Scalable Multiple Description Coding and Distributed Video Streaming Over 3G Mobile Networks

Download as pdf or txt
Download as pdf or txt
You are on page 1of 144

Scalable Multiple Description Coding and

Distributed Video Streaming over


3G Mobile Networks



by



Ruobin Zheng




A thesis

presented to the University of Waterloo

in fulfilment of the

thesis requirement for the degree of

Master of Applied Science

in

Electrical and Computer Engineering



Waterloo, Ontario, Canada, 2003



Ruobin Zheng 2003
ii

I hereby declare that I am the sole author of this thesis. This is a true copy of the thesis,
including any required final revisions, as accepted by my examiners.

I understand that my thesis may be made electronically available to the public.
iii
Abstract

In this thesis, a novel Scalable Multiple Description Coding (SMDC) framework is
proposed. To address the bandwidth fluctuation, packet loss and heterogeneity problems
in the wireless networks and further enhance the error resilience tools in Moving Pictures
Experts Group 4 (MPEG-4), the joint design of layered coding (LC) and multiple
description coding (MDC) is explored. It leverages a proposed distributed multimedia
delivery mobile network (D-MDMN) to provide path diversity to combat streaming video
outage due to handoff in Universal Mobile Telecommunications System (UMTS). The
corresponding intra-RAN (Radio Access Network) handoff and inter-RAN handoff
procedures in D-MDMN are studied in details, which employ the principle of video
stream re-establishing to replace the principle of data forwarding in UMTS. Furthermore,
a new IP (Internet Protocol) Differentiated Services (DiffServ) video marking algorithm
is proposed to support the unequal error protection (UEP) of LC components of SMDC.
Performance evaluation is carried through simulation using OPNET Modeler 9.0.
Simulation results show that the proposed handoff procedures in D-MDMN have better
performance in terms of handoff latency, end-to-end delay and handoff scalability than
that in UMTS. Performance evaluation of our proposed IP DiffServ video marking
algorithm is also undertaken, which shows that it is more suitable for video streaming in
IP mobile networks compared with the DiffServ video marking algorithm (DVMA)
proposed in [71] [72].
iv
Acknowledgements

I would like to express my deepest gratitude to my supervisor, Professor W. Zhuang,
for her guidance throughout this work. She was always available, encouraging and
professional. From her, I have learned not only how to do good research, but also how to
be a better person. Her encouragement, patience and support made my thesis possible. In
addition, I would like to thank Professors X. Shen and P. Ho for their careful and
thorough reading of this thesis, and for their valuable and helpful comments for my
research work.
v
Contents

1 INTRODUCTION......................................................................................................1
1.1 MOTIVATION OF THIS RESEARCH..............................................................................1
1.2 THESIS ORGANIZATION.............................................................................................7
2 HANDOFF DESIGN ISSUES IN MEDIA DELIVERY........................................8
2.1 GPRS NETWORK ARCHITECTURE.............................................................................8
2.2 3GPP UMTS NETWORK ARCHITECTURE..................................................................9
2.3 UMTS HANDOFF PROCEDURES FOR MEDIA STREAMING........................................11
2.3.1 Assumptions ...................................................................................................11
2.3.2 Handoff Procedures in UMTS Release 4 .......................................................12
2.4 HANDOFF PROBLEMS IN GPRS/UMTS ...................................................................19
2.4.1 Bearer Service QoS and Seamless Handoff ...................................................19
2.4.2 Transfer Delay ...............................................................................................21
2.4.3 Handoff Latency (Media Stream Interruption) ..............................................21
2.4.4 Media Synchronization ..................................................................................25
2.4.5 Handoff Scalability ........................................................................................27
3 SYSTEM MODEL...................................................................................................29
3.1 LAYERED CODING...................................................................................................29
3.2 MULTIPLE DESCRIPTION CODING............................................................................33
3.3 DISTRIBUTED MULTIMEDIA DELIVERY MOBILE NETWORK....................................37
3.3.1 Concept of Content Delivery Network...........................................................37
3.3.2 Proposed Mobile Network Model ..................................................................38
3.3.3 MPEG-4 Video Streaming over IP DiffServ..................................................42
3.4 PROTOCOL STACK OF END SYSTEMS.......................................................................47
3.4.1 MPEG-4 Media Transport .............................................................................48
3.4.2 MPEG-4 Media Control.................................................................................49
4 PROPOSED SOLUTIONS FOR VIDEO MOBILITY........................................51
4.1 JOINT DESIGN OF MDC AND LC .............................................................................51
vi
4.1.1 Architecture of Proposed SMDC Framework................................................52
4.1.2 Scalability Structure of SMDC Framework...................................................54
4.1.3 Advantages of Proposed LC Component .......................................................57
4.2 MPEG-4 VIDEO STREAMING OVER IP DIFFSERV....................................................62
4.2.1 QoS Mapping between UMTS and IP DiffServ..............................................62
4.2.2 IP DiffServ MPEG-4 Video Marking Algorithm............................................63
4.2.3 Evolution of System Model for Native IP DiffServ ........................................66
4.3 PROPOSED HANDOFF PROCEDURES FOR VIDEO STREAMING...................................70
4.3.1 Proposed intra-RAN Handoff Procedure.......................................................70
4.3.2 Proposed inter-RAN Handoff Procedure.......................................................71
4.3.3 Handoff Enhancement for Streaming Services ..............................................73
5 SIMULATIONS .......................................................................................................76
5.1 SIMULATION MODELS .............................................................................................76
5.2 SYSTEM SETUP AND TEST CONDITIONS ..................................................................83
5.2.1 Rate Shaping (Filtering) and Packetization ..................................................83
5.2.2 BER over Rayleigh Fading Channels ............................................................84
5.2.3 Delay-constrained ARQ.................................................................................86
5.2.4 Traffic Profile.................................................................................................87
5.2.5 Test Cases ......................................................................................................92
5.3 SIMULATION RESULTS AND ANALYSIS....................................................................95
5.3.1 Effect of BER on FER over Wireless Channels .............................................95
5.3.2 Effect of AF Queue Size .................................................................................98
5.3.3 Effect of Video Marking Algorithm................................................................98
5.3.4 Effect of Streaming Video Handoff ..............................................................100
6 CONCLUSIONS AND FUTURE WORK...........................................................120
APPENDIX.....................................................................................................................122
PROPOSED INTER-CELL, INTRA-RNS HANDOFF PROCEDURE..........................................122
ACRONYMS..................................................................................................................123
REFERENCES...............................................................................................................128
vii
List of Tables

