . . . . .
5
1.1.2 LTE user throughput and spectrum efficiency requirements . . . .
6
1.1.3 Interruption time requirements, LTE-GSM and LTE-WCDMA. . .
7
1.1.4 Logical channel characterized by the transferred information. . . .
14
1.1.5 Transport channel characterized by how the data is transferred
over radio interface. . . . . . . . . . . . . . . . . . . . . . . . . . .
15
1.3.6 E-UTRA Numerology . . . . . . . . . . . . . . . . . . . . . . . . .
18
3.3.1 Simulations parameters about the solution proposed in Section 3.2. 56
4.2.1 Common parameters of all simulation . . . . . . . . . . . . . . . .
65
5.4.1 Parameters for simulation results and mathematical model comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
xv
75
Abbreviations
3GPP
AIPN
AM
AMC
AQM
ARQ
CDM
CN
CQI
CWR
DL
DAE
DSCP
eNB
ECE
ECN
EDGE
EPC
EPS
E-UTRA
E-UTRAN
FDM
FDD
FFT
GPRS
GSM
3rd Generation Partnership Project
All IP Network
Acknowledged Mode
Adaptive Modulation and Coding
Active Queue Management
Automatic Repeat Request
Code Division Multiplexing
Core Network
Channel Quality Information
Congestion Window Reduced
Downlink
Differential Algebraic Equation
Differentiated Services Codepoint
eNodeB
ECN-echo
Explicit Congestion Notification
Enhanced Data rates for GSM Evolution
Evolved Packet Core
Evolved Packet System
Evolved UMTS Terrestrial Radio Access
Evolved UMTS Terrestrial Radio Access Network
Frequency Division Multiplexing
Frequency Division Duplex
Fast Fourier Transform
General Packet Radio Service
Global System for Mobile communications
xvii
Abbreviations
GTP
GW
HARQ
HO
HSPA
HSS
IFFT
IP
LTE
NAS
MAC
ME
MIMO
MISO
MME
MMOG
NAS
NRT
OFDM
PDCP
PDN
PDN GW
PHY
PLMN
QoS
RACH
RAN
RAT
RED
RLC
RNC
ROHC
RRC
RT
RTO
xviii
GPRS Tunneling Protocol
Gateway
Hybrid ARQ
Handover
High Speed Packet Access
Home Subscriber Server
Inverse FFT
Internet Protocol
Long Term Evolution
Non-Access Stratum
Medium Access Control
Mobile Equipment
Multiple Input Multiple Output
Multiple Input Single Output
Mobility Management Entity
Multimedia Online Gaming
Non-Access Stratum
Non-Real Time
Orthogonal Frequency Division Multiplexing
Packet Data Convergence Protocol
Packet Data Network
Packet Data Network Gateway
Physical
Public Land Mobile Network
Quality of Services
Random Access Channel
Radio Access Network
Radio Access Technology
Random Early Discard
Radio Link Control
Radio Network Controller
Robust Header Compression
Radio Resource Control
Real Time
Retransmission Timeout
Abbreviations
RTT
RS
SAE
SDU
SeNB
SFN
SGW
SIMO
SISMO
SNR
TA
TB
TDD
TEID
TeNB
TM
TCP
TDM
TOS
TTI
UE
UL
UM
UMTS
VIP
WCDMA
WFQ
Round Trip Time
Reference Symbol
System Architecture Evolution
Service Data Unit
Source eNodeB
Single frequency network
Serving Gateway
Single Input Multiple Output
Single Input Single Output
Signal-to-Noise Ratio
Tracking Area
Transport Block
Time Division Duplex
Tunnel Endpoint Identifier
Target eNodeB
Transparent Mode
Transport Control Protocol
Time Division Multiplexing
Type Of Service
Transmission Time Interval
User Equipment
Uplink
Unacknowledged Mode
Universal Mobile Telecommunications System
Voice over IP
Wireless Coded Division Multiple Access
Weighted Fair Queue
xix
Chapter 1
Background
In this chapter we describe the key aspects of LTE. The chapter is then concluded with a detailed description of the handover mechanism of LTE.
1.1
LTE (Long Term Evolution)
The recent increase of mobile data usage and the emergence of new applications, such as Multimedia Online Gaming (MMOG), mobile TV, Web 2.0, streaming contents, have motivated the 3rd Generation Partnership Project (3GPP) to
work on the Long Term Evolution (LTE). LTE is the latest standard in the mobile network technology tree, which previously implemented the GSM/EDGE
and UMTS/HSxPA network technologies now account for over 85% of all mobile subscribers. LTE will ensure 3GPP’s competitive edge over other cellular
technologies.
LTE, whose radio access is called Evolved UMTS Terrestrial Radio Access
Network (E-UTRAN), is expected to substantially improve end-user throughputs and sector capacity also to reduce user plane latency, bringing significantly
improved user experience with full mobility. With the emergence of Internet
Protocol (ip) as the protocol of choice for carrying all types of traffic, LTE is
scheduled to provide support for ip-based traffic with end-to-end Quality of ser-
3
1. Background
vice (QoS). Voice traffic will be supported mainly as Voice over ip (VIP) enabling
better integration with other multimedia services. Initial deployments of LTE are
expected by 2010 and commercial availability on a larger scale 1-2 years later.
Unlike High Speed Packet Access (HSPA), which was accommodated within
the Release 99 UMTS architecture, 3GPP is specifying a new Packet Core, the
Evolved Packet Core (EPC) network architecture to support the E-UTRAN
through a reduction in the number of network elements, simpler functionality, improved redundancy but most importantly allowing for connections and handover
to other fixed line and wireless access technologies, giving the service providers
the ability to deliver a seamless mobility experience.
LTE has been set aggressive performance requirements that rely on physical layer technologies, such as: Orthogonal Frequency Division Multiplexing
(OFDM), Multiple-Input Multiple-Output (MIMO) systems and Smart Antennas to achieve the baseline targets. The main objectives of LTE are to minimize
the system and User Equipment (UE) complexities, allow flexible spectrum deployment in existing or new frequency spectrum and to enable co-existence with
other 3GPP Radio Access Technologies (RATs).
1.1.1
Requirements and performance goals for LTE
E-UTRA is expected to support different types of services including web
browsing, FTP, video streaming, VIP, online gaming, real time video, push-totalk and push-to-view. Therefore, LTE is being designed to be a high data rate
and low latency system as indicated by the key performance criteria shown in
Table 1.1.1. The bandwidth capability of a UE is expected to be 20MHz for both
transmission and reception. The service provider can however deploy cells with
any of the bandwidths listed in the table. This gives flexibility to the service
providers to tailor their offering dependent on the amount of available spectrum
4
LTE (Long Term Evolution) 1.1
Metric
Peak data rate
Mobility support
Requirement
DL: 100 Mbps
UL: 50 Mbps
(for 20 MHz spectrum)
Up to 500 km/h but optimized
for low speeds from 0 to 15 km/h
< 100 ms (for idle to active)
Control plane latency
(Transition time to active state)
User plane latency
< 5 ms
Control plane capacity
> 200 users per cell
(for 5 MHz spectrum)
Coverage (Cell sizes)
5-100 km with slight
degradation after 30 km
Spectrum flexibility
1.25, 2.5, 5, 20 MHz
Table 1.1.1. LTE performance goal
or the ability to start with limited spectrum for lower upfront cost and grow the
spectrum for extra capacity.
During 2005 the 3GPP activity on 3G evolution was setting the requirement
for LTE. These are documented in [8] and were divided into seven categories:
• capabilities,
• system performance,
• deployment-related aspects,
• architecture and migration,
• radio resource management,
• complexity,
• general aspects.
5
1. Background
Performance measure
Downlink target
Uplink target
relative to baseline relative to baseline
3x-4x
2x-3x
Average user throughput
(per MHz)
Cell-edge user throughput 2x-3x
(per MHz, 5th percentile)
Spectrum efficiency
3x-4x
(bit/s/Hz/cell)
2x-3x
2x-3x
Table 1.1.2. LTE user throughput and spectrum efficiency requirements
Capabilities
The targets are 100 Mbit/s for downlink and 50 Mbit/s for uplink when operating in 20 MHz spectrum allocation. Thus, the requirements can be expressed
as 5 bit/s/Hz for the downlink and 2.5 bit/s/Hz for uplink.
There are two kind of latency requirements: control-plane requirements and
user-plane requirements. The control-plane latency requirements address the delay for transiting from different non-active terminal states to an active state.
The user-plane latency requirement is expressed as the time it takes to transmit
a small ip packet from the terminal to the RAN edge node or vice versa measured
on the ip layer.
System performance
The LTE system performance design targets user throughput, spectrum efficiency, mobility and coverage. The first two are summarized in Table 1.1.2
The mobility requirements focus on the mobile terminals speed. For speeds
up to 15 km/h there are the best performances; for speeds up to 120 km/h there
should be high achievement and for speeds above 120 km/h the system should
be able to keep the connection. The maximum speed allowed in the LTE system
is set to 350 km/h or 500 km/h depending on the frequency band.
6
LTE (Long Term Evolution) 1.1
Non-real-time (ms) Real-time (ms)
relative to baseline relative to baseline
LTE to WCDMA 500
300
LTE to GSM
500
300
Table 1.1.3. Interruption time requirements, LTE-GSM and LTE-WCDMA.
The coverage requirements focus on the cell range. The cells up to 5 km of
radius allow non-interference-limited scenarios; for cells range up to 30 km, a
slight performances degradation are tolerated and cell ranges up to 100 km are
not precluded but no performance requirements are stated yet.
Deployment-related aspects
The requirement on the deployment scenario includes both the case when the
LTE system is deployed as a stand-alone system and the case when it is deployed
together with WCDMA/HSPA and/or GSM. For mobile terminals supporting
those technologies, Table 1.1.3 lists the interruption requirements, that is, longest
acceptable interruption in the radio link when moving between the different radio
access technologies, for both real-time and non-real-time services.
The basis for the requirements on spectrum flexibility is the requirement for
LTE to be deployed in existing IMT-2000 frequency bands (coexistence with the
systems that are already deployed in those bands), that is LTE should support
both Frequency Division Duplex (FDD), and Time Division Duplex (TDD) (Figure 1.1.1).
Architecture and migration
LTE RAN architecture should be packet based (and also support real-time
class traffic), simplify and minimize the interface, support an end-to-end QoS
and designed to minimize the jitter (for example, for tcp/ip traffic type) [8].
7
1. Background
TDD-Based radio access
FDD-Based radio access
Uplink
1900
1920
Downlink
1980
2010
2025
2110
2170
Frequency (MHz)
Figure 1.1.1. FDD and TDD operating bands.
Radio resource management
The radio resource management requirements are divided into enhanced support for end-to-end QoS, efficient support for transmission of higher layers, and
support of load sharing and policy management across different radio access technologies.
Complexity and general aspects
LTE complexity requirements imply that the number of options should be
minimized with no redundant required features. This has an impact also to
the cost and service related aspects that, specific to the cost, it is desirable to
minimize it maintaining the desired performance.
1.1.2
System architecture description
The architecture consists of the following functional elements:
Evolved Radio Access Network (RAN)
The evolved RAN for LTE is composed of a single node, i.e., the eNodeB (eNB)
that interfaces with the UE. The eNB hosts the PHYsical (PHY), Medium Access
Control (MAC), Radio Link Control (RLC), and Packet Data Control Protocol
(PDCP) layers that include the functionality of user-plane header-compression
8
LTE (Long Term Evolution) 1.1
and encryption. It also offers Radio Resource Control (RRC) functionality corresponding to the control plane. It performs many functions including radio resource management, admission control, scheduling, enforcement of negotiated UL
QoS, cell information broadcast, ciphering/deciphering of user and control plane
data, and compression/decompression of DL/UL user plane packet headers.
User Plane
App
TCP
IP
IP
PDCP
PDCP
GTP
GTP
RLC/MAC
RLC/MAC
UDP/IP
UDP/IP
PHY
PHY
L1
L1
S-GW
S1-UP
eNodeB
S1-CP
ME
IP transport
network
MME
Control Plane
X2
NAS
NAS
RRC
RRC
S1-AP
S1-AP
PDCP
PDCP
SCTP
SCTP
RLC/MAC
RLC/MAC
IP
IP
PHY
PHY
L1
L1
Figure 1.1.2. LTE architecture.
Serving Gateway (SGW)
The SGW routes and forwards user data packets, while also acting as the
mobility anchor for the user plane during inter-eNB handovers and as the anchor
for mobility between LTE and other 3GPP technologies (terminating S4 interface
[7] and relaying the traffic between 2G/3G systems and Packet Data Network
Gateway, PDN GW). For idle state UEs, the SGW terminates the DL data path
9
1. Background
and triggers paging when DL data arrives for the UE. It manages and stores
UE contexts, e.g. parameters of the ip bearer service, network internal routing
information. It also performs replication of the user traffic in case of lawful
interception.