Table 1-1 Error characteristics of video communication............................................4
Table 2-1 QoS attributes required by audio and video media streams in 3GPP.......20
Table 2-2 The required amount of transcoding state information to be transferred .24
Table 3-1 Comparison of different scalability modes in MPEG-2............................33
Table 3-2 Single Description versus Multiple Description.......................................42
Table 4-1 Methods of adapting bit rate of each description.....................................60
Table 4-2 Mapping of UMTS QoS classes to DiffServ PHB classes .......................62
Table 4-3 Proposed IP DiffServ MPEG-4 video marking algorithm .......................66
Table 4-4 Summary of handoff solution comparison...............................................74
Table 5-1 Test Cases of BER and FER in wireless channels ...................................86
Table 5-2 Traffic profile of each cell for evaluation of video marking algorithm....88
Table 5-3 UMTS bearer service attributes of streaming class..................................88
Table 5-4 Traffic profile of each cell for evaluation of streaming video handoff ....88
Table 5-5 WFQ profile for proposed IP DiffServ video marking algorithm............92
Table 5-6 Test Cases of Effect of Queue Size ..........................................................92
Table 5-7 Queue scheduling configuration...............................................................94
Table 5-8 WRED profile for AF queue in DVMA...................................................94
Table 5-9 Test Cases of Video Marking Algorithm.................................................94
Table 5-10 Test Cases of Handoff Procedures..........................................................95
Table 5-11 Comparison of QoS (e.g., FER) guarantee in three solutions ................99
Table 5-12 Comparison of handoff performance in UMTS and MDMN...............101

viii
List of Figures

Figure 1-1 Unicast and Multicast video distribution..................................................5
Figure 2-1 GPRS Network Architecture.....................................................................8
Figure 2-2 UMTS Network Architecture..................................................................10
Figure 2-3 UMTS Rel4 network model of intra-RAN handoff (Data plane) ...........12
Figure 2-4 Intra-RAN handoff procedure in UMTS Rel4 (Control plane)...............14
Figure 2-5 UMTS Rel4 network model of inter-RAN handoff (Data plane) ...........16
Figure 2-6 Inter-RAN handoff procedure in UMTS (Control plane: Phase I) .........17
Figure 2-7 Inter-RAN handoff procedure in UMTS (Control plane: Phase II&III) .18
Figure 2-8 Temporal dependency in a MPEG-based coding stream........................22
Figure 2-9 Types of handoff in GSM/GPRS/UMTS ................................................27
Figure 3-1 Block diagram of layered coding with transport prioritization...............29
Figure 3-2 Block diagram of MDC coding and decoding ........................................34
Figure 3-3 Proposed network model of intra-RAN handoff (Data plane) ...............39
Figure 3-4 Proposed network model of inter-RAN handoff (Data plane) ................40
Figure 3-5 Single Description versus Multiple Description.....................................41
Figure 3-6 Queue management and scheduling mechanisms of AF PHBs ..............44
Figure 3-7 Protocol stack in UMTS and the Proposed D-MDMN (Data plane) ......45
Figure 3-8 Protocol stack of network-aware end systems ........................................47
Figure 4-1 Proposed SMDC architecture..................................................................52
Figure 4-2 Single Description versus Multiple Description.....................................54
Figure 4-3 Proposed SMDC scalability structure .....................................................55
Figure 4-4 Video illustration of SMDC layered structure ........................................56
Figure 4-5 Balanced and Unbalanced MD Operation...............................................58
Figure 4-6 Comparison of Unbalanced MDC with Unbalanced SMDC ..................59
Figure 4-7 Packet loss effect on frame error rate......................................................65
Figure 4-8 UMTS network model in IP native mode with 3G AR...........................67
Figure 4-9 D-MDMN network model in IP native mode with 3G AR.....................68
Figure 4-10 Evolution of Protocol stack (Data plane) ..............................................69
Figure 4-11 Propose intra-RAN handoff procedure (Control plane) ........................70
Figure 4-12 Proposed inter-RAN handoff procedure (Control plane) ......................72
Figure 5-1 UMTS simulation model (Intra-RAN Handoff) .....................................78
ix
Figure 5-2 MDMN simulation model (Intra-RAN Handoff)....................................78
Figure 5-3 UMTS simulation model (Inter-RAN Handoff) .....................................79
Figure 5-4 MDMN simulation model (Inter-RAN Handoff)....................................79
Figure 5-5 Node Model of RNS................................................................................80
Figure 5-6 Node Model of SGSN (or GGSN) ..........................................................80
Figure 5-7 Node Model of SGSN in the proposed MDMN......................................80
Figure 5-8 Node Model of MS..................................................................................81
Figure 5-9 Radio Transceiver Pipeline Model..........................................................82
Figure 5-10 Radio Transceiver Attributes for Specifying Pipeline Stages ...............82
Figure 5-11 Level of rate shaping.............................................................................83
Figure 5-12 BER performance over flat Rayleigh fading channels..........................85
Figure 5-13 Layered Video Traffic (Video Client1).................................................89
Figure 5-14 Video Traffic (Video Client1)...............................................................89
Figure 5-15 Background Traffic (Video Client2).....................................................90
Figure 5-16 Background Traffic (25 Voice Users) ...................................................90
Figure 5-17 Background Traffic (Web Users) ..........................................................90
Figure 5-18 Video Traffic (Handoff occurs 28 times) ..............................................91
Figure 5-19 Background Traffic (Video Clients) .....................................................91
Figure 5-20 Background Traffic (5 Voice Users) .....................................................91
Figure 5-21 Background Traffic (Web Users) ..........................................................91
Figure 5-22 Bit Error Rate over the Rayleigh fading channel (Test Case )...........96
Figure 5-23 Bit Error Rate over the Rayleigh fading channel (Test Case )...........96
Figure 5-24 Bit Error Rate over the Rayleigh fading channel (Test Case ) ...........96
Figure 5-25 FER over the Rayleigh fading channel (Test Case ) ..........................97
Figure 5-26 FER over the Rayleigh fading channel (Test Case ) ..........................97
Figure 5-27 FER over the Rayleigh fading channel (Test Case )...........................97
Figure 5-28 Average of End-to-End Delay (Test Case 1~9: BER = 10
-5
)..............103
Figure 5-29 Average of Packet Loss Ratio (Test Case 1~9: BER = 10
-5
) ..............103
Figure 5-30 Average of End-to-End Delay (Test Case (1)~ (9): BER = 10
-4
)........104
Figure 5-31 Average of Packet Loss Ratio (Test Case (1)~ (9): BER = 10
-4
)........104
Figure 5-32 FER (Test Case A: Best Effort, FIFO, BER = 10
-5
)............................105
Figure 5-33 FER (Test Case B: DiffServ, WRED, BER = 10
-5
) ............................105
Figure 5-34 FER (Test Case C: DiffServ, WFQ, BER = 10
-5
) ...............................106
x
Figure 5-35 FER (Test Case a: Best Effort, FIFO, BER = 10
-4
).............................106
Figure 5-36 FER (Test Case b: DiffServ, WRED, BER = 10
-4
) .............................107
Figure 5-37 FER (Test Case c: DiffServ, WFQ, BER = 10
-4
) ................................107
Figure 5-38 End-to-end Delay (Test Case A: Best Effort, FIFO, BER = 10
-5
) ......108
Figure 5-39 End-to-end Delay (Test Case B: DiffServ, WRED, BER = 10
-5
) .......108
Figure 5-40 End-to-end Delay (Test Case C: DiffServ, WFQ, BER = 10
-5
) ...........108
Figure 5-41 End-to-end Delay (Test Case a: Best Effort, FIFO, BER = 10
-4
) .......109
Figure 5-42 End-to-end Delay (Test Case b: DiffServ, WRED, BER = 10
-4
).........109
Figure 5-43 End-to-end Delay (Test Case c: DiffServ, WFQ, BER = 10
-4
)...........109
Figure 5-44 Delay Jitter (Test Case A: Best Effort, FIFO, BER = 10
-5
) ................110
Figure 5-45 Delay Jitter (Test Case B: DiffServ, WRED, BER = 10
-5
) .................110
Figure 5-46 Delay Jitter (Test Case C: DiffServ, WFQ, BER = 10
-5
) ....................110
Figure 5-47 Delay Jitter (Test Case a: Best Effort, FIFO, BER = 10
-4
) .................111
Figure 5-48 Delay Jitter (Test Case b: DiffServ, WRED, BER = 10
-4
)..................111
Figure 5-49 Delay Jitter (Test Case c: DiffServ, WFQ, BER = 10
-4
).....................111
Figure 5-50 End-to-End Delay (Test Case I: UMTS, Intra-RAN, BER = 10
-5
) .....112
Figure 5-51 Delay Jitter (Test Case I: UMTS, Intra-RAN, BER = 10
-5
)................112
Figure 5-52 End-to-End Delay (Test Case II: MDMN, Intra-RAN, BER = 10
-5
) ..113
Figure 5-53 Delay Jitter (Test Case II: MDMN, Intra-RAN, BER = 10
-5
).............113
Figure 5-54 End-to-End Delay (Test Case III: UMTS, Inter-RAN, BER = 10
-5
) ..114
Figure 5-55 Delay Jitter (Test Case III: UMTS, Inter-RAN, BER = 10
-5
) .............114
Figure 5-56 End-to-End Delay (Test Case IV: MDMN, Inter-RAN, BER = 10
-5
) 115
Figure 5-57 Delay Jitter (Test Case IV: MDMN, Inter-RAN, BER = 10
-5
) ...........115
Figure 5-58 End-to-End Delay (Test Case i: UMTS, Intra-RAN, BER = 10
-4
) .....116
Figure 5-59 Delay Jit ter (Test Case i: UMTS, Intra-RAN, BER = 10
-4
) ................116
Figure 5-60 End-to-End Delay (Test Case ii: MDMN, Intra-RAN, BER = 10
-4
) ..117
Figure 5-61 Delay Jitter (Test Case ii: MDMN, Intra-RAN, BER = 10
-4
) .............117
Figure 5-62 End-to-End Delay (Test Case iii: UMTS, Inter-RAN, BER = 10
-4
) ...118
Figure 5-63 Delay Jitter (Test Case iii: UMTS, Inter-RAN, BER = 10
-4
)..............118
Figure 5-64 End-to-End Delay (Test Case iv: MDMN, Inter-RAN, BER = 10
-4
)..119
Figure 5-65 Delay Jitter (Test Case iv: MDMN, Inter-RAN, BER = 10
-4
) ............119
Figure 0-1 Proposed inter-cell, intra-RNS handoff procedure (Control plane) ......122