Mobility Management Entity (MME)
The MME is the key control-node for the LTE access network. It is responsible
for idle mode UE tracking and paging procedure including retransmissions. It
is involved in the bearer activation/deactivation process and is also responsible
for choosing the SGW for a UE at the initial attach and at the time of intraLTE handover involving Core Network (CN) node relocation. It is responsible
for authenticating the user (by interacting with the Home Subscriber Server,
HSS). The Non-Access Stratum (NAS) signaling terminates at the MME and it
is also responsible for generation and allocation of temporary identities to UEs.
It checks the authorization of the UE to camp on the service provider’s Public
Land Mobile Network (PLMN) and enforces UE roaming restrictions. The MME
is the termination point in the network for ciphering/integrity protection for
NAS signaling and handles the security key management. Lawful interception
of signaling is also supported by the MME. The MME also provides the control
plane function for mobility between LTE and 2G/3G access networks with the
S3 interface terminating at the MME from the SGSN. The MME also terminates
the S6a interface towards the home HSS for roaming UEs.
1.1.3
Key Features
EPS to EPC
A key feature of the EPS is the separation of the network entity that performs
control-plane functionality (MME) from the network entity that performs bearer-
10
LTE (Long Term Evolution) 1.1
plane functionality (SGW) with a well defined open interface between them (S11).
Since E-UTRAN will provide higher bandwidth to enable new services as well
as to improve existing ones, separation of MME from SGW implies that SGW
can be based on a platform optimized for high bandwidth packet processing,
where as the MME is based on a platform optimized for signaling transactions.
This enables selection of more cost-effective platforms for, as well as independent
scaling, of each of these two elements. Service providers can also choose optimized
topological locations of SGWs within the network independent of the locations of
MMEs in order to optimize bandwidth, reduce latencies and avoid concentrated
points of failure.
S1-flex Mechanism
The S1-flex concept provides support for network redundancy and load sharing
of traffic across network elements in the CN, the MME and the SGW, by creating
pools of MMEs and SGWs and allowing each eNB to be connected to multiple
MMEs and SGWs in a pool.
1.1.4
Network Sharing
The LTE architecture enables service providers to reduce the cost of owning
and operating the network by allowing the service providers to have separate CN
(MME, SGW, PDN GW) while the E-UTRAN (eNBs) is jointly shared by them.
This is enabled by the S1-flex mechanism by enabling each eNB to be connected
to multiple CN entities. When a UE attaches to the network, it is connected to
the appropriate CN entities based on the identity of the service provider sent by
the UE.
In this section, we describe the functions of the different protocol layers and
their location in the LTE architecture. In Figure 1.1.2, the NAS protocol, which
11
1. Background
runs between the MME and the UE, is used for control-purposes such as network
attach, authentication, setting up of bearers, and mobility management. All NAS
messages are ciphered and integrity protected by the MME and UE. The RRC
layer in the eNB makes handover decisions based on neighbor cell measurements
sent by the UE, pages for the UEs over the air, broadcasts system information,
controls UE measurement reporting such as the periodicity of Channel Quality
Information (CQI) reports and allocates cell-level temporary identifiers to active
UEs. It also executes transfer of UE context from the Source eNB to the Target
eNB during handover, and does integrity protection of RRC messages. The RRC
layer is responsible for the setting up and maintenance of radio bearers.
In the user-plane, the PDCP layer is responsible for compression/decompression the headers of user plane ip packets using Robust Header Compression
(ROHC) to enable efficient use of air interface bandwidth. This layer also performs ciphering of both user plane and control plane data. Because the NAS
messages are carried in RRC, they are effectively double ciphered and integrity
protected, once at the MME and again at the eNB.
The RLC layer is used to format and transport traffic between the UE and the
eNB. RLC provides three different reliability modes for data transport: Acknowledged Mode (AM), Unacknowledged Mode (UM), or Transparent Mode (TM).
The UM mode is suitable for transport of Real Time (RT) services because such
services are delay sensitive and cannot wait for retransmissions. The AM mode,
on the other hand, is appropriate for non-RT (NRT) services such as file downloads. The TM mode is used when the PDU sizes are known a priori such as for
broadcasting system information. The RLC layer also provides in-sequence delivery of Service Data Units (SDUs) to the upper layers and eliminates duplicate
SDUs from being delivered to the upper layers. It may also segment the SDUs
depending on the radio conditions.
12
LTE (Long Term Evolution) 1.1
Furthermore, there are two levels of re-transmissions for providing reliability,
namely, the Hybrid Automatic Repeat reQuest (HARQ) at the MAC layer and
outer ARQ at the RLC layer. The outer ARQ is required to handle residual errors
that are not corrected by HARQ, that is kept simple by the use of a single bit
error-feedback mechanism. An N-process stop-and-wait HARQ is employed that
has asynchronous re-transmissions in the DL and synchronous re-transmissions
in the UL. Synchronous HARQ means that the re-transmissions of HARQ blocks
occur at pre-defined periodic intervals, hence, no explicit signaling is required to
indicate to the receiver the retransmission schedule. Asynchronous HARQ offers
the flexibility of scheduling re-transmissions based on air interface conditions.
The Figure 1.1.3 show the structure of layer 2 for DL. The PDCP, RLC and
MAC layers together constitute layer 2.
Radio Bearers
ROCH
ROCH
ROCH
ROCH
Security
Security
Security
Security
Segm.
ARQ
Segm.
Segm.
Segm.
ARQ
ARQ
ARQ
PDCP
RLC
BCCH
PCCH
Logical Channels
Scheduling / Priority Handling
MAC
Multiplexing UE1
Multiplexing UEn
HARQ
HARQ
Transport Channels
Figure 1.1.3. Layer 2 structure for Downlink.
In LTE, there is significant effort to simplify the number and mappings of
logical and transport channels. The different logical and transport channels in
LTE are illustrated in Table 1.1.4 and Table 1.1.5, respectively.
The transport channels are distinguished by the characteristics (e.g. adap-
13
1. Background
Channel type
Control channels
(carry control
plane info)
Channel Name
Broadcast Control
Channel (BCCH)
Paging
Control
Channel (PCCH)
Common
Control
Channel (CCCH)
Multicast Control
Channel (MCCH)
Dedicated Control
Channel (DCCH)
Traffic channels
(carry user plane
info)
Dedicated
Traffic
Channel (DTCH)
Multicast
Traffic
Channel (MCCH)
Carried information
DL channel for broadcasting system control info
DL channel for transferring paging
UL channel for transmitting control info and used
by UE without RRC connection
DL
point-to-multipoint
channel for transmitting
MBMS control info
DL point-to-point bidirectional channel for
exchanging control information and used by UE
with RRC connection
Bi-directional
channel
dedicated to a single UE
DL
point-to-multipoint
channel for transmission
of MBMS data
Table 1.1.4. Logical channel characterized by the transferred information.
14
LTE (Long Term Evolution) 1.1
Channel type
Channel Name
Broadcast Channel
(BCH)
Downlink
Shared
Channel (DL-SCH)
Downlink
channels
Paging
Channel
(PCH)
Multicast Channel
(MCCH)
Uplink
Channel
SCH)
Shared
(UL-
Uplink channels
Random
Access
Channel (RACH)
Table 1.1.5.
Carried information
fixed transport format
HARQ, dynamic link
adaptation,
support
for UE DRX, dynamic
and semi-static resource
allocation
required to be broadcast
Support for SFN combining and semi-static resource allocation
HARQ, dynamic link
adaptation,
support
for UE DRX, dynamic
and semi-static resource
allocation
limited control information, collision risk
Transport channel characterized by how the data is transferred
over radio interface.
tive modulation and coding) with which the data are transmitted over the radio
interface. The MAC layer performs the mapping between the logical channels
and transport channels, schedules the different UEs and their services in both UL
and DL depending on their relative priorities, and selects the most appropriate
transport format.
The logical channels are characterized by the information carried by them.
The mapping of the logical channels to the transport channels is shown in Figure 1.1.4.
The physical layer at the eNB is responsible for protecting data against chan-
15
1. Background
Downlink or uplink
DTCH
DCCH
Downlink only
BCCH
PCCH
BCH
PCH
MTCH
MCCH
Logical Channels
Transport Channels
DL-SCH
UL-SCH
MCH
Figure 1.1.4. Example of mapping of logical channels to transport channels.
nel errors using adaptive modulation and coding (AMC) schemes based on channel conditions. It also maintains frequency and time synchronization and performs RF processing including modulation and demodulation. In addition, it
processes measurement reports from the UE such as CQI and provides indications to the upper layers.
The minimum unit of scheduling is a time-frequency block corresponding to
one sub-frame (1 ms) and 12 sub-carriers. The scheduling is not done at a subcarrier granularity in order to limit the control signaling. QPSK, 16QAM and
64QAM will be the DL and UL modulation schemes in E-UTRA. For UL, 64-QAM
is optional at the UE. Each radio frame is 10ms long containing 10 sub-frames
with each sub-frame capable of carrying 14 OFDM symbols. For more details on
these access schemes, refer to [4].
Multiple antennas at the UE are supported with the 2 receive and 1 transmit
antenna configuration being mandatory. MIMO (multiple input multiple output)
is also supported at the eNB with two transmit antennas being the baseline
configuration. Orthogonal Frequency Division Multiple Access (OFDMA) with
a sub-carrier spacing of 15 kHz and Single Carrier Frequency Division Multiple
Access (SC-FDMA) have been chosen as the transmission schemes for the DL
and UL, respectively.
16
Mobility management 1.2
1.2
Mobility management
Power-Up
LTE_ACTIVE
LTE_DETACHED
• IP
• No IP address
• Position not known
address assigned
to known cell
• Connected
OUT_OF_SINK
• DL reception possible
• No UL transmission
LTE_IDLE
• IP address assigned
• Position partially known
• DL DRX period
IN_SINK
• DL reception possible
• UL transmission possible
Figure 1.2.1. Mobility states of the UE in LTE.
Mobility management can be classified based on the radio technologies of the
source and the target cells, and the mobility-state of the UE. From a mobility perspective, the UE can be in one of the three states: LTE DETACHED, LTE IDLE,
and LTE ACTIVE as shown in Figure 1.2.1. LTE DETACHED state is typically
a transitory state in which the UE is powered-on but is in the process of searching
and registering with the network. In the LTE ACTIVE state, the UE is registered
with the network and has an RRC connection with the eNB. In LTE ACTIVE
state, the network knows the cell to which UE belongs and can transmit/ receive
data from the UE. The LTE IDLE state is a power-conservation state for the UE,
where typically the UE is not transmitting or receiving packets. In LTE IDLE
state, no context about the UE is stored in the eNB. In this state, the location
of the UE is only known at the MME and only at the granularity of a tracking
area (TA) that consists of multiple eNBs. The MME knows the TA in which the
UE last registered and paging is necessary to locate the UE to a cell.
17
1. Background
Transmission BW (MHz)
Sub-frame duration
Sub-carrier spacing
Sampling frequency (MHz)
FFT size
Number of occupied sub-carriers
Normal
CP length (µs)
Extended
1.4
1.92
128
72
3
5
10
15
1.0 ms
15KHz
3.84 7.68 15.36 23.04
256 512 1024 1536
180 300
600
900
4.69×6, 5.21×1
16.6
20
30.72
2048
1200
Table 1.3.6. E-UTRA Numerology
1.3
1.3.1
LTE Technologies
OFDM
In the downlink, OFDM is selected to meet efficiently E-UTRA performance
requirements. With OFDM, it is straightforward to exploit frequency selectivity of the multi-path channel with low complexity receivers. This allows frequency selective in addition to frequency diverse scheduling and one cell reuse of
available bandwidth. Furthermore, due to its frequency domain nature, OFDM
enables flexible bandwidth operation with low complexity. Smart antenna technologies are also easier to support with OFDM, since each sub-carrier becomes
flat faded and the antenna weights can be optimized on a per sub-carrier (or
block of sub-carriers) basis. In addition, OFDM enables broadcast services on
a synchronized Single Frequency Network (SFN) with appropriate cyclic prefix
design. This allows broadcast signals from different cells to combine over the air,
thus significantly increasing the received signal power and supportable data rates
for broadcast services.
To provide great operational flexibility, E-UTRA physical layer specifications
are bandwidth agnostic and designed to accommodate up to 20 MHz system
bandwidth. Table 1.3.6 provides the downlink sub-frame numerology for different
spectrum allocations. Sub-frames with one of two cyclic prefix (CP) durations
18
LTE Technologies 1.3
may be time-domain multiplexed, with the shorter designed for unicast transmission and the longer designed for larger cells or broadcast SFN transmission.