1
1 Introduction
With the emergence of broadband wireless networks and increasing demand of
multimedia information on the Internet, wireless video communications have received
great interest from both industry and academia, and wireless multimedia services are
foreseen to become widely deployed in this decade.
Real-time transport of live video or stored video is the predominant part of real- time
multimedia. There are two concepts for delivery of stored video over the Internet or the
wireless networks, namely the downloadable video and the streaming video [1].
The video download is the same concept as the file download, but a large file. The
entire video file is expected to be downloaded on the local machine, where it could be
played back using the standard media software. It allows simple delivery mechanisms,
e.g., Transmission Control Protocol (TCP). However, it usually suffers long and perhaps
unacceptable transfer time and large storage space. Also, download before viewing
requires the users patient.
In contrast, the streaming video is partitioned into packets. It needs not be
downloaded in full, but is being played out simultaneously during video delivery. There
is relatively low delay (e.g., Real and Microsoft use 5-15 seconds) of starting playback
before viewing. It also minimizes the storage requirement.
In this thesis, we are more concerned with the streaming video, which refers to real-
time transmission of stored video. Its relevant techniques are also applicable to the
delivery issues of live video.
1.1 Motivation of This Research
1. Bandwidth, packet loss and heterogeneity problems
Due to its real-time nature of transmission of stored video, video streaming typically
has quality of service (QoS) requirements, e.g., bandwidth, delay and error requirements.
However, unreliability, bandwidth fluctuations and high bit error rate of wireless
2
channels can cause severe degradation to video quality. In addition, for video multicast,
network heterogeneity and receiver heterogeneity make it difficult to achieve efficiency
and flexibility. The bandwidth problems, packet loss and transmission error, and
heterogeneity problems will be discussed in detail as follows.
To address the bandwidth fluctuation, packet loss and heterogeneity problems,
scalable coding (i.e., layered coding) and rate shaping are employed for transport delay-
and bandwidth-sensitive video. Furthermore, to enhance the video quality in the presence
of unavoidable packet loss and/or bit error, open- loop error control (e.g., multiple
description coding) and close- loop error control (e.g., delay-constrained retransmission)
are explored in this thesis.
1) Bandwidth problems
To achieve acceptable presentation quality, a streaming application typically has
minimum bandwidth requirement. However, the current Internet offers only the best-
effort service and does not provide any bandwidth reservation mechanism. It is well
known that fluctuations of the Internet traffic have a fractal- like scaling behavior over
time scales [25]. Due to the self- similar nature of traffic fluctuations in the Internet, the
available bandwidth is unknown and dynamic.
In the wireless networks, the wireless channel suffers from both bandwidth
fluctuation and bandwidth limitation: (1) The throughput of a wireless channel may be
reduced due to multipath fading, shadowing, co-channel interference, and noise
disturbances; (2) When a mobile terminal moves between different networks (e.g. from a
wireless local area network (LAN) to a wireless wide area network (WAN)), the
available bandwidth may vary drastically (e.g., from a few megabits per second to a few
kilobits per second); (3) When a handoff takes place, a base station may not have enough
unused radio resources to meet the demand of a newly admitted mobile host. Thus, the
available bandwidth of wireless channel is time-varying and even unknown.
If the transmission rate of streaming video is faster than the available bandwidth, the
congestion will occur, resulting in bursty loss, excessive delay and severe drop in video
quality. On the contrary, it invokes the inefficient utilization of available bandwidth and
the sub-optimal video quality. Thus, it is desirable for streaming video application to
employ congestion control mechanisms to match video bit rate with available bandwidth.
3

2) Packet loss and transmission error
In the wired link of a mobile network or the Internet, most errors are caused by
packet loss due to congestion or misrouting. The effects of packet loss greatly depend on
the types of packet loss [47]: isolated single packet loss, burst packet loss and temporary
outage (loss of communication). Misrouting may occur in the downlink during handoffs,
which will incur bursty packet loss and temporary outage.
The wireless channels are typically more error-prone at the bit level. The wireless
link of a mobile network suffers from very high bit error rate (BER) due to multipath
fading, shadowing, co-channel interference, noise disturbances and handoff. The types of
transmission bit error also can be classified into three groups: isolated single bit error,
burst bit error, and temporary outage (e.g., due to handoff).
The effects of packet loss or bit error are significant for video streaming due to error
propagation [27]. Predictive video-encoding algorithms employ motion compensation to
achieve high compression by reducing temporal redundancies between successive frames.
When this motion information is lost to the decoder, a reconstruction error can occur.
Such errors can propagate temporally and spatially if the affected region is subsequently
used as a prediction for motion compensation. Furthermore, differential encoding is also
employed to reduce statistical redundancies, for example in the encoding of motion and
quantizer information. Loss of such information can cause additional spatial degradation
throughout the affected frames by producing incorrectly predicted motion vectors and
quantizer levels. Because of motion compensation, these errors also can propagate
temporally and spatially.
Because of error propagation of streaming video, isolated single packet loss or bit
error is converted to burst packet loss or bit error. Also, the video packet which arrives
beyond a delay bound is useless and has to be considered lost. Such loss or error can
potentially make the visual presentation displeasing to human eyes or even make the
presentation impossible.
From a video communication perspective, it is important to reduce or eliminate the
effects of burst loss/error and outage. The error characteristics of video communication in
different environments are roughly summarized in Table 1-1 [28] [38].
4
Table 1-1 Error characteristics of video communication
Application Error Characteristics
Video phone over PSTN (H.324) Very few bit errors and packet losses
Video conferencing over ISDN (H.320) Practically error free (BER=10
-10
~10
-8
)
Video conferencing over ATM (H.310, H.321) Almost error free (CLR=10
-6
~10
-4
)
Digital television Almost error free (after FEC)
Terrestrial/cable/satellite TV Almost error free (depend on weather)
Video phone over the Internet (H.323) BER = 0, Packet loss of 0~30%
Mobile video phone (H.324 wireless) BER = 10
-5
~10
-3
, burst errors