The useful symbol duration is constant across all bandwidths. The 15 kHz subcarrier spacing is large enough to avoid degradation from phase noise and Doppler
(250km/h at 2.6 GHz) with 64QAM modulation.
The downlink reference signal structure for channel estimation, CQI measurement, and cell search/acquisition is shown in Figure 1.3.1. Reference symbols
(RS) are located in the 1st OFDM symbol (1st RS) and 3rd to last OFDM symbol
(2nd RS) of every slot. For FDD, it may be possible to reduce overhead by not
transmitting the 2nd RS for at least low to medium speeds, since adjacent subframes can often be used to improve channel estimation performance. This dual
TDM (or TDM) structure has similar performance to a scattered structure in 0.5
ms sub-frames, and an advantage in that low complexity channel estimation (interpolation) is supported as well as other excellent performance low-complexity
techniques, such as IFFT-based channel estimators. To provide orthogonal signals for multi-antenna implementation, FDM is used for different TX antennas
of the same cell, and CDM is used for different cells.
1.3.2
Multiple antenna techniques
Central to LTE is the concept of multiple antenna techniques - often loosely
referred to as MIMO - which take advantage of spatial diversity in the radio
channel. Multiple antenna techniques are of three main types: diversity, MIMO,
and beamforming. These techniques are used to improve signal robustness and
to increase system capacity and single-user data rates. Each technique has its
own performance benefits and costs.
Figure 1.3.2 illustrates the range of possible antenna techniques from the
simplest to the most complex, indicating how the radio channel is accessed by
19
1. Background
R0
R0
R1
R0
R0
R0
R0
R1
Frequency
Frequency
R0
R1
R1
R1
R0
l=6
l=0
Slot
R1
l=0
l=6
R1
l=6
l=0
Slot
l=0
Slot
(a) Antenna port 0.
Figure 1.3.1.
R1
l=6
Slot
(b) Antenna port 1.
Downlink reference signal structure - normal cyclic prefix, two
transmit antennas
the system’s transmitters and receivers.
Transmit
antennas
The radio channel
Receive
antennas
Transmit
antennas
The radio channel
(a) SISO system.
(b) SIMO system.
(c) MISO system.
(d) MIMO system.
Receive
antennas
Figure 1.3.2. Radio channel access modes
SISO
The most basic radio channel access mode is single input single output (SISO)
in which, only one transmit antenna and one receive antenna are used. This is
the form of communication that has been the default since radio began and is
the baseline to which all the multiple antenna techniques are compared.
20
LTE Technologies 1.3
MISO
Slightly more complex than SISO is multiple input single output (MISO)
mode, which uses two or more transmitters and one receiver. (Figure 1.2(c) shows
only two transmitters and one receiver for simplicity.) MISO is more commonly
referred to as transmit diversity. The same data is sent on both transmitting
antennas but coded such that the receiver can identify each transmitter. Transmit diversity increases the robustness of the signal to fading and can increase
performance in low signal-to-noise ratio (SNR) conditions; however, it does not
increase data rates as such, but rather supports the same data rates using less
power. Transmit diversity can be enhanced with closed loop feedback from the
receiver to indicate the balance of phase and power used for each antenna.
SIMO
Figure 1.2(b) is single input multiple output (SIMO), which - in contrast to
MISO - uses one transmitter and two or more receivers. SIMO is often referred
to as receive diversity. Similar to transmit diversity, it is particularly well suited
for low SNR conditions in which a theoretical gain of 3dB is possible when two
receivers are used. As with transmit diversity, there is no change in the data
rate since only one data stream is transmitted, but coverage at the cell edge is
improved due to the lowering of the usable SNR.
MIMO
Finally the Figure 1.2(d) shows full MIMO, which requires two or more transmitters and two or more receivers. This mode is not just a superposition of SIMO
and MISO since multiple data streams are now transmitted simultaneously in the
same frequency and time, taking full advantage of the different paths in the radio
channel. For a system to be described as MIMO, it must have at least as many
21
1. Background
receivers as there are transmit streams. The number of transmit streams should
not be confused with the number of transmit antennas. Consider the Tx diversity (MISO) case in which two transmitters are present but only one data stream.
Adding receive diversity (SIMO) does not turn this into MIMO, even though there
are now two Tx and two Rx antennas involved. SIMO + MISO 6= MIMO. It is
always possible to have more transmitters than data streams but not the other
way around. If N data streams are transmitted from less than N antennas, the
data cannot be fully descrambled by any number of receivers since overlapping
streams without the addition of spatial diversity just creates interference. However, by spatially separating N streams across at least N antennas, N receivers
will be able to fully reconstruct the original data streams provided the crosstalk
and noise in the radio channel are low enough.
One other crucial factor for MIMO operation is that the transmissions from
each antenna must be uniquely identifiable so that each receiver can determine
what combination of transmissions has been received. This identification is usually done with pilot signals, which use orthogonal patterns for each antenna.
The spatial diversity of the radio channel means that MIMO has the potential
to increase the data rate. The most basic form of MIMO assigns one data stream
to each antenna.
In this form, one data stream is uniquely assigned to one antenna. The channel
then mixes up the two transmissions such that at the receivers, each antenna
sees a combination of each stream. Decoding the received signals is a clever
process in which the receivers, by analyzing the patterns that uniquely identify
each transmitter, determine what combination of each transmit stream is present.
The application of an inverse filter and summing of the received streams recreates
the original data.
The theoretical gains from MIMO are a function of the number of transmit and
22
LTE Technologies 1.3
receive antennas, the radio propagation conditions, the ability of the transmitter
to adapt to the changing conditions, and the SNR. The ideal case is one in which
the paths in the radio channel are completely uncorrelated, almost as if separate,
physically cabled connections with no crosstalk existed between the transmitters
and receivers. Such conditions are almost impossible to achieve in free space, and
with the potential for so many variables, it is neither helpful nor possible to quote
MIMO gains without stating the conditions. The upper limit of MIMO gain in
ideal conditions is more easily defined, and for a 2x2 system with two simultaneous
data streams a doubling of capacity and data rate is possible. MIMO works best
in high SNR conditions with minimal line of sight.
Beamforming
Beamforming uses the same signal processing and antenna techniques as
MIMO but rather than exploit de-correlation in the radio path, beamforming
aims to exploit correlation so that the radiation pattern from the transmitter is
directed towards the receiver. This is done by applying small time delays to a
calibrated phase array of antennas. The effectiveness of beamforming varies with
the number of antennas. With just two antennas little gain is seen, but with
four antennas the gains are more useful. Obtaining the initial antenna timing
calibration and maintaining it in the field are challenge.
Combining beamforming and MIMO
The most advanced form of multiple antenna techniques is probably the combination of beamforming with MIMO. In this mode MIMO techniques could be
used on sets of antennas, each of which comprises a beamforming array. Given
that beamforming with only two antennas has limited gains, the advantage of
combining beamforming and MIMO will not be realized unless there are many
23
1. Background
antennas. This limits the practical use of the technique on cost grounds.
1.3.3
Scheduling
Scheduling controls the allocation of the shared resources among the users and
often is jointed to link adaptation; if this happens we have a channel-dependent
scheduling (Figure 1.3.3). In LTE a part of the scheduler is the rate adaptation,
so it determines the data rate to be used for each link.
Effective channel variations
seen by the base station
Effective channel variations
seen by the base station
User #1
User #1
User #2
Channel quality
Channel quality
User #2
#1
#2
#1
#2
(a) Time domain scheduling.
Time
#1
#2
#1
#2
#1
#2 Frequency
(b) Frequency domain scheduling.
Figure 1.3.3. Channel-dependent scheduling.
LTE has also, in addition to the time domain, access to the frequency domain,
due to the use of OFDM. In other words, scheduling in LTE can take channel
variations into account not only in the time domain, as HSPA, but also in the
frequency domain and this meant that for LTE, scheduling decisions can be taken
as often as once every 1 ms and the granularity in the frequency domain is 180
kHz.
1.4
Handover
In case of a handover the protocol endpoints that are located in the eNodeB
will need to be moved from the Source eNodeB to the Target eNodeB. Then,
it is an option whether the full protocol status of the Source eNodeB is transferred to the Target eNodeB or the protocols are reinitialized after the handover.
24
Handover 1.4
This raises the question for example, whether the HARQ and ARQ window state
is discarded and reset or transferred during the handover. Since, it would be
overly complex and not always feasible to transfer the whole protocol state, it
is an assumption in LTE that the RLC/MAC protocols are reset after a handover. The message sequence diagram of the LTE handover procedure is shown
in Figure 1.4.1.
The figure shows both the control plane messages (black and blue solid arrows)
and the flow of the user plane packets (purple dashed arrows). See also [4] for a
description of the procedure. The UE sends measurement reports to the Source
eNodeB, which may decide on the execution of a handover. The Source eNodeB
requests the preparation of a handover at the Target eNodeB. The Target eNodeB
can perform admission control to check whether the established QoS bearers of
the UE can be accommodated in the target cell.
Next, the Source eNodeB sends the Handover Command to the UE, which
includes all necessary information for the UE to access the target cell. At the
same time the Source eNodeB suspends the RLC/MAC protocols and it may start
to forward the SDUs that have not yet been successfully sent to the UE toward
the Target eNodeB. That is, partially transmitted SDUs at the HARQ/ARQ
layers will be forwarded along with the buffered and not yet transmitted SDUs,
also including all the incoming SDUs from the GW. Whether SDU forwarding is
employed at all by the eNodeB is left as a vendor specific implementation detail.
At the same time, the UE starts to execute the handover to reconnect at
the Target eNodeB. The UE needs to perform a random access procedure on
the RACH (Random Access Channel) in the target cell. The UE also needs to
get uplink time alignment assigned, which means that the eNodeB has to measure on the uplink transmission of the UE (on the RACH) and determine the
timing advance that the UE has to use for its uplink transmissions in order for
25
1. Background
Source
eNB
UE
Target
eNB
MME
S-GW
0. Area Restriction Provided
1. Measurement Control
Packet Data
Legend
Packet Data
L3 Signaling
UL allocation
User Data
2. Measurement Report
L1/L2 Signaling
3. HO decision
Handover Preparation
4. Handover Request
5. Admission Control
6. Handover Request Ack
DL allocation
7. Handover Command
Detach from old cell and
synchronize to new cell
Deliver buffered and in
transit packets to target
eNB
Handover Execution
8. SN Status Tranfer
Data Forwarding
Buffer packets from
Source eNB
9. Syncronization
10. UL Allocation + TA for UE
11. Handover Confirm
12. Path Switch Request
13. User Plane Update Req
End Marker
16. Path Switch Request Ack
17. Relase Resource
Flush DL buffer, continue
delivering in transit packets
Data Forwarding
End Marker
18. Relase Resources
Packet Data
Packet Data
Figure 1.4.1. Message chart of the LTE handover procedure.
26
Handover Completion
14. Switch DL Path
15. User Plane Update Reply
Handover 1.4
the received signals from multiple UEs arrive time synchronized at the eNodeB.
Note that in OFDM [5] it is important to keep an accurate time synchronization
between carriers in order to maintain orthogonality. The UE also needs to get
UL/DL resources scheduled in order to be able to commence user data transmission. These L1/L2 access procedures, i.e., the radio interruption time at a
handover, are estimated to take approximately 30 ms [6].
Next, the UE sends the Handover Complete message in the target cell, which
is used by the Target eNodeB to verify that it is the right UE that is accessing the
target cell. At that point the Target eNodeB can start sending DL data to the
UE. The Target eNodeB sends the path switch command to the GW and finally,
it triggers the release of resources at the Source eNodeB. Note that there can be
a time interval after the path switch when packets both on the direct path and
also on the forwarding path arrive in parallel at the Target eNodeB. This may
give rise to the problem of out of order packet delivery as explained below.
In fact, during the handover procedure there is a time interval when packets
on the direct path (GW - Target eNodeB) and packets on the forwarding path
(GW - Source eNodeB - Target eNodeB) may arrive in parallel at the Target
eNodeB. This gives rise to the potential problem of out of order packet delivery
to the UE, since packets on the direct path typically travel a shorter distance and
arrive earlier to the Target eNodeB than packets on the forwarding path. Note
that out of order packet arrivals may have a bad impact on tcp performance and
may result in a loss of system utilization. When three or more packets arrive out
of order, it results in duplicate acks received at the tcp sender, which will react
with a retransmission and congestion window halving.
The problem mentioned above has been analyzed and solved in [6], where the
solution consists in a packet reordering feature inside the TeNB to avoid the out
of order delivering.
27
1. Background
1.5
Summary
In this chapter we have seen that the main objectives of Long Term Evolution
(LTE) are (1) instantaneous peak data rate 100 Mbps in downlink and 50 Mbps
in uplink within a 20 MHz spectrum allocation, the data rate is scalable with
the spectrum allocation; (2) user throughput improved by factor of 2-3 in uplink
and downlink respectively; reduced control-plane latency; (3) user-plane latency
below 5 ms for small ip packet in an unloaded network; (4) improved spectrum
efficiency by factor of 2-3 in uplink and downlink respectively; (5) co-existence
and interworking with legacy standards; (6) reduced cost.
A detailed description of the handover procedure in LTE is given by a message
chart that includes the control plane messages and the flow of the user plane packets. In the next chapter, we give a description of TCP with particular reference
to LTE.
28
Chapter 2
TCP and LTE
In this chapter we describe the Transmission Control Protocol and his application to LTE.
The most commonly used transport layer protocols are tcp and udp. udp[12]
is a connectionless protocol that does not guarantee delivery of data, it does not
make any error check on the payload. It is used in server client type request-reply
situations and in applications where prompt delivery of data is more important
than accurate delivery, like video streaming.
The Transmission Control Protocol (tcp) is one of the core protocols of the
Internet Protocol Suite. tcp is so central that the entire suite is often referred
to as “tcp/ip”. Whereas ip handles lower-level transmissions from computer
to computer as a message makes its way across the Internet, tcp operates at a
higher level, concerned only with the two end systems, for example a Web browser
and a Web server.
tcp/ip manages the end-to-end reliability in the Transport layer. The topic
of this section is the Transmission Control Protocol (tcp) which is the predominant transport protocol of the Internet today carrying more than 80% of the
total traffic volume. tcp is used by a range of different applications such as
web-traffic (www), file transfer (ftp, ssh), e-mail and even streaming media ap-
29
2. TCP and LTE
plication with “almost real-time” constraints such as, e.g., Internet radio. Among
its management tasks, tcp controls message size, the rate at which messages are
exchanged, and network traffic congestion.
2.1
Importance of TCP
tcp provides a communication service at an intermediate level between an
application program and the Internet Protocol (ip). That is, when an application
program desires to send a large chunk of data across the Internet using ip, instead
of breaking the data into ip-sized pieces and issuing a series of ip requests, the
software can issue a single request to tcp and let tcp handle the ip details.
ip works by exchanging pieces of information called packets. A packet is a
sequence of bytes and consists of a header followed by a body. The header describes the packet’s destination and, optionally, the routers to use for forwarding
- generally in the right direction - until it arrives at its final destination. The
body contains data which ip is transmitting. When ip is transmitting data on
behalf of tcp, the content of the ip packet body is tcp payload.
Due to network congestion, traffic load balancing, or other unpredictable network behavior, ip packets can be lost or delivered out of order. tcp detects these
problems, requests retransmission of lost packets, rearranges out-of-order packets,
and even helps minimize network congestion to reduce the occurrence of other
problems. Once the tcp receiver has finally reassembled a perfect copy of data
originally transmitted, it passes that datagram to the application program. Thus,
tcp abstracts the application’s communication from the underlying networking
details.
30
TCP key features 2.2
2.2
TCP key features
tcp[11] is a reliable transport protocol with connection procedure that provides a in-sequence data delivering from a sender to a receiver. Some of the
tcp properties may be desired by certain applications.
The tcp reliability is achieved using ARQ mechanism based on positive acknowledgments. tcp protocol provides transparent segmentation and reassembly
of user data and handles both data flow and congestion. tcp packets are cumulatively acknowledged when they arrive in sequence; out of sequence packets cause
the generation of duplicate acknowledgments. For each segment transmission, a
retransmission timer is started; retransmission timers are continuously updated
on a weighted average of previous round trip time (rtt) measurements, i.e. the
time it takes from the transmission of a segment until the paired acknowledgment
is received. tcp sender detects a loss either when multiple duplicate acknowledgments (3 is the default value) arrive, this imply that the packet after the last
acknowledged one has been lost, or when a retransmission timeout (RTO) expires.
The RTO value is calculated dynamically based on rtt measurements. Its accuracy is critical, delayed timeouts slow down recovery or redundant retransmissions
can occur due to mistakes in the RTO evaluation.
2.2.1
Flow control
For the flow control tcp uses a sliding window mechanism. It limits the
amount of data that can be sent without waiting for acknowledgement (ack)
from the receiver and when the window is constant, this results in the so called
ack-clock.
31
2. TCP and LTE
Sliding window
The window size is the amount of allowed outstanding data, as illustrated in
the Figure 2.2.1. The sliding window, only after the ack for an “in sequence”
packet permits the sender transmission of one further packet, and the window of
outstanding data slides forward one packet.
0
9
1
9
0
1
8
2
8
2
7
3
7
3
6
5
4
(a) Before the ack of packet 2.
6
5
4
(b) After the ack of packet 2.
Figure 2.2.1. Sliding window mechanism.
The initial value of the congestion window can be chosen between one and
four segments, see [17]. The receiver maintains an advertised windows (rwnd )
which indicates the maximum number of bytes its buffer can accept. The value
of the rwnd is sent back to the sender together with each segment going back. At
any moment, the amount of outstanding data (wnd ) is limited by the following
rule
wnd = min (rwnd, cwnd) .
(2.2.1)
ACK-clock
When tcp’s congestion window is kept constant, the sender transmits one new
packet for each received ack. This is referred to as the ack-clock. The number of
32
TCP Applicability 2.3
packets inside the network (be they data packets or acks) is kept constant. The
ack-clock is based on the idea that by controlling the number of packets inside
the network, we can control the load of the network. The average transmission
rate is one window of data per rtt. In tcp congestion control, the control signal
is the window size, and the actual sending rate is controlled only indirectly by
adjusting the window. This can be seen as a cascaded control system, where
the ack-clock is a fast inner-loop, operating on a per-packet timescale, which is
combined with an outer-loop that adjusts the window size, and operates on an
per-rtt timescale.
2.3
TCP Applicability
tcp is used extensively by many of the Internet’s most popular application
protocols and resulting applications, however, because tcp is optimized for accurate delivery rather than timely delivery, tcp sometimes incurs relatively long
delays (in the order of seconds) while waiting for out-of-order messages or retransmissions of lost messages, and it is not particularly suitable for real-time
applications such as Voice over ip. For such applications, protocols like the Realtime Transport Protocol (rtp) running over the User Datagram Protocol (udp)
are usually recommended instead.
tcp is a reliable stream delivery service that guarantees delivery of a data
stream sent from one host to another without duplication or losing data. Since
packet transfer is not reliable, a technique known as positive acknowledgment
with retransmission is used to guarantee reliability of packet transfers. This
fundamental technique requires the receiver to respond with an acknowledgment
message as it receives the data. The sender keeps a record of each packet it sends,
and waits for acknowledgment before sending the next packet. The sender also
33
2. TCP and LTE
keeps a timer starting from the moment the packet was sent, and retransmits a
packet if the timer expires (rto). The timer is needed in case a packet gets lost
or corrupt.
tcp consists of a set of rules: for the protocol, that are used with the Internet
Protocol, and for the ip, to send data “in a form of message units” between computers over the Internet. At the same time that the ip takes care of handling the
actual delivery of the data, the tcp takes care of keeping track of the individual
units of data “packets” (or more accurately, “segments1 ”) that a message is divided into for efficient routing through the net. For example, when a html file is
sent from a Web server, the tcp program layer of that server takes the file as a
stream of bytes and divides it into segments, numbers the segments, and then forwards them individually to the ip program layer. The ip program layer then turns
each tcp segment into an ip packet by adding a header which includes (among
other things) the destination ip address. Even though every packet has the same
destination ip address, they can get routed differently through the network. When
the client program in your computer gets them, the tcp stack (implementation)
reassembles the individual segments and ensures they are correctly ordered as it
streams them to an application.
Over wireless networks, where losses are mainly caused by mobility and intermittent poor channel conditions [14], poor tcp performance is experienced.
Common tcp version misinterpret delays as congestion indications, sometimes
useless retransmissions are experienced and much time is wasted during the slow
start and the congestion avoidance phases. Some solutions have been found by
1
tcp accepts data from a data stream, segments it into chunks, and adds a tcp header
creating a tcp segment. The tcp segment is then encapsulated into an IP packet. A tcp segment is “the packet of information that tcp uses to exchange data with its peers.” Note that
the term tcp packet is now used interchangeably with the term tcp segment, although in the
original RFC segment usually referred to the tcp unit of data
34
TCP Applicability 2.3
introducing changes in the tcp paradigm. Westwood tcp for example tries to improve the classic tcp behavior to keep it applicable over both wireless and wired
networks. tcp selective acknowledgments (sack)[16] is proposed to alleviate the
inefficiency of tcp in handling multiple drops in a single data window.
2.3.1
Congestion control
tcp was initially designed to be used in wired networks which has extremely
low transmission losses (BERs lower than 10−10 ), it assumes that all losses are
due to congestion. Therefore, when tcp detects packet losses, it reacts both
retransmitting the lost packet and reducing the transmission rate. In this way it
allows a reduction of the router queues length. Afterwards, it gradually increases
the transmission rate to probe the network’s capacity.
The purpose of slow start and congestion avoidance is to prevent the congestion from occurring varying the transmission rate. For this reason tcp uses
a congestion window (cwnd), which represents the number of segments that the
network might deliver without congestion occurring.
Today all tcp implementations are provided of congestion control algorithm
such slow start, congestion avoidance, fast retransmit and fast recovery[13].
Slow start
In the slow start phase, the congestion window is increased by one segment
for each acknowledgment received. This phase is used both at the beginning of a
new connections and after retransmissions due to RTO expiring. During the slow
start phase cwnd experiences an exponential increase until a timeout occurs or a
threshold value (ssthresh) is reached. In the first case the cwnd will be set to one
packet, instead in the second case, the slow start phase ends and the congestion
avoidance phase starts.
35
2. TCP and LTE
In this state, cwnd is increased by one packet for each non-duplicate ack. The
effect is that for each received ack, two new packets are transmitted. This implies
that the congestion window, and also the sending rate, increases exponentially,
doubling once per rtt.
The motivation for the slow start state is that when a new flow enters the
network, and there is a bottleneck link along the path, then the old flows sharing
that link need some time to react and slow down before there is room for the new
flow to send at full speed.
Congestion avoidance
In the congestion avoidance state, cwnd is increased by one packet per rtt (if
cwnd reaches the maximum value, it stays there). This corresponds to a linear
increase in the sending rate. On timeout, tcp enters the exponential back-off
state2 , and on three duplicate acks, it enters the fast recovery state.
The motivation for this congestion avoidance mechanism is that since tcp does
not know the available capacity, it has to probe the network to see at how high
a rate data can get through.
The main difference between slow start and congestion avoidance algorithm is
the way to update the cwnd value (the first it is exponential while for the second
it is linear) as shown in
cwnd =
2

 cwnd + 1,
Slow Start;
 cwnd + M SS · M SS , Congestion avoidance.
cwnd
(2.3.1)
In the exponential back-off mode, once timeout occurred, tcp takes the following actions:
(1) The lost packet is retransmitted. (2) The state variables are updated as follow: cwnd is
set at 1 packet and ssthresh is set to the value of cwnd/2. (3) The rto value is doubled.
The motivation for the exponential back-off mechanism is to avoid congestion collapse due to
timeout, that is a sign of severe network congestion.
36
TCP Applicability 2.3
Timeout
ce
idan
vo
nA
stio
20
gre
Con
SStresh
16
e
anc
id
Avo
tion
s
gre
Con
12
New SStresh
w
Sta
rt
8
4
Slo
Congestion Window (Segments)
24
w
Slo
0
0
1
2
3
4
5
6
7
8
9
10
St
11
t
ar
12
13
14
15
16
17
Round Trip Times
Figure 2.3.1. tcp slow start and congestion avoidance.
where M SS is the Maximum Segment Size and is expressed in byte.
tcp opens the congestion window quickly to reach the capacity limit of the
link as rapid as possible, the congestion avoidance algorithm is conceived to
transmit at a safe operating point and increase the congestion window slowly to
probe the network for more bandwidth becoming available.
As shown in Figure 2.3.1, if a timeout occurs, the ssthresh is updated to a
half of the current window size according to
sstresh =
min (rwnd, cwnd)
.
2
(2.3.2)
The congestion window is reduced to one MSS (Maximum Segment Size), and
the slow start phase in entered again.