To enhance the video quality in presence of unavoidable packet loss or bit error,
error control mechanisms should be used.
3) Heterogeneity problems
Before addressing the heterogeneity problems, we first compare unicast with
multicast. The unicast delivers streaming media through point-to-point transmission,
where only one sender and one receiver are involved. In contrast, multicast delivers
streaming media through point-to- multipoint transmission, where one sender and multiple
receivers are involved. For streaming applications such as video conferencing and
Internet television, multicast delivery can achieve high bandwidth efficiency since the
receivers can share links. On the other hand, unicast delivery of such applications is
inefficient in terms of bandwidth utilization. An example is given in Figure 1-1, where,
for unicast, five copies of the video content flow across Link 1 and three copies flow
across Link 2 as shown in Figure 1-1 (a); For multicast in Figure 1-1 (b), there is only one
copy of the video content traversing any link in the network, resulting in substantial
bandwidth savings. However, the efficiency of multicast is achieved at the cost of losing
the service flexibility of unicast (i.e., in unicast, each receiver can individually negotiate
service parameters with the source). Such lack of flexibility in multicast can be
problematic in a heterogeneous network environment. For example, the receivers in
Figure 1-1 (b) may attempt to request different video quality with different bandwidth.
5
But only one copy of the video content is sent out from the source. As a result, all the
receivers have to receive the same video content with the same quality.


Figure 1-1 Unicast and Multicast video distribution
In a public land mobile network, there are two kinds of heterogeneity, namely,
network heterogeneity and receiver heterogeneity [2]. Network heterogeneity refers to the
different domains (e.g., wireless domain and wired domain) having unevenly distributed
resources (e.g., processing, bandwidth, storage, and congestion control policies). Network
heterogeneity can make different users experience different packet loss/delay
characteristics. Receiver heterogeneity means that receivers have different or even
varying latency requirements, visual quality requirements, and/or processing capability.
It is a challenge to design a multicast mechanism that not only achieves efficiency in
network bandwidth, but also meets the various requirements of the receivers.
6
2. Handoff design issues in media delivery
The problem of handoff in the wireless network is well-know, however it is largely
unexplored in the applications of streaming media. There are a serial of problems or
requirements associated with media streaming during seamless handoff, such as handoff
latency (or media stream interruption), end-to-end delay (or service delivery time), media
synchronization and Handoff scalability. However, the handoff procedures in UMTS R99
and Release 4 [57] [58] [59]may not satisfy the requirements of seamless handoff for
media streaming services. The handoff design issues of media streaming will be
discussed in details in Chapter 2.
To address the handoff problems in media streaming, a Scalable Multiple
Description Coding framework is proposed together with distributed video storage in the
DiffServ mobile network to support streaming video handoff. It leverages the distributed
multimedia delivery mobile network to provide path diversity to combat outage due to
handoff. Since the media streaming services are pushed to the edge of core network so
that the streaming media is sent over a shorter network path, it also reduces the media
service delivery time, the probability of packet loss, and the total network resource
occupation with relatively consistent QoS in all scenarios.
3. Error resilience enhancement for MPEG-4 video
To further explore the error resilience and concealment tools in MPEG-4, the shape,
motion, and texture information in the bit-stream of an object-based video are re-
organized into different layers in the proposed SMDC scheme to support the
classification and priority assignment in the DiffServ network. Moreover, due to the joint
design of LC and MDC, it is possible to overcome the drawbacks of LC and MDC
4. UMTS QoS and IP DiffServ
It is challenging to provide QoS attribute translation and mapping between the IP
networks and the UMTS systems and to implement the IP differentiated services for the
traffic encapsulated and isolated by tunneling in UMTS. In order to support the unequal
error protection for layered video, a UMTS-to-DiffServ QoS mapping scheme and its
marking algorithm for MPEG-4 scalable video are proposed in the thesis. Furthermore, it
spurs the evolution of UMTS toward its final all-IP phase for the purpose of addressing
the DiffServ tunneling issue in UMTS.
7
This thesis studies the architecture of third generation mobile networks and the
handoff procedures for video delivery in UMTS. In order to address the handoff issues in
video streaming, as well as the bandwidth fluctuation, packet loss and heterogeneity
problems in the wireless networks, and to further enhance the error resilience tools in
MPEG-4, a scalable multiple description coding framework together with a distributed
multimedia delivery mobile network is proposed. The corresponding intra-RAN handoff
and inter-RAN handoff procedures in D-MDMN are studied. Furthermore, a new IP
DiffServ video marking algorithm is explored to support the UEP of SMDC. Simulation
results show that the proposed scheme achieves performance improvements compared
with the original UMTS and DVMA solutions.
1.2 Thesis Organization
The remainder of this thesis is organized as follows.
Chapter 2 overviews and discusses the General Packet Radio Service (GPRS) and
UMTS mobile network architecture, handoff protocols and handoff problems for media
streaming.
In Chapter 3, the concepts of layered coding and multiple description coding are
introduced, followed by the proposed mobile system model (i.e., the distributed
multimedia delivery mobile network) and the protocol stack of network-aware end
system.
Chapter 4 presents the details of the proposed scalable multiple description coding,
the corresponding handoff procedures in the D-MDMN and a novel IP DiffServ MPEG-4
video marking algorithm.
The simulation models, system setup, test conditions, and simulation results are
presented and analyzed in Chapter 5.
Finally, Chapter 6 gives conclusions of this work and suggestions for further
research.
8
2 Handoff Design Issues in Media Delivery
Handoff is a basic mobile network capability for dynamic support of terminal
migration. In order to illustrate the proposed solution to solve the handoff problems of
media streaming, the GPRS/UMTS mobile systems and handoff protocols are first
studied.
2.1 GPRS Network Architecture
The GPRS network configuration is outlined in TS 23.002 [60] [64] and illustrated
in Figure 2-1.


Figure 2-1 GPRS Network Architecture
Current services (voice and circuit-switched data) are supported via the base station
subsystem (BSS) and network subsystem (NSS). The BSS consists of the base transceiver
station (BTS) that handles the radio physical layer and the base station controller (BSC)
9
that deals with radio resource management and handover. The NSS for circuit-switched
services consists of the mobile switching center (MSC), the visitor location register
(VLR) integrated in the MSC, and the home location register (HLR).
GPRS provides packet-switched services over the Global System for Mobile
communications (GSM) radio. The major new element introduced by GPRS is an NSS
(GPRS backbone) that processes all the data traffic. It comprises two network elements:
Serving GPRS support node (SGSN), which keeps track of the location of
individual mobile stations and performs security functions and access control.
Gateway GPRS support node (GGSN), which encapsulates packets received
from external packet networks (Internet) and routes them toward the SGSN.
The interface between the BSS and the SGSN is based on the frame relay transport
protocol. The SGSN and GGSN are interconnected via an IP network. No layer-two
technology has been specified.
2.2 3GPP UMTS Network Architecture
The telecommunication system standardised by the Third Generation Partnership
Project (3GPP) consists of a core network and a radio access network that may be either
GERAN or UMTS Terrestrial Radio Access Network (UTRAN), or both. The UMTS
network [60] [64], shown in Figure 2-2, consists of two independent subsystems
connected over a standard interface:
Radio access network, which may be either GSM/EDGE radio access
network (GERAN) or UMTS terrestrial radio access network (UTRAN), or
both. UTRAN composes of node B and a radio network controller (RNC).
Node B is functionally similar to the GSM BTS, and RNC is similar to the
GSM BSC.
UMTS core network (CN), which is equivalent to the GSM/GPRS NSS.
The separation of the radio access network (RAN) from the core network is the
fundamental concept of the cellular system.
The UMTS core network reuses as much as possible the GSM/GPRS NSS:
10
Packet switched (PS): an evolution of the GPRS SGSN/GGSN with a more
optimized functional split between the UTRAN and core network.
Circuit switched (CS): an evolution of the NSS with the transcoder function
moved from the BSS to the core network.