Fast recovery
A different behavior is assumed by tcp in case of three duplicate acks in a
row. This can happen when the receiver detects an out-of-order packet and sends
back to the sender the sequence number of the next expected packet. Therefore,
37
2. TCP and LTE
tcp performs a direct retransmission of the missing segment after the reception
of the third duplicate-ack even if the retransmission timer has not expired yet.
The ssthresh is set to the same value as in the case of timeout (2.3.2). After
the retransmission, fast recovery is performed until all lost data are recovered.
The congestion window is set three segments more than ssthresh (to take into
account the number of segments that have already left the network which are
buffered by the receiver). For each additional duplicate ack received, the cwnd is
incremented by one packet according to (2.3.3)
cwnd = cwnd + 1.
(2.3.3)
The fast recovery phase ends when one non-duplicate acknowledgment is detected.
The congestion window is then set ssthresh and a congestion avoidance phase
starts (Figure 2.3.2). If no ack for the retransmitted packet is received within
the rto interval, tcp enters the exponential back-off state. The main problem
in this behavior is that if more than one packet is lost within the same window,
the original fast recovery procedure of tcp Reno is limited in that it can recover
only one packet per rtt.
With fast retransmit and fast recovery, tcp avoids unnecessary slow starts
phases due to minor congestion incidents duplicate acks are used as indicators
of some kind of network congestion).
The motivation for the fast recovery mechanism is that the reception of duplicate acks indicates that the network is able to deliver new data to the receiver, on
the other hand, the loss of a packet also indicates that the network is on the border of congestion. For these reasons halving the cwnd also implies that tcp will
stay silent for about half an rtt, waiting for acks that reduce the number of
outstanding packets.
38
TCP Applicability 2.3
3rd duplicate Ack
ce
idan
Avo
n
o
sti
gre
20
New Ack
Con
SStresh
16
(fast retransmit)
e
anc
Fast Recovery
12
Con
sti
gre
void
on A
New SStresh
w
Sta
rt
8
4
Slo
Congestion Window (Segments)
24
0
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
Round Trip Times
Figure 2.3.2. tcp fast retransmit and fast recovery.
2.3.2
Explicit Congestion Notification (ECN)
ECN is an extension to the Internet Protocol and is defined in [19]. ECN
allows end-to-end notification of network congestion without dropping packets.
It is an optional feature and is only used when both endpoints signal that they
want to use it (ECN negotiation occur during the three way handshaking).
Traditionally, tcp/ip networks signal congestion by dropping packets. When
ECN is successfully negotiated, an ECN-aware router may set one bit in the
ip header instead of dropping a packet in order to signal the beginning of congestion. The receiver of the packet echoes the congestion indication to the sender,
which must react as though a packet drop were detected. Obviously this policy
is useful only if it is used in a network where the nodes have an Active Queue
Management (AQM) algorithm.
ECN uses two bits in the ipv4 Type Of Service (TOS) field in the ip header.
These two bits can be used to encode one of the values ECN-unaware transport,
ECN-aware transport or congestion experienced as shown in Figure 2.3.3.
In addition to the two ECN bits in the ip header, tcp uses two flags in the
39
2. TCP and LTE
0
1
2
3
4
5
6
7
Differentiated Service CodePoint
ECN
ECT
CE
0
0
ECN-unaware
0
1
ECN-aware
1
0
ECN-aware
1
1
Congestion Experienced
Figure 2.3.3. The TOS field in ip header and ECN signaling.
tcp header to signal the sender to reduce the amount of information it sends.
These are the ECN-echo (ECE) and Congestion Window Reduced (CWR) bits
(Figure 2.3.4).
0
1
2
Header Length
3
4
5
6
Reserved
7
8
9
10
11
12
13
14
15
CWR
ECE
URG
ACK
PSH
RST
SYN
FIN
Figure 2.3.4. Bytes 13 and 14 of the tcp header.
Once the receiver get a tcp segment with the Congestion Experienced codepoint, the tcp receiver sends an acknowledgement with the ECN-echo (ECE)
flag set. The ECN-echo bit indicates congestion to the sender, which reduces its
congestion window as for a packet drop. It then acknowledges the congestion
indication by sending a segment with the Congestion Window Reduced (CWR)
codepoint.
2.4
Evaluation of TCP Performance
tcp performance can be evaluated in different ways depending on the context,
such as the system used and the application carried over the tcp connection. The
key parameters to evaluate tcp Performance are [18]:
40
Evaluation of TCP Performance 2.4
Effective throughput also called effective bandwidth. It is the data transmission rate of the application in bit/s. The effective throughput is more
significant than the communication channel throughput, since it takes into
account the effective amount of data delivered at the tcp layer.
Throughput variation The throughput variation over a given timescale, depending on the application, is very important to assess end-to-end performance of certain application.
File transfer time This parameter is related to the effective throughput.
Round trip time (rtt) The time between the transmission of a segment and
the reception of the paired acknowledgment. This parameter includes the
delay in the intermediate network nodes and is an estimation of the network
latency. It can limit the effective throughput.
Delay variation or jitter in other words, the variation of the rtt. This variation can have a direct impact on the appearance of triple duplicate ack or
timeout events (sometimes spurious) and it results in throughput limitation
unappropriate use of resources, meant as a scarce use of them.
Fairness is an important characteristic, especially when tcp applications are
carried over wireless systems. Unfair resource allocation can have drastic
effects on the effective throughput.
Resource consumption The amount of resources (e.g., buffer space, transmission power) needed to achieve a given effective throughput.
41
2. TCP and LTE
2.5
Bottleneck in the core network
The LTE system supports QoS mechanisms over radio and in the transport
network. However no flow control mechanism are supported between the base
station and the gateway. This implies that packet buffering/dropping for elastic
services, like tcp, could occur during congestion in any node of the core network
(e.g., in the base station or a transport network router).
The 3GPP standard assume that the end-to-end tcp protocol will govern
the congestion control and adapt to the varying network conditions and handle
packet loss.
As already depicted in the Section 1.1.2 the user plane connection crosses
the core network exploiting an ip tunnel, between serving gateway and eNodeB,
based on GPRS Tunneling Protocol (GTP).
Due to this tunneling protocol any explicit congestion signaling is not allowed
even if the user connection is able to handle it. The ip header handled by the
network router is belonging to the tunnel application rather than the user and
the signaling will be trashed once the packet has reached the end of the tunnel.
2.5.1
GPRS Tunneling Protocol (GTP)
GPRS Tunneling Protocol[15] is an ip based protocol created to be used within
GSM and UMTS networks as a solution to manage the mobility of the MEs. It
can be used with udp or tcp but generally the first solution is adopted.
We can distinguish two separate protocols for user data and control data:
GTP-C is used inside the core network for signalling between inner Support
Nodes. This allows the network to activate a session on behalf of users
(PDP context activation), to deactivate the same session, to adjust Quality
of Service parameters or to update a session for a subscriber.
42
Bottleneck in the core network 2.5
GTP-U is used for carrying user data within mobile core network and between
the Radio Access Network and the core network. The user data transported
can be packets in any of ipv4, ipv6 or other formats.
Let focus our attention on the user plane. GTP-U is a relatively simple
ip based tunneling protocol which permits many tunnels between each set of end
points. When used in LTE, as in the UMTS network, each subscriber will have one
or more tunnel, one for each PDP context they have active plus, possibly separate
tunnels for specific connections with different Quality of Service requirements.
User
Application
Carry the effective payload
IP (user)
It will be the IP header once the packet leave the tunnel connection
GTP
Header used to distinguish and manage different tunnel
UDP
Connectionless transport protocol for GTP
Header used to manage the tunnel application
IP
Depending on the Network Architecture
Layer 2
Figure 2.5.1. GTP-U Protocol Stack.
Figure 2.5.1 show the GTP protocol stack for user data. It shows that a second
ip header is added to manage the tunnel connection while the user ip header is
moved inside the payload of the GTP protocol that works like an application
protocol.
In Figure 2.5.2 we can see the GTP header structure and below there is a
quick description of all field function:
Version is a 3-bit field that indicate the GTP protocol version. For GTPv1, this
has a value of 1.
43
2. TCP and LTE
0
Bits 0-2
3
4
5
6
7
8-15
Version
Protpcol
Type
Reserved
Next
Extension
Header
Sequence
Number
N-PDU
Number
Type
16-23
24-31
Total Lentht
TEID
32
Sequence Number
64
N-PDU Number
Next Extension
Header
Figure 2.5.2. GTPv1 header (default and optional fields).
Protocol Type differentiates GTP (value 1) from GTP’ (value 0).
Reserved this field must be 0.
Extension Header (E) is set if there is an Extension Header optional field.
Sequence Number (S) is set if there is a Sequence Number optional field.
N-PDU number (PN) is set if there is a N-PDU number optional field.
Type differentiates the packet type
Total Length is a 16-bit field that states the length of the packet being encapsulated by GTP (not including the GTP header itself, but including the
optional fields).
Tunnel Endpoint Identifier (TEID) used to multiplex different connections
in the same GTP tunnel.
We can find also some optional fields available only if any of the E, S, or PN bits
are on. These fields are:
Sequence Number must be interpreted only if the S bit is set to 1.
N-PDU number must be interpreted only if the PN bit is high
44
Summary 2.6
Next Extension Header must be interpreted only if the E bit is on
The separate tunnels are identified by a Tunnel Endpoint Identifier (TEID) in
the GTP-U messages, which should be a dynamically allocated random number.
If this random number is of cryptographic quality, then it will provide a measure of
security against certain attacks. Even so, the requirement of the 3GPP standard
is that all GTP traffic, including user data should be sent within secure private
networks, not directly connected to the Internet.
2.6
Summary
In this chapter, an accurate description of the tcp protocol is given by focusing
on the flow control and congestion control. Particular attention is given to the
GTP tunnel protocol used by LTE during the handover procedure, where the
ip tunnel between the GW and the radio base station associated with the moving
mobile is switched in the GW towards the target base station when the terminal
arrives there. In the next chapter, we see how to manage TCP during LTE
handover.
45
Chapter 3
Management of LTE handover
In this chapter we investigate the problems that could occur during the handover procedure in LTE network and we will show how to solve or mitigate them.
3.1
Forwarding procedure
0.25
0.2
RTT [s]
0.15
0.1
0.05
0
1
1.5
2
2.5
Time[s]
3
3.5
4
Figure 3.1.1. rtt behavior during the LTE handover
In Chapter 1 we have shown that, during the handover procedure, the user
data are forwarded from the SeNB to the TeNB. This mechanism has negative
effects for the in the rtt , because the forwarded packet experiments an dramatic
increase of delay, especially when the network is congested.
47
3. Management of LTE handover
Figure 3.1.1 shows the rtt values for each sent packet. It is a result of a
simulation performed by our LTE simulator in which are present 2 mobiles on
SeNB, 2 mobiles on TeNB with ftp traffic. One of the mobiles connected to the
SeNB is moving towards the TeNB following the procedure stated in Figure 1.4.1.
As we can see, the forwarded packets have an rtt higher then the other. This
could carry an useless waste of resources because if the rtt increases a lot, what
could happen is that the cwnd could halve its value, or at worst set its value to 1
MSS.
Next, we examine both cwnd halving and timeout occurrence, compared with
the optimal case, where cwnd does’t suffer of these problems.
CWND
CWNDMAX
al
Ide
g
n
lvi
Handover
CWNDHO
se
ca
ow
ha
ind
W
SStresh
Timeout
SStreshHO
RTT1
RTT
RTT2
Figure 3.1.2. cwnd during handover in case of optimal case, window halving
and timeout
Figure 3.1.2 shows the possible behavior that cwnd assumes during the handover. We assume that the cwnd is in the steady state and we focus on the time
interval between rtt1 and rtt2 . During this interval we want to quantify the
possible throughput loss caused by the handover.
We distinguish three possible situations to quantify such a loss: an optimal
48
Forwarding procedure 3.1
case, a halving case, and a timeout occurrence case. We analyze them in the
sequel.
3.1.1
Optimal case
In the optimal case the congestion window keeps on following the steady-state
course independently from the handover event. The amount of data delivered
(DI ) during the period of our interest is
DI =
rtt
X2
cwnd
MAX
X
cwnd(i) =
i=rtt1
i=
i=SSthresh
cwndX
MAX /2
cwndMAX
+1 +
i=
2
i=0
3 cwndMAX cwndMAX
= ·
+1
2
2
2
cwndMAX
·
=
2
(3.1.1)
in which cwndMAX is the maximum value assumed by the cwnd in the steady-state
cwndMAX
phase and SSthresh =
according with the tcp standard.
2
3.1.2
Window halving case
This event is due to an out-of-order delivering. That can happen if the router
buffer exceeds the maximum length during the packet forwarding.