Figure 2-2 UMTS Network Architecture
The UTRAN consists of several possibly interconnected radio network subsystems
(RNSs). An RNS contains one RNC and at least one node B. The RNC is in charge of the
overall control of logical resources provided by the node Bs. RNCs can be interconnected
in the UTRAN (i.e., an RNC can use resources controlled by another RNC). In case of a
WCDMA RAN, the RNC provides soft handover, combining and splitting between
streams from different base stations belonging to the same mobile station.
Node B provides logical resources, corresponding to the resources of one or more
cells, to the RNC. It is responsible for radio transmission and reception in the cells
maintained by this node B. A node B controls several cells.
11
2.3 UMTS Handoff Procedures for Media Streaming
2.3.1 Assumptions
Before the presentation of the handoff procedures in UMTS Release 4 for media
streaming, the following assumptions are made [52].
1. The handoff procedures are designed specially for media streaming services in
packet switched service domain. The QoS attributes required by audio and
video media streams are defined in 3GPP TS 23.107 V5.6.0 [61], as discussed
in Section 2.4.1.
2. This research focuses on the hard handoff procedures. Soft handoff may provide
better performance for media streaming. However, hard handoff is required
when there are no connections between source RNC and target RNC within the
mobile network, especially under the consideration of network heterogeneity
and receiver heterogeneity (such as interworking between UTRAN and GERAN
or UTRAN and IEEE 802.11 (WLAN)).
3. The handoff procedures are instantiated in GPRS. However, they are also
applicable to either GSM (with the SGSN/GGSN being replaced by
MSC/GMSC), or UMTS (with the BSC/BTS being replaced by RNC/Node B).
4. Mobile assisted handoff is adopted. That is, the mobile station (MS) assists the
network by taking periodic measurements on the downlink and relaying them
back to the network for handoff decision making.
5. The inherent GPRS tunneling protocol (GTP) of GSM/GPRS/UMTS [51] [70]
is used to support mobility.
6. Based on the measurements on both downlink and uplink which are performed
by MS and BTS/Node B respectively, the BSC/RNC makes handoff decisions.
7. In both int ra-RAN and inter-RAN handoffs, the MSC/SGSN determines the
readiness of the new access point to accommodate the handoff; in inter-cell,
intra-RNS handoff, the BSC/RNC does so; while in intra-cell handoff, the
BTS/Node B does so.
12
8. For media streaming, the BTS/Node B ensures that the handoff algorithm
maintains packet sequencing after handoff.
9. The handoff procedures in UMTS Release 4 are presented as follows only in the
scenarios of intra-RAN handoff and inter-RAN handoff. The details of other
scenarios, such as inter-cell, intra-RNS handoff, and intra-cell handoff are
similar and will not be repeated in this section.
2.3.2 Handoff Procedures in UMTS Release 4
2.3.2.1 Intra-RAN Handoff Procedure


Figure 2-3 UMTS Rel4 network model of intra-RAN handoff (Data plane)
Figure 2-3 illustrates a typical UMTS network model under the IP transport mode in
the scenario of intra-RAN handoff. In UMTS Release 4, the media service provider is
13
outside the CN which consists of SGSNs and GGSN, and far from the MS. The media
streams should first get through CN and then can feed into the RNS A or B.
In the IP transport mode, the destination IP address of an end-user packet is not used
to make the packet forwarding decision. Instead, the packets are encapsulated in an
intermediate layer (e.g., frame relay transport protocol (FP) in the RAN and GTP in the
CN, which may be specific to the chosen wireless technology. The encapsulated data
units are then transported, between the nodes in the segment, over another IP layer. Most
of the existing proposals espouse this approach, which allows the mobile operator to keep
many of the legacy components of the 2G network untouched while upgrading just the
transport layer from point-to-point lines or an ATM network to an IP-based network.
The control plane of handoff procedure [57] [58] [59] consists of three phases, as
shown in Figure 2-4, while the data plane of these three phases is shown in Figure 2-3.
l Phase I: Preparation of RNS handoff and resource allocation
The MS sends its periodic measurement reports (signal #1). Based on these reports
and its own measurement and on current traffic conditions, the source RNS (sRNS) makes
the decision to perform a handoff and sends an HO-required message (signal #2) to inform
the SGSN about the identifier of the target RNS (tRNS) to which the MS attempts to make
a handoff. The SGSN then shall generate an HO-request message (signal #3) to the selected
tRNS and requests the allocation of resources for the MS. The tRNS checks if enough radio
resources are available and activates a physical channel at the tRNS to prepare for the
arrival of the MS. Once resource allocation has been completed by the tRNS, it shall return
an HO-request-ack message (signal #4) to the SGSN. When this message is received by
the SGSN, it starts to set up a link (i.e., GTP tunnel) to the tRNS, indicates the completion
of the preparation phase on the core network side for the handoff by sending an HO-
command message (signal #5) to the sRNS.
Note that: The HO-request-ack (signal #4) from the tRNS contains the complete
radio interface message that shall be sent by the sRNS to the MS in the HO-command
(signal #6), the SGSN transparently passes this radio interface message onto the sRNS.
For the data plane of handoff phase I, upon receiving the signal #5 at the end of the
preparation phase, the sRNS stops transmitting downlink data to the MS and should store
all downlink data which continue to arrive from the SGSN to the sRNC.
14

Figure 2-4 Intra-RAN handoff procedure in UMTS Rel4 (Control plane)
l Phase II: Moving the serving RNS role to target RNS
On receipt of the HO-command (signal #5), the sRNS will issue the radio interface
message HO-command (signal #6), containing a Handover Reference Number previously
allocated by tRNS, to the MS. The MS will then break its old radio link and access the
new radio resource using the Handover Reference Number contained in the HO-access
message (signal #7). The number will be checked by the tRNS to ensure that it is as
expected and that the correct MS has been captured. If this is the correct MS, the tRNS
shall send an HO-detect message (signal #8) to the SGSN.
15
For the data plane of handoff phase II, upon receiving the signal #5 in Phrase I and
at the beginning of the execution phase, the sRNS starts to forward all the buffered data
(including state information for session migration) to the tRNS, via SGSN.
Once data forwarding is started, the tRNS stores all GTP Protocol Data Unit (GTP-
PDUs) forwarded from the sRNS. When Serving RNS operation is initiated, the tRNS
starts the downlink data processing and transmission from the first forwarded GTP-PDU.
After the GTP tunnel is created between the tRNS and the SGSN, the uplink flow is
switched from the old path to the new path.
l Phase III: Releasing resource reservation in the old path
For correct resequencing, the sRNS and the SGSN should forward the Sequence
Number information respectively to the tRNS as defined in Release 99, so that the tRNS
can judge whether the data forwarding has been completed or not. This requires
triggering the GTP Sequence Number field in the GTP header for each video packet.
When the MS is successfully communicating with the tRNS and after the data
forwarding is complete, an HO-complete message (signal #9) will be sent by the MS to
the tRNS. The tRNS will then send an HO-complete message (signal #10) to the SGSN.
After the SGSN has received the signal #10 from the tRNS, it shall begin to release the
resources reserved on sRNS for the MS in the old path. In Figure 2-4 the resources are
released by using the Clear-command message (signal #11) and Clear-complete message
(signal #12).
On the data plane of handoff phase III, at the beginning of the releasing phase, the
downlink media flow is redirected from the old path to the new path. The tRNS should
store the redirected data until the transmission of all the forwarded data to the MS is
completed, such that the correct packet sequencing can be ensured. The functionality of
resequencing is implemented in the BSs.
2.3.2.2 Inter-RAN Handoff Procedure
Figure 2-5 illustrates a typical UMTS Release 4 network model under the IP
transport mode in the scenario of inter-RAN handoff. Similarly, the media service
provider is outside the CN which consists of SGSNs and GGSN, and far from the MS.
The media streams should first get through CN and then feed into the Domain A or B.
16
Note that the handoff latency gets worse since the data forwarding has to take place
between the RNSs which are separated by a transport network.


Figure 2-5 UMTS Rel4 network model of inter-RAN handoff (Data plane)
The control plane of handoff procedure [57] [58] [59] also consists of three phases,
as shown in Figure 2-6 and Figure 2-7, while the data plane of these three phases is
shown in Figure 2-5. The differences are described as follows in contrast with the intra-
RAN handoff.
l Phase I: Preparation of RNS handoff and resource allocation
The sRNS in RAN A informs the target SGSN (tSGSN) about the MS which
attempts to make a handoff to the tRNS in RAN B. The source SGSN (sSGSN) sets up a
link (i.e., GTP tunnel) to the tRNS through the tSGSN, and requests the allocation of
resources for the MS.
17
For the data plane of handoff phase I, at the end of the preparation phase, the sRNS
stops transmitting downlink data to MS and should store all downlink data which
continue to arrive from the sSGSN to the source RNC.


Figure 2-6 Inter-RAN handoff procedure in UMTS (Control plane: Phase I)
l Phase II: Moving the serving RNS role to target RNS
Under the handoff command from the sRNS, the MS access to the tRNS.
Meanwhile, the downlink and uplink GTP tunnel is updated between the tSGSN and the
GGSN through Update-PDP-context-request message (signal #12) and Update-PDP-
context-response message (signal #13), so that the downlink and uplink flow can use the
new route in the next phase.
On the data plane of handoff phase II, at the beginning of the execution phase, the
sRNS starts to forward all the buffered data including state information for session
migration to the tRNS, via a GTP tunnel between the RNSs. When data forwarding starts,
the tRNS stores all GTP-PDUs forwarded from sRNS. When serving RNS operation is
initiated, the tRNS starts the downlink data processing and transmission from the first
18
forwarded GTP-PDU. After updating of the uplink GTP tunnel, the uplink flow is
switched from the old path to the new path.


Figure 2-7 Inter-RAN handoff procedure in UMTS (Control plane: Phase II&III)
l Phase III: Switching of downlink flow in CN
The handoff is completed and the sRNS then releases the resources reserved for the
MS in the old path.
For the data plane of handoff phase III, at the beginning of the path-optimization
phase, the downlink media flow is redirected from the old path to the new path. The
tRNS should store the redirected data until the transmission of all the forwarded data to
the MS is completed, such that the correct packet sequencing can be ensured.
For correct resequencing, the sRNS and the GGSN should forward the Sequence
Number information respectively to the tRNS as defined in Release 99, so that tRNS can
19
judge whether the data forwarding has been completed or not. This requires triggering the
GTP Sequence Number field in the GTP header for each video packet.
Note that: The mechanism shown in Figure 2-5 assumes that the downlink GTP port
used for a given media stream in tRNS is the same for all arriving GTP-PDUs regardless
of their arrival routes.
2.4 Handoff problems in GPRS/UMTS
There are several problems or requirements as follows associated with media
streaming during handoff.
2.4.1 Bearer Service QoS and Seamless Handoff
Handoff management for streaming applications is the process of initiating and
ensuring a seamless handoff, in which the radio access network changes the radio
transmitters or radio access mode or radio system used to provide the bearer services,
while maintaining a defined bearer service QoS. Seamless handoff means a handoff
without perceptible interruption of the radio connection according to the definition given
in 3GPP in [37]. For seamless handoff, it assumes that there is no need to buffer any
downlink or uplink traffic in the involved nodes considering that packet loss (or frame
loss) is tolerated to some degree in the streaming application.
Because of the limitation of the cost and the physical size, a mobile handset
generally can afford only limited buffer size. It is liable to extend the playout buffer in
base stations. Also due to the mismatch between high transmission rate over wired links
and low transmission rate over wireless links, the packet buffer in base stations can be
used for rate matching and packet resequencing. When handoffs occur, the buffered data
are then forwarded from the sRNS buffers to the tRNS.
In addition, packet loss can already occur over the radio or due to congestion in the
wired link. From Table 1-1, the BER of wireless video can be up to 10
-3
. Therefore any
packet loss due to handoff is in addition to the packets lost over the radio or in the wired
link.
20
In order to maintaining a defined bearer service QoS (BER10
-3
) in Table 2-1, a
data buffering and forwarding mechanism in UMTS R99 should be reused also in UMTS
Release 4 for streaming services requiring seamless handoff.
The values in Table 2-1 are indicative of the QoS attributes required by audio and
video media streams in 3GPP [61], including BER and frame erasure rates (FER), for the
Adaptive Multi Rate (AMR) speech codec and the MPEG-4 video codec as examples.
However, the handoff procedures in UMTS R99 and Release 4 under above
considerations may not satisfy the requirements of transfer delay, media stream
interruption, signalling traffic and scalability, as follows.
Table 2-1 QoS attributes required by audio and video media streams in 3GPP
Type of
payload
QoS attributes
Bit rate: 4.75 ~ 12.2 kbit/s
Delay: end-to-end delay not to exceed 100ms (codec frame length is 20ms)
BER: 10
-4
for Class 1 bits
10
-3
for Class 2 bits
For some applications, a higher BER class (~10
-2
) might be feasible.
AMR
speech
codec
payload
FER < 0.5% (with graceful degradation for higher erasure rates)
Bit rate: variable, average rate scalable from 24 to 128 kbit/s and higher
Delay: end-to-end delay between 150 and 400ms
video codec delay is typically less than 200 ms
BER: 10
-6
- no visible degradation
10
-5
- little visible degradation
10
-4
- some visible artefacts
10
-3
- limited practical application
MPEG-
4 video
payload
Packet loss rate: for further study
21
2.4.2 Transfer Delay
Transfer delay (i.e., end-to-end delay of media service delivery) is used to specify
the delay tolerated by the media application. It allows UTRAN to set transport formats
and Automatic Repeat reQuest (ARQ) parameters. For interactive media services, it
should be minimized. For streaming audio/video, the transfer delay depends on the
playout buffer length in the Internet. However, as presented in Table 2-1, the transfer
delay of UMTS bearer service was bounded stringently in 3GPP. In the UMTS network
model, shown in Figure 2-5, the media providers are separated from the UTRAN by the
CN. The media streams should first get through CN and then feed into the Domain A or
B. Thus, the transfer delay requirement may not be satisfied under the current model and
should be studied further.
2.4.3 Handoff Latency (Media Stream Interruption)
Handoff latency is defined as time between the last packet transmitted from the old
base station (BS) and the first packet transmitted from the new BS. As mentioned
previously, buffered data forwarding during seamless handoff may result in relatively
large handoff latency (or media stream interruption) and a large amount of additional
traffic.
In the case of inter-RAN handoff, for downlink media streams, there are two
possible situations wheren media stream gap or overlapping may happen:
1. The media stream overlap/gap may be introduced when tRNS takes the serving
RNS role and starts to produce the downlink data from forwarded GTP-PDUs.
In this case the estimated gap/overlap for hard handoff is equal to the delay of the
GTP tunnel used for data forwarding. This first instance of media stream overlap
coincides with radio hard handoff.
If the transport bearer delay difference is smaller than the air interface Transmission
Time Interval (TTI) (10, 20, 40 or 80 ms depending on the service), the amount of
gap/overlap most likely does not exist.
22
2. The additional media stream gap may be introduced when the CN transport is
optimized.
In this case the gap will exist only if the delay via the optimized route is larger than
the delay via the forwarding route.
The above two types of media stream overlap/gap can also happen during a soft
handoff. Due to the existence of the Iur interface (i.e., the interface between sRNC and
tRNC), the first type of overlap/gap during soft handoff should be smaller in theory.
However, note that the Iur interface is only logical interface, which may be provided via a
transport network. Thus, in the real world, the handoff latency may be worse due to the
introduction of the transport network between RNCs.
The effects of media stream interruption get worse when the network heterogeneity
and the receiver heterogeneity are considered. Usually, video transcoders [33] have to be
deployed at the edge of networks, such as at base stations, in order to address the
heterogeneity problems. However, it incurs the extra state information migration besides
data forwarding and the temporal dependence issue because of the nature of predictive
encoding and motion compensation.
Temporal Dependence