The difference of throughput between the optimal case and this case is shown
in Figure 3.1.3.
We need to subtract the patterned area (U nsentW H ) in Figure 3.1.3 to (3.1.1)
to obtain the data delivered:
DW H = DI − U nsentW H = DI −
cwnd
MAX
X
(cwndHO − SSthreshHO ) =
i=cwndHO
(3.1.2)
cwndHO
= DI −
· (cwndMAX − cwndHO + 1)
2
where cwndHO is the value assumed by the cwnd at the handover starting time
49
3. Management of LTE handover
CWNDMAX
CWND
Handover
CWNDHO
UnsentWH
SStresh
SStreshHO
RTT1
RTT
RTT2
Figure 3.1.3. Difference between optimal case and window halving
and SSthreshHO =
cwndHO
.
2
By using such a model, the allowed values for cwndHO are those that belong
to the [SSthresh, cwndMAX ] interval that corresponds to all real possible values.
3.1.3
Timeout occurrence case
This is the worst event that can occur during the handover. In practice, it is a
consequence of spurious timeout due to the high jitter increase even exacerbated
by the data forwarding.
Figure 3.1.4 shows that in this case we have more loss than any other case. The
unsent amount of data can be divided in two different contributions: U nsentW H
due to the cwnd halving, and U nsentSS due to the slow-start phase.
50
Forwarding procedure 3.1
CWNDMAX
CWND
Handover
CWNDHO
UnsentWH
SStresh
k
SStreshHO
UnsentSS
RTT1
RTT
RTT2
Figure 3.1.4. Difference between optimal case and timeout verification
Then the throughput can be written as
DSS = DI − U nsentW H − U nsentSS =
= DW H − k · (cwndMAX − cwndHO + 1) −
k−1 X
cwndHO
i=0
2
i
−k+i−2
(3.1.3)
The U nsentSS quantity depends on k, which represent the total number of rtt in
which the slow-start phase is on duty.
The SSthreshHO belongs to the interval stated in the (3.1.4)
SSthreshHO ∈
SSthresh
cwndMAX
, SSthresh where SSthresh =
(3.1.4)
2
2
and k can be expressed as
k = dlog2 SSthreshHO e + 1.
(3.1.5)
The cwndHO values must belong to the [SSthresh, cwndMAX − 2k + 2] interval to use this model. That interval does not correspond to all possible values
51
3. Management of LTE handover
100
Normalized Throughput [%]
95
90
85
80
75
70
65
60
Ideal
WH
SS
55
50
40
45
50
55
60
65
70
CWNDHO [Packets]
75
80
(a) cwndMAX = 80.
100
Normalized Throughput [%]
95
90
85
80
75
70
65
60
Ideal
WH
SS
55
50
70
80
90
100
CWNDHO [Packets]
110
120
(b) cwndMAX = 128.
100
Normalized Throughput [%]
95
90
85
80
75
70
65
60
Ideal
WH
SS
55
50
140
160
180
200
CWNDHO [Packets]
220
240
(c) cwndMAX = 256.
Figure 3.1.5. Normalized throughput during the time interval (rtt1 , rtt2 )
considering optimal case, window halving and timeout for different values of
cwndMAX
52
Forwarding procedure 3.1
(2k − 2 other possible values are missing), but we have to consider to extend
our observation period to the next rtt and take into account further throughput
differences.
In Figure 3.1.5 we can see the normalized throughput for all the cases. If we
vary the cwndMAX we can see that the performance of the window halving case is
almost the same for all the cases. On the contrary, for the timeout case we can
say that an cwndMAX increasing leads to a better performance, which is essentially
due to the incidence of the slow start phase (k increases as log2 cwndMAX ).
100
Normalized Throughput [%]
95
90
85
80
75
70
65
Ideal
WH
SS
60
55
40
50
60
70
80
90
CWNDMAX [Packets]
100
110
120
Figure 3.1.6. Different averages of the normalized throughput on varying the
cwndMAX parameter
Figure 3.1.6 shows two different kind of average that could be done from the
model described above. The continuous lines represent the average made from
the normalized throughput on varying the cwndMAX , the dashed lines assume that
in average the handover occurs in the middle of interval (SSthresh, cwndMAX ).
The basic idea is to find a way to avoid the packet forwarding and then prevent
the rtt increasing. A possible way to do that could be to modify the path switch
time or to predict the handover with a good accuracy to the cwnd halving and
timeout occurrence.
53
3. Management of LTE handover
3.2
Two improving solutions
A first solution presented here is to modify the message chart of the standard,
shown in Figure 1.4.1. We discuss next the details.
3.2.1
Advancing the path switch command
Source
eNB
UE
Target
eNB
MME
S-GW
HO decision
Legend
L3 Signaling
Handover Request
Admission Control
User Data
Handover Request Ack
L1/L2 Signaling
DL allocation
Handover Command
Path Switch Request
User Plane Update Req
End Marker
Switch DL Path
Figure 3.2.1. Advancing the path switch command in the handover message
chart
The first proposal is to send the path switch command to the SAE gateway
immediately after the handover command to reduce the amount of forwarded
data (Figure 3.2.1).
A modification is to send the path switch command to the SAE gateway
immediately after the handover command to reduce the amount of forwarded
data.
In Figure 3.2.1 is summarized how the first part of the message chart should be
by this modification. This solution should decrease the amount of user data that
needs to be forwarded carrying to better performance in terms of rtt variation.
54
Two improving solutions 3.2
3.2.2
Predicting the handover occurrence
The second proposal is using a handover prediction mechanism. Even if in the
LTE standard this feature is not used yet, it seems to be an appealing solution
for a 3G LTE system.
The basic idea of this solution concerns a prediction of the handover event to
allow the generation of the path switch command before the handover command.
This algorithm allow the emptying of the ME pipe before the handover command
and avoid the data forwarding procedure, as shown in Figure 3.2.2.
Source
eNB
UE
Target
eNB
MME
S-GW
1. Measurement Control
Packet Data
Legend
UL allocation
L3 Signaling
2. Measurement Report
User Data
HO prediction
L1/L2 Signaling
Handover Request
Admission Control
Necessary Prediction time interval
Path Switch Request
User Plane Update Req
End Marker
Switch DL Path
Handover Request Ack
DL allocation
Handover Command
Figure 3.2.2. Modification to the handover message chart to predict the handover.
In this case the modification in the message chart involves that the path switch
request should be sent to the MME before the handover (Figure 3.2.2). In that
way the amount of user data that need to be forwarded should decrease, carrying
to better performance in terms of rtt variation.
For this solution, as well as for the one depicted in the Section 3.2.1, a procedure is needed for resource allocation on the Target eNB after a request made
55
3. Management of LTE handover
by the network rather than the mobile.
To examine the behavior of tcp protocol when LTE handover occurs and to
validate all this solution proposal we have used the ns-2 network simulator. In
the next section a description of the simulator is given.
3.3
Validation of the solutions
Here we check the benefits of the solutions to improve TCP performance
during the handover, as proposed in 3.2.
Table 3.3.1 shows the key parameters used for the simulations. We considered
the case in which the ME making the handover is going towards a TeNB that is
more congested than the SeNB.
Parameter
Total simulation time
Nodes attached to SeNB
Nodes attached to TeNB
Router queue size
cwndMAX
Handover Time
Table 3.3.1.
Value
5s
1
5
90
44 ' 64 KB
2÷3s
Simulations parameters about the solution proposed in Section
3.2.
Figure 3.3.1 shows the congestion window update and the rtt experienced by
all the arrived packets in all the cases: (a) and (b) describe the case of standard
LTE handover procedure in which a spurious timeout occur due to rtt increasing,
(c) and (d) depict the application of the fast path switch procedure which is not
able to avoid the spurious timeout event while better performance is achieved
with the handover prediction (Figure 3.3.1(e) and (f)) that avoids the spurious
timeout and carries only to the window halving due the packet dropping by the
56
0.2
45
0.18
40
0.16
35
0.14
25
0.12
RTT [s]
30
0.1
20
0.08
15
0.06
10
0.04
5
0.02
0
0
0.5
1
1.5
2
2.5
Time[s]
Handover
50
Handover
CWND [packets]
Validation of the solutions 3.3
3
3.5
4
4.5
0
5
0
0.2
45
0.18
40
0.16
35
0.14
30
0.12
RTT [s]
0.08
15
0.06
10
0.04
5
0.02
0
0.5
1
1.5
2
2.5
Time[s]
3
3.5
4
4.5
0
5
0
0.2
45
0.18
40
0.16
35
0.14
RTT [s]
0.08
0.06
10
0.04
5
0.02
0.5
1
1.5
2
1
1.5
0.1
15
0
0.5
0.12
20
0
2.5
Time[s]
3
3.5
4
(e) cwnd handover prediction case.
3
3.5
4
4.5
5
2
2.5
Time[s]
3
3.5
4
4.5
5
4.5
5
Handover
50
25
2.5
Time[s]
(d) rtt fast path switch case.
Handover
CWND [packets]
(c) cwnd fast path switch case.
30
2
0.1
20
0
1.5
Handover
50
25
1
(b) rtt standard procedure case.
Handover
CWND [packets]
(a) cwnd standard procedure case.
0.5
4.5
5
0
0
0.5
1
1.5
2
2.5
Time[s]
3
3.5
4
(f) rtt handover prediction case.
Figure 3.3.1. cwnd and rtt of the reference ME during the handover.
57
3. Management of LTE handover
2900
Direct OLD
Forwarding
Direct NEW
2800
Handover
Sequence Number
2850
2750
2700
2650
2600
2.6
2.7
2.8
2.9
Time[s]
3
3.1
3.2
2.6
2.7
(a) Standard procedure case.
2400
Direct OLD
Forwarding
Direct NEW
2300
Handover
Sequence Number
2350
2250
2200
2150
2100
2.1
2.2
2.3
2.4
2.5
Time[s]
(b) Fast path switch case.
Received Sequence Number
2300
Direct OLD
Forwarding
Direct NEW
Handover
Sequence Number
2250
2200
2150
2100
2050
2000
2
2.1
2.2
2.3
Time[s]
2.4
2.5
2.6
(c) Handover prediction case
Figure 3.3.2. Reference ME received sequence number.
58
Summary 3.4
router buffer.
In Figure 3.3.2 we can see that between standard handover procedure and
fast path switch procedure there is a negligible difference in terms of amount of
data forwarded (green line in the Figure 3.3.2(a) and (b)), this is essentially due
to the fact that outstanding data always corresponds to the window, intended as
in (2.2.1), and at the handover time such an amount of data has been already
pushed in the network. On the contrary, there are no forwarded data in the
handover prediction procedure (Figure 3.3.2(c)) and that is the main reason that
carrying to better performance.
3.4
Summary
In this chapter the main issues of the LTE handover procedure are investigated. After a study regarding the possible tcp behavior, two solutions proposed
and validated by simulation in ns-2. These solutions avoids the data forwarding
during the LTE handover, and thus they improve the performance of the tcp protocol as seen by the end-user.
59
Chapter 4
LTE simulator: an overview
In this chapter a brief description of the ns-2 LTE simulator and the nsMiracle framework is given to show how we implemented our LTE simulator. This
simulator is the first one developed to evaluate LTE performance in ns2. We use
it to validate the solution proposed in Chapter 3 and the model proposed in the
next chapter.
4.1
LTE simulator through ns-2 and nsMiracle
ns began as a variant of the REAL network simulator in 1989 and has evolved
substantially over the past few years. In 1995 ns development was supported
by DARPA through the VINT project [27] at LBL, Xerox PARC, UCB, and
USC/ISI. Currently ns development is support through DARPA with SAMAN
and through NSF with CONSER, both in collaboration with other researchers
including ACIRI. ns has always included substantial contributions from other
researchers, including wireless code from the UCB Daedelus and CMU Monarch
projects and Sun Microsystems. We used the version 2 of ns, called ns-2.
ns-2 is a discrete event simulator targeted at networking research that provides
substantial support for simulation of tcp, routing, and multicast protocols over
wired and wireless (local and satellite) networks. ns-2 is written in C++ and an
61
4. LTE simulator: an overview
Object oriented version of Tcl called OTcl.
The tools included in the simulator suite are: (1) the network animator (nam)
that is a graphical visualizer showing network topology, packet flows, queued and
dropped packets at buffers and (2) Xgraph, which is another tool that allow to
graphs simulation results.
The scheduler is the most important component ns-2. It is the core of the
simulator because it handles all the events that occur during the simulation and
keep trace of them in the trace file.
ns-2 does not currently support multiple radio interfaces and does not have
flexible tools for the cross-layer control of communication systems. Moreover,
it is not possible to reuse the existing implementations of the wireless channel
because they do not work for LTE radio interfaces (they are all ad-hoc solution
to the problem which are implemented for). Therefore, we have implemented our
LTE simulator by using nsMiracle.
.....
.....
Connector
Module
Layer N-1
.................
NodeCore
Module
Module
Layer N
PlugIn
Module
Module
.....
Module
.....
Layer N-2
PlugIn
Module
Layer N+1
Figure 4.1.1. One node of nsMIracle.
The framework Multi InteRfAce Cross Layer Extension for ns-2 (nsMiracle)
[28] is conceived as a set of dynamic libraries which are loaded to add support for
multi-technology and cross-layering. It exploits also a patch that facilitates the
use of dynamic libraries in ns-2 [29], which allow to load in the simulator only
62
LTE simulator through ns-2 and nsMiracle 4.1
the necessary modules and makes the simulation faster.
This extension allow an easy way to customize a node as shown in Figure 4.1.1.
nsMiracle also facilitates the connection of different modules and uniforms the
plug-in procedure of multiple protocol into the same node.
NodeCore in nsMiracle is a structure which all Modules connected with. Its
role are: (1) enable communication among Modules and thus to facilitate crosslayer design,(2) handle information and (3) provide functionalities of common
interest for all Modules. nsMiracle architecture includes the PlugIn class that is
attached to the NodeCore and is able to communicate with all Modules.
For these reason nsMiracle has been chosen to develop the first ns-2 extension
for the LTE network.
To represent our network topology with the required detail level in the first
layer of the protocol stack and to investigate the tcp behavior during the LTE
handover, we extended nsMiracle with the following components:
LTE Module represent PDCP and RLC/MAC layers and provides some basic
schedulers for all mobiles connected to the eNB. In the scheduling policy
can be chosen among Round Robin (RR) and Weighted Fair Queue (WFQ).
The calculation of the transport block (TB) dimension, we used a probabilistic distribution provided by “RedHawk” LTE simulator, which has been
conceived to capture almost all detail level of the lowest layer of the protocol stack. Such a simulator is a proprietary software provided by Ericsson
Ltd. LTE Module implements also a basic handover predictor to evaluate
the performance improvement of this mechanism.
LTE header contains some fields regarding data forwarding during the handover
and all information about fragmentation and merging of the packets done
by the LTE Module.
63
4. LTE simulator: an overview
Forwarding Module implements the data forwarding during LTE handover.
Dumb Channel Module links all the MEs to the eNBs.
Queue Monitor Porting uses the Queue Monitor available in ns-2.
Web Cross-traffic Generator Porting investigates differences produced by
unequal cross-traffic characteristic.
Also other modifications have been done to Node Core, Link Module and
Routing Module to manage the tunnel connection and the mobility trough the
network [1].
4.2
Network topology
Source
Cell
Router
eNodeB
Server
ME
Target Cell
Figure 4.2.1. Reference simulation scenario.
The topology used to perform the simulations consists of two cells served by
two different eNB (TeNB and SeNB), a variable number of MEs, a router and a
64
Summary 4.3
collection of servers that are the traffic source for each tcp connection. All these
part of the network are interconnected as shown in Figure 4.2.1.
We consider that all the tcp connections are tcp Reno. This choice is due
to its large use in the mobile applications and in Internet.
The basic parameters used in each simulation are shown in Table 4.2.1. These
are not the only parameters to set in the LTE simulator, but these are common
to all simulations performed.
Parameter
Value
Cross-traffic
FTP
Bit rate of router links 12 M b/s
Bit rate of server links 100 M b/s
Latency of wired links 2 ms
eNodeB queue length
200
Reordering
Enabled
Detach time
30 ms
Scheduling policy
WFQ
Packet size
1500 byte
Table 4.2.1. Common parameters of all simulation
In the next sections we will show which are the issues that the handover LTE
could carry and we will analyze and validate all the tcp performance proposals
to improve this mechanism.
4.3
Summary
In this chapter an overview of our LTE simulator is given, focusing to a
new framework of ns-2, called nsMiracle. The needs of use nsMiracle is due to
impossibility to reuse the existing implementations in ns-2 of the wireless channel,
because they do not work for LTE radio interfaces. A short description of the
65
4. LTE simulator: an overview
nsMiracle node is also given and the network topology implemented is shown with
some implementation parameters set to achieve the aims of this project.
66
Chapter 5
Mathematical model of TCP over LTE
In this Chapter we propose a mathematical model of the tcp during the
handover in LTE. The model is then validated by ns-2 simulations.
To develop a model that is able to capture the important dynamics of the real
system and, at the same time, be simple enough for mathematical analysis, we
have to know which class of models we want to use. There are three classes of
mathematical models fo congestion control: packet-level, fluid flow and hybrid.
The first one sees the network from a discrete-events point of view; so, it takes
into account the location of each individual packet. The second one describes
the network system using differential equation; this mean that this model does
not take into account the packet boundaries, but sees the data transport as a
continuous flow. All the quantities of interest, such as sending rates, window
sizes and queue sizes, are considered as continuous functions of time. The last
one is a result of discrete events, together with continuous changes between events.
Next a deep description of the analytical model is given taking into account the
LTE an tcp concept explained in the first chapters.
67
5. Mathematical model of TCP over LTE
5.1
Existing models
In the literature, the following models to describe the behavior of congestion
control can be found:
1. Integrator model
2. Static link model
3. Joint link model
4. DAE link model
Before analyzing the models listed above, we make some useful considerations.
First, when analyzing congestion control methods, it is useful to divide the system traffic into two types: Traffic from flows that use congestion control, and
other cross traffic, x(t) (cross traffic). Second, in fluid flow modeling of queueing
networks, data rates are the central quantities of interest. In the tcp family
of congestion control, the main control variable is the congestion window. The
sending rate is determined indirectly by the window size, under the control of the
ack-clock. For equilibrium or average quantities, that relation between window
size and the resulting sending rate is simple:
sending rate =
window size
.
roundtrip time
(5.1.1)
At this point is better to understand what the “windows size” is. We mean by
“windows size” (w(t)) the actual amount of in-flight data, which are packets that
have been transmitted but for which no ack has been received yet. Sometimes
this definition may create a confusion with congestion window in tcp, cwnd, which
is determined by the window update mechanism, but there may be a significant
discrepancy between w and cwnd.
68
Existing models 5.1
Regarding the propagation delay τ , it can be divided into forward delay τf
which is the time it takes for a transmitted packet to arrive to the bottleneck
queue, and backward delay τb which is the time it takes for a packet transmitted
on the bottleneck link to arrive to the receiver, and for the corresponding ack to
travel back to the sender.
Integrator model
The integrator model consist in the following equations:
w(t)
;
rtt(t)
w(t − τf )
q 0 (t) =
+ x(t) − c,
rtt(t)
r(t) =
(5.1.2)
where rtt(t) = τ + q(t − τ )/c which is the roundtrip time.
Static link model
The second approach is based on the observation that the window size is the
amount of data that is in transit in the network. Let us first consider the case
with no cross traffic, namely x(t) = 0. Of data in transit, all but at most cτ must
reside in the queue at the bottleneck link. This leads to the static link model
q(t) = w(t) − cτ ;
(5.1.3)
0
r(t) = c + w (t).
Such an expression for the rate can be understood as follows: Packets are
transmitted onto the bottleneck link with rate c. Then the rate of received jacks
will also be c, regardless of the window size [10].
Joint link model
As we have seen in previous paragraphs, the integrator model and the static
model describe the congestion control from two different points of view: the
69
5. Mathematical model of TCP over LTE
integrator model is accurate when the bottleneck link is dominated by crosstraffic, while the static model is accurate when the cross-traffic uses only a small
proportion of the capacity.
The Joint link model, is more accurate for any level of cross-traffic. It can be
summarized in the following equations:
w(t − τ )
+ w0 (t);
rtt(t)
w(t − τ − τf )
q 0 (t) =
+ w0 (t − τf ) + x(t) − c,
rtt(t − τf )
r(t) =
(5.1.4)
where rtt include the propagation delay τ and queueing delay.
This model, proposed by Möller in 2008 [2], is slightly more complex than the
previous ones, but the added complexity justifies the reached accuracy.
DAE link model
The DAE (Differential Algebraic Equation) link model gives quite good results. Such a model is also more complicated than the previous ones. It was
proposed by Jacobsson in 2008 [3], where it was observed that the models above
fail to capture the significant oscillations in the queue due to the large differences
in rtt. This is in contrast with the DAE link model which shows almost perfect
agreement with the packet level simulation, even at time scales finer than the
rtts (sub-rtt time scales).
5.2
Model adaptation
To show how the LTE handover procedure works in the congestion control
scenario we are using a fluid-flow model. In particular we choose the joint link
model proposed by Möller in [2]. Our scenario is shown in Figure 5.2.1, where
the ME(0) is the mobile that is doing handover (moving from Source eNodeB SeNB - to Target eNodeB - TeNB -) following the procedure stated in [4].
70
Model adaptation 5.2
SeNB
INTERNET
TeNB
ME(0)
ME(n)
ME(1)
ME(i)
Figure 5.2.1. LTE handover scenario.
We are assuming to use tcp Reno. Suppose the roundtrip propagation delay
as a constant and denoting it by τ (without taking into account the time spent
in the queue) tcp Reno allows us to send one full windows (w) of data each τ ,
consisting of w/m packets, where m is the packet size.
Let p be the packet loss probability for each packet. There are two possible
window updates:
1. With probability (1 − p), an ack is received, and the additive increase
mechanism increases the window by ∆w(t) = m2 /w(t).
2. With probability p, the packet is lost, and the multiplicative decrease mechanism reduces the window size by ∆w(t) = −w(t)/2.
In that way it is easy writing the derivative of the continuous function w(t)
as the ratio of his expected value
(1 − p)m2 pw(t)
E[w(t)] =
−
w(t)
2
71
5. Mathematical model of TCP over LTE
and the average inter-packet time T = τ m/w:
(1 − p)m pw2 (t)
dw
=
−
dt
τ
2τ m
.
(5.2.1)
By extending this result with the time-varying queueing delay, (5.2.1) becomes
dw
(1 − p)m
pw2 (t)
=
−
,
dt
τ + q(t − τb )/c 2m (τ + q(t − τb )/c)
(5.2.2)
where c is the link capacity and τb is the backward delay.
5.2.1
Single flow and multiple flow model
In the previous section is shown that for single flow traversing a single bottleneck, the state equations are
w(t − τ )
+ w0 (t);
rtt(t)
w(t − τ − τf )
q 0 (t) =
+ w0 (t − τf ) + x(t) − c,
rtt(t − τf )
r(t) =
(5.2.3)
where w0 (t) is defined by (5.2.2) for tcp Reno, x(t) is the cross traffic (it includes
all non tcp traffic type) and rtt(t) = τ + q(t − τb )/c, where τ = τf + τb . τf and
τb are the forward and backward delay, respectively.
When we have multiple flows traversing a single bottleneck, every k-th ones
have a window size wk (t), sending rate rk (t), and a roundtrip propagation delay
τk . Let us extend the results found in (5.2.3), below is shown how we can easily
write the equation for multiple flows. Without loss of generality, we can assume
that τf,k = 0.
wk (t − τk )
+ wk0 (t),
rttk (t)
X wk (t − τk ) X
q 0 (t) =
+
wk0 (t) + x(t) − c,
rttk (t)
k
k
rk (t) =
where now rttk (t) = τ + q(t − τk )/c, in which τk = τb,k .
72
(5.2.4)
Application scenario 5.3
Flows with the same roundtrip propagation delay behave as a single aggregated flow, where the rate and the window size of the aggregated packets are the
sums of the rate and window sizes of the individual flows. With heterogeneous
delays, the dynamics becomes more complex.
5.3
Application scenario
In this section we apply the model illustrated before to our scenario by dividing
the problem in three phases:
1. Handover preparation: the ME(0) is still attached to the network by the
SeNB and other mobiles are connected with TeNB;
2. Handover actuation: this phase is the time in which the mobile ME(0)
starts the handover procedure (at time tHO ) and his data - stored in the
SeNB’s buffer - start to be forwarded to TeNB through the core network
(Router);
3. Handover completion: the handover has finished and the ME(0) is attached to the network by the TeNB with all the other mobiles and share
the radio resource with them.
5.3.1
Handover preparation
In the beginning we suppose that the ME(0) is the only mobile attached to
SeNB, so we start the analysis by studying the queue state over the link between
the router and SeNB (see Figure 5.2.1) taking into account that all the connection
between MEs and sources are tcp ones, so we can handle this problem as a single
tcp flow traversing a single bottleneck. The state equation are given by (5.2.3).