Figure 2-8 Temporal dependency in a MPEG-based coding stream
23
MPEG coding is based on both inter- frame and intra- frame coding, which produces
three types of compressed frames, shown in Figure 2-8 as a group of pictures (GOP) in a
frame-based MPEG coding stream:
l I-frame: Intra-coded frame, coded independently of all other frames.
l P-frame: Predictively coded frame, coded with reference to a previous I-frame or
P-frame.
l B-frame: Bi-directionally predicted frame, coded with reference to both previous
I-frame and future P- frame.
During the period of combing a group of blocks (GOB) into a medium unit, the
packet loss caused by handoff makes the combination and decompression incomplete if
the medium unit is split into different transmitted packets. If the information- lost part due
to handoff belongs to I- frame I
0
, and then the whole media sequence becomes erroneous,
which results in a period of dummy video accompanied with some strange audio.
Similarly, if the information- lost part due to handoff belongs to P- frame P
6
, B- frames B
4
,
B
5
, B
7
, B
8
will be erroneously decoded. However, the damage of B- frames does not affect
the decoding of any other-type frames in the GOP. Thus, the incomplete combination of a
medium unit makes a multimedia presentation interrupted, asynchronous and
discontinuous, and results in larger handoff latency.
State Information Migration
Typically, three types of state information should be reliably migrated for a
successful handoff of a media stream:
1) The session description:
The session description is typically present at the beginning of the video stream as
part of specific header information. It is possible to cache it on the old BS and transfer it
from there to the new BS. Sometimes this information is provided as part of the control
handshake, for example, using the Session Description Protocol (SDP).
2) The session parameters at the handoff decision point:
The session parameters at the handoff decision point include specifications from the
clients request, the current position (offset) in the video stream, and some way of locating
the video object, e.g., via a Uniform Resource Locator (URL). The media server should
preferably have the ability to seek a specified stream offset.
24
3) The transcoding state information:
There are two types of state information associated with the transcoding process:
reconstructible state information and dependent state information. Reconstructible state
information can be recreated given an input stream. It consists of reference frame data
(both original and down-sampled ones) and macroblock- level side information.
Dependent state information includes data derived from the output stream. For example,
the rate control module takes the number of bits consumed so far to evaluate the bit
budget that can be allocated for the next coding unit. The volume of reconstructible state
information is usually much larger than that of the dependent state information. In
general, a transcoder needs to maintain and communicate at least dependent state
information for a session handoff since the output stream is not shared between the
transcoders in different BSs. Table 2-2 [10] gives the amount of data required for the
transcoding state transfer, where CCIR refers to Consultative Committee for International
Radiocommunication, CIF refers to Common Interleaved Frame, and QCIF refers to
Quarter Common Interleaved Frame.
Table 2-2 The required amount of transcoding state information to be transferred
Total Transferred*
Source
Format
Resolution
(pixels)
Reference
Frame Data
(bytes)
Macroblock- level
side information
(bytes)
CIF
(bytes)
QCIF
(bytes)
CCIR 601 720 480 2,073,600 648,000 3,352,064 2,895,872
CIF 352 288 608,256 190,080 - 972,608
QCIF 176 144 152,064 47,520 - -
* The total transferred includes fixed overheads.
The data forwarding has to wait until the migration of the above state information
totally completes, which results in a larger handoff latency.
The handoff protocols in UMTS R99/Rel 4 are based on measuring the signals
quality to determine the time and place for initiating the handoff procedures, which do
not consider the temporal dependence issue and the extra state information migration.
25
In [37], the concept of glue-point was proposed for resolving the temporal
dependence issue. Handoff can only occur at glue-point so that the transcoding state
information can be minimized. The glue-point delimits the boundary of two consecutive
medium units (e.g., GOB in H.263 or Video Object Planes (VOP) in MPEG-4) or some
type of boundary that is relatively less temporally dependent.
However, this solution will not work if the mobile terminal suddenly loses contact
with the current base station before glue-point arrives, due to deep deterioration of the
wireless channel condition and the high speed of the mobile user.
2.4.4 Media Synchronization
Media synchronization [5] refers to maintaining the temporal relationships within
one data stream and between various media streams. There are three levels of
synchronization, namely, intra-stream, inter-stream, and inter-object synchronization. The
three levels of synchronization correspond to three semantic layers of multimedia data as
follows [1].
1) Intra-stream synchronization
The unit of the compression layer of the MPEG encoder is a logical data unit such as
a video/audio frame, which adheres to strict temporal constraints to ensure acceptable
user perception at playback. Synchronization at this layer is referred to as intra-stream
synchronization, which maintains the continuity of logical data units. Without intra-
stream synchronization, the presentation of the stream may be interrupted by pauses or
gaps.
2) Inter-stream synchronization
The unit of the synchronization layer is of the MPEG encoder a whole stream.
Synchronization at this layer is referred to as inter-stream synchronization, which
maintains temporal relationships among different continuous media. Without inter-stream
synchronization, skew between the streams may become intolerable. For example, users
could be annoyed if they notice that the movements of the lips of a speaker do not
correspond to the presented audio (lip synchronazition).