Meanwhile, studying the queue state over the link between the router and
TeNB (see Figure 5.2.1), we can suppose to handle the problem as a multiple
73
5. Mathematical model of TCP over LTE
tcp flows traversing a single bottleneck; so the state equation will become (5.2.4)
with 1 ≤ k < n, where n is the total number of attached mobiles to the TeNB.
5.3.2
Handover actuation
At the handover time, tHO , we suppose that all the data in the SeNB’s buffer
are moved to the TeNB’s buffer in a priority queue to avoid the out of order
packets arrival to ME(0). Hence, the packet amount that are forwarded to the
TeNB at time tHO , is the in-flight data at tHO added to the data sent from the
sources, between the attach time (tAT T ) and path sw time (tP S ), which is 10ms
for LTE. We call by FW this packet amount. We model this data transfer as the
time that a link with capacity c uses to transmit FW data; in that way, the integral
is exactly the data to forwards. Specifically, for the ME(0) in the TeNB, we have
(denoting his flow by k = 0) that
w(t − τ0 )
+ w00 (t);
rtt0 (t)
w(t − τ0 )
t − (tHO + FW/2c)
0
0
+ w (t − τ0 ) + x(t) − c + c · rect
q0 (t) =
,
rtt0 (t)
FW/c
r0 (t) =
(5.3.1)
where
Z
tP S
FW = w(tHO ) +
r0 (t)dt
(5.3.2)
tAT T
and for all the other flows the equations are given by (5.2.4).
5.3.3
Handover completion
After handover at TeNB the situation will be the same as before, with the
exception of the presence of a further mobile. Therefore, the equations for all the
flows are given by (5.2.4) but now 0 ≤ k < n. At SeNB we don’t have anymore
tcp connections.
74
Simulations and Results 5.4
5.4
Simulations and Results
The simulations are performed using Matlab/Simulinkr environment and are
compared with LTE simulator results. The parameters used in both simulations
are shown in the Table 5.4.1.
Parameter
Total simulation time
Nodes attached to SeNB
Nodes attached to TeNB
Queue size
cwndMAX
Handover Time
Value
10 s
1
4
No Limit
44 ' 64 KB
4s
Table 5.4.1. Parameters for simulation results and mathematical model comparison
The model shows the behavior of the bit rate, buffer status, rtt and cwnd. In
all the figures the continuous lines are the simulation results and the dot-dashed
lines are the models results.
In Figure 5.4.1 is shown the bit rate of the mobile that does the handover
from one SeNB to which is connected, towards a TeNB in which there are other
four mobiles connected. The mobile that does the handover is represented with
the red solid line (LTE simulator) and the black dot-dashed line. The other two
plots depict the bit rate of one of the four mobile connected to the TeNB. As
we can see, our model gives a good approximation of the real behavior of the
handover procedure in LTE.
In Figure 5.4.2 and in Figure 5.4.3 show the cwnd status with the same notation used in Figure 5.4.1. In Figure 5.4.2 we nave not reported the slow start
phase because it is not implemented in the model. For this reason in the LTE
simulator the initial SSthresh is set to 22 packets. In Figure 5.4.3 we see a good
75
5. Mathematical model of TCP over LTE
15
BW0 Math
BW1 Math
BW0
Bitrate [Mb/s]
BW1
10
5
0
0
1
2
3
4
5
Time[s]
6
7
8
9
10
Figure 5.4.1. Bit rate comparison.
50
45
40
CWND [packets]
35
30
25
20
15
ME0 Math
10
ME1 Math
ME0
5
ME1
0
0
1
2
3
4
5
Time[s]
6
7
Figure 5.4.2. cwnd comparison.
76
8
9
10
Simulations and Results 5.4
250
Queues[packets]
200
150
100
eNB0 Math
50
eNB1 Math
eNB0
eNB1
0
0
1
2
3
4
5
Time[s]
6
7
8
9
10
Figure 5.4.3. Buffer status comparison.
approximation of the queue status with the exception of slight oscillations. The
buffer utilization of the router are referred to the link between router and TeNB
and between router and SeNB. The first one is drawn with green solid line and
blue dot-dashed line for the LTE simulation results and the model, respectively.
The second queue status is plotted with red solid line and black dot-dashed line
for the LTE simulation results and the model, respectively.
rtt behavior of the packets that are departed from tcp source is reported
in Figure 5.4.4. Before the handover, the model (blue and black dot-dashed
lines) follows well the simulation results (red and green dotted lines). After the
handover time, all the mobiles are attached to the TeNB and the rtt values
have slight oscillations that the analytical model is not able to capture. These
oscillations are due to the scheduler in the TeNB, which causes a random time to
assign the radio resource to each ME to schedule for transmission.
77
5. Mathematical model of TCP over LTE
0.35
0.3
RTT [s]
0.25
0.2
0.15
0.1
ME0 Math
ME1 Math
0.05
ME0
ME1
0
0
1
2
3
4
5
Time[s]
6
7
8
9
10
Figure 5.4.4. rtt comparison.
5.5
Summary
This chapter gives the main contribution of the thesis because an analytical
model is proposed to describe the performances of the tcp protocol in the LTE
handover procedure. The description is given by an overview of the existing
models and the reason for the choice of the Joint link model. Here is shown how
to adapt this model for our scenario and to describe with the valid accuracy all
the dynamics of the real system. A validation of such a model is performed by
simulations in ns-2 and it is shown that it follows quite well the behavior of the
simulation results.
78
Chapter 6
Conclusion and future work
6.1
Conclusion
In this thesis we have studied the impacts on the end-user and system performance when users with high bit rates tcp services are moving through the
network. In particular we have focused on the tcp performances during the LTE
handover procedure.
To reach such an aim we proposed a simple modification of the analytical
model based on the fluid flow model Joint link proposed by Möller in 2008 [2].
The validation of this model has been done by simulations performed with ns-2
and the framework nsMiracle which to our best knowledge is the first simulator
developed to evaluate LTE performance.
The result showed a good agreement analysis-simulations of the real behavior
of the handover procedure in LTE.
6.2
Future work
Developments for future studies could be the extension of the proposed model
by evaluating the tcp performances during the LTE handover with a finite buffer
size of the router, because the model proposed in this thesis does not take into
79
6. Conclusion and future work
account such a case. This involves a previous study of packet loss probability,
which the buffer size is dependent on.
80
Bibliography
[1] Davide Pacifico, “Analysis and Performance Improvement of tcp during
Handover of LTE ”, Msc Thesis, February 2009.
[2] Niels Möller, “Window-based congestion control - Modeling, analysis and
design”, Phd Thesis, January 2008
[3] Krister Jacobsson, “Dynamic modeling of Internet congestion control ”, Phd
Thesis, April 2008
[4] 3GPP TS 36.300, “Evolved UTRA and evolved UTRAN, overall description”,
http://www.3gpp.org/ftp/Specs/archive/36_series/36.300/, 2006.
[5] R. van Nee, R. Prasad, “OFDM for Multimedia Wireless Communications”,
Artech House, 2000.
[6] A. Racz, A. Temesvary, N. Reider, “Handover Performance in 3GPP Long
Term Evolution (LTE) Systems”, Mobile and Wireless Communications
Summit, 2007. 16th IST
[7] 3GPP TS 23.402, “Architecture enhancements for non-3GPP accesses”,
http://www.3gpp.org/ftp/specs/archive/23_series/23.402/, 2006.
81
Bibliography
[8] 3GPP TS 25.913, “Requirements for Evolved UTRA (E-UTRA) and Evolved
UTRAN (E-UTRAN)”, http://www.3gpp.org/ftp/specs/archive/25_
series/25.913/, 2006.
[9] 3GPP TS 36.101, “E-UTRA and UE radio transmission and reception”,
http://www.3gpp.org/ftp/Specs/archive/36_series/36.300/, 2008.
[10] J. Wang, D. X. Wei, and S. H. Low. “Modelling and stability of fast tcp.
In Proceedings of IEEE Infocom, Miami, March 2005.
[11] J. Postel, “RFC 793: Transmission Control Protocol ”, IETF, Tech. Rep.,
September 1981.
[12] J. Postel, “RFC 768: User Datagram Protocol ”, IETF, Tech. Rep., August
1980.
[13] M. Allman, V. Paxson, and W. R. Stevens, “RFC 2581: tcp congestion
control ”, IETF, Tech. Rep., April 1999. A.
[14] Dahlen and P. Ernstrm, “tcp over umts”, RVK02, 2002.
[15] 3GPP TS 29.060, “GPRS tunnelling protocol across the Gn and Gp interface (Rel 7)”, http://www.3gpp.org/ftp/specs/archive/29_series/29.
060/, 2006.
[16] M. Mathis, J. Mahdavi, S. Floyd, and A. Romanow, “RFC 2018: tcp selective acknowledgment options”, IETF, Tech. Rep., April 1996.
[17] M. Allman, S. Floyd and C. Partridge, “RFC 3390: Increasingtcp’s initial
window ”, IETF, Tech. Rep., October 2002.
82
Bibliography
[18] G. Xylomenos, G. C. Polyzos, M. Saaranen, and P. Mähönen, “tcp/ip Performance over Wireless Networks”, M. HASSAN and R. JAIN, Eds. PRENTICE HALL, 2002.
[19] K. Ramakrishnan, S. Floyd and D. Black, “RFC 3168: The Addition of Explicit Congestion Notification (ECN) to IP ”, IETF, Tech. Rep., September
2001.
[20] L. Le, J. Aikat, K. Jeffay and F. D. Smith, “The effects of active queue management on web performance”, In Proceedings of the 2003 conference on
Applications, technologies, architectures, and protocols for computer communications, Karlsruhe, 2003.
[21] S. Floyd and V. Jacobson, “Random Early Detection Gateways for Congestion Avoidance”, Transactions on Networking, August 1993.
[22] S. Floyd, “RED: Discussions of Setting Parameters”, November 1997, http:
//www.aciri.org/floyd/REDparameters.txt.
[23] W. Feng, D. Kandlur, D. Saha, and K. Shin, “A Self-Configuring RED Gateway”, Infocom, March 1999.
[24] Tae-Hyong Kim, Qiping Yang, Jae-Hyoung Lee, Soon-Gi Park and YeonSeung Shin, “A Mobility Management Technique with Simple Handover Prediction for 3G LTE Systems”, Vehicular Technology Conference, 2007.
[25] Hyeyeon Kwon, Mi-Jeong Yang, Ae-Soon Park and S.Venkatesan, “Handover
Prediction Strategy for 3G-WLAN Overlay Networks”, Network Operations
and Management Symposium, April 2008.
[26] Nicola Baldo, Federico Maguolo, Marco Miozzo, Michele Rossi and Michele
Zorzi, “ns2-MIRACLE: a Modular Framework for Multi-Technology and
83
Bibliography
Cross-Layer Support in Network Simulator 2 ”, NSTools ’07, October 2007,
Nantes, France.
[27] “The network simulator ns-2 ”, http://www.isi.edu/nsnam/ns/.
[28] “ns-MIRACLE: Multi InteRfAce Cross Layer Extension for ns-2 ”, http:
//www.dei.unipd.it/ricerca/signet/tools/nsmiracle.
[29] “Dynamic Modules in ns-2 ”, http://telecom.dei.unipd.it/ns/miracle/
ns_dynamic_libraries/.
[30] T. Berners-Lee, R. Fielding and H. Frystyk, “RFC 1945: Hypertext Transfer
Protocol (HTTP/1.0)”, IETF, Tech. Rep., May 1996
[31] R. Fielding, J. Gettys and H. Frystyk, “RFC 2616: Hypertext Transfer Protocol (HTTP/1.1)”, IETF, Tech. Rep., May 1996
[32] F. Xinz and A. Jamalipour, “TCP throughput and fairness performance in
presence of delay spikes in wireless networks”, International Journal of Communication Systems, March 2005
[33] A. Speranzon Fischione C. and K.H. Johansson, “Distributed and collaborative estimation over wireless sensor networks”, IEEE Conference on Decision
and Control, 2006.
[34] A. Speranzon Fischione C. and K.H. Johansson, and A. SangiovanniVincentelli, “A Distributed Minimum Variance Estimator for Sensor Networks”, IEEE Journal on Selected Areas in Communications, Special Issue
on Control and Communication, May 2008, p. 609-621, vol. 26.
84
Bibliography
[35] Speranzon Fischione C. and K.H. Johansson, and A. Sangiovanni-Vincentelli,
“Peer-to-Peer Estimation over Wireless Sensor Networks via Lipschitz Optimization”, ACM/IEEE IPSN, 2009.
85
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

Related manuals

Download PDF

advertising