26
3) Inter-object synchronization
In MPEG-4, the spatio-temporal location of audio-visual objects is defined by scene
description using a tree-based structure in the Compress Layer. Synchronization
according to this tree-based structure is referred to as inter-object synchronization. The
objective of inter-object synchronization is to start and stop the presentation of the time-
independent data within a tolerable time interval, if some previously defined points of the
presentation of a time-dependent media object are reached. Without inter-object
synchronization, for example, the audience of a slide show could be annoyed if the audio
is commenting one slide while another slide is being presented.
The media stream interruption introduced due to handoff is typically unpredictable.
The incurred delays and delay variations can disrupt intra- media, inter- media, and inter-
object synchronization. Therefore, media synchronization mechanisms are required to
ensure proper rendering of the multimedia presentation at the client.
For the purpose of media synchronization, the playout buffer in either BS or MS is
deployed to eliminate the side effects that result from the wired network jitter or handoff
latency. Due to the limitation of playout buffer, however, the buffered media units may
be used up if the wired network jitter or handoff latency exceeds the expected value. The
synchronization strategies thereby are proposed as follows when the expected media units
are not available at the expected presentation time.
Nonblocking Strategy
If an expected video unit does not arrive at the expected time and the playout buffer
in MS is not empty, the expected one is considered lost and thus ignored; continues to the
next one. If an expected video unit does not arrive at the expected time and the playout
buffer in MS is empty, the most recently displayed video unit is repeated until the video
stream is available again.
Blocking Strategy
If an expected audio unit does not arrive at the expected time and the playout buffer
in MS is not empty, the expected one is considered lost and thus ignored; continues to the
next one. If an expected audio unit does not arrive at the expected time and the playout
buffer in MS is empty, block the current presentation until the audio stream is available
again.
27
2.4.5 Handoff Scalability
Typically, there are four possible handoff scenarios [55] [68] in GPRS/UMTS, as
illustrated in Figure 2-9.


Figure 2-9 Types of handoff in GSM/GPRS/UMTS
1. Inter-RAN handoff: The calls are transferred between two cells belonging to
different MSCs/SGSNs. Both MSCs/SGSNs perform the handover together.
2. Intra-RAN handoff: The calls are transferred between two cells belonging to
different RNSs with the same MSC/SGSN. This handoff then has to be
controlled by the MSC/SGSN.
3. Inter-cell, intra-RNS handoff: The calls are transferred between two cells but
stays within the control of the same BSC/RNC. The BSC/RNC then performs a
handoff, assigns a new radio channel in the new cell, and releases the old one.
4. Intra-cell handoff: The calls are transferred within the same cell. The BSC/RNC
makes the handoff decision.

28
In UMTS, the types of handoff can also be classified into:
1. Inter-system handoff, between cells belonging to different radio access
technologies (e.g., UMTS and GSM/EDGE) or different radio access modes
(e.g., FDD/WCDMA and TDD/TD-CDMA).
2. Intra-system handoff, which can be further subdivided into:
l Intra- frequency handoff, between cells belonging to the same WCDMA carrier;
l Inter- frequency handoff, between cells operating on different WCDMA carriers.
In addition, UMTS supports both hard handoff and soft handoff. The soft handoff is
fully performed within UTRAN, without involving the core network due to the existence
of Iur interface. The hard handoff may be also performed within UTRAN or GERAN, or
between GERAN and UTRAN, or the core network may be involved if the Iur or Iur- g
interface between RNSs does not exist. Note that the Iur interface is only logical
interface, which may be provided via a transport network. Thus, in the real world, the soft
handoff latency may be worse due to the introduction of the transport network between
RNCs.
Usually for streaming video services, video sessions are long-term sessions. During
a long-term video presentation, mobile users may have to experience all or parts of the
above scenarios. It is important to maintain relatively consistent handoff latency (delay
jitter) in all the above scenarios so that there are no perceptible video or audio quality
fluctuations during such a presentation. The handoff procedures are supposed to be
specially designed to meet this challenge.
In summary, this chapter overviews the GPRS/UMTS mobile network architecture
and the corresponding handoff protocols. The problems and requirements in UMTS
associated with media streaming during handoff, such as bearer service QoS requirement
and seamless handoff, transfer delay, handoff latency, media synchronization and handoff
scalability are discussed in details.
29
3 System Model
Before the description of the proposed mobile system model and the protocol stack
of network-aware end system, the concepts of layered coding and multiple description
coding will be introduced first.
3.1 Layered Coding
From a video-coding point-of- view, scalability plays a crucial role in delivering the
best possible video quality over unpredictable best-effort networks or time- varying
wireless channels. From a networking point-of-view, scalability is needed to enable a
large number of users to view any desired video stream, at anytime, and from anywhere.
So far, layered coding with unequal error protection is the most popular and effective
scheme for facilitating error resilience in a video transport system.


Figure 3-1 Block diagram of layered coding with transport prioritization
Principle of layered coding
Layered coding produces a hierarchy of bitstreams, where the first or base layer is
coded independently at a coarser but acceptable level of quality, and subsequent
enhancement layers are coded dependently. Each enhancement layer of the hierarchy can
30
increase the frequency, spatial, and temporal resolution over that of the previous layer
and incrementally improve the quality. Figure 3-1 shows the block diagram of a generic
two- layer coding and transport system.
Furthermore, layered coding has inherent error-resilience benefits, particularly when
the base- layer bitstream can be transmitted with higher priority, guaranteeing a basic
quality of service, and the enhancement- layer bitstreams can be transmitted with lower
priorities, refining the quality of service. This approach is commonly referred to as
layered coding with transport prioritization. By itself, layered coding is a way to enable
users with different bandwidth capacity or decoding powers to access the same video at
different quality levels. Therefore, layered coding is also called Scalable Coding.
Unequal Error Protection will be discussed later.
Implementation mechanisms
Basically, there are four scalable mechanisms (data partitioning, temporal
scalability, SNR scalability, and spatial scalability) depending on the way the video
information is partitioned.
1) Data partitioning (Frequency domain partitioning)
In transform or subband based coding, the coder can include the low- frequency
coefficients or low-frequency band subsignals in the base layer while leaving the high-
frequency signal in the enhancement layer.
2) Temporal scalability (Spatial resolution refinement)
Temporal scalability is a technique to code a video sequence into two layers at the
same spatial resolution, but different frame rates. The base layer is coded at a lower frame
rate. The enhancement layer provides the missing frames to form a video with a higher
frame rate. Coding efficiency of temporal scalability is high and very close to nonscalable
coding.
3) Signal- to-noise ratio (SNR) scalability (Successive amplitude refinement)
SNR scalability is a technique to code a video sequence into two layers at the same
frame rate and the same spatial resolution, but different quantization accuracy. The base
layer can also encode the DCT coefficients of each block with a coarser quantizer,
leaving the fine details (the error between the original and the coarsely quantized value)
to be specified in the enhancement layer. A higher accuracy DCT coefficient is obtained
31
by adding the base- layer reconstructed DCT coefficient and the enhancement- layer DCT
residue.
4) Spatial scalability (Spatial resolution refinement)
Spatial scalability is a technique to code a video sequence into two layers at the same
frame rate, but different spatial resolutions. The base layer is coded at a lower spatial
resolution and the enhancement layers contain additional information for obtaining higher
spatial resolution. At the decoder, the reconstructed base-layer picture is up-sampled to
form the prediction for the high-resolution picture in the enhancement layer.
Fine granularity scalability (FGS)
To provide more flexibility in meeting different demands of streaming (e.g.,
different access link bandwidths and different latency requirements), a new scalable
coding mechanism, called fine granularity scalability (FGS), was proposed to MPEG-4
[6] [7] [8]. An FGS encoder also compresses a raw video sequence into two substreams,
i.e., a base layer bit-stream and an enhancement bit-stream. Different from an SNR-
scalable encoder, an FGS encoder uses bitplane coding to represent the enhancement
stream. With bitplane coding, an FGS encoder is capable of achieving continuous rate
control for the enhancement stream. This is because the enhancement bit stream can be
truncated anywhere to achieve the target bit-rate.
A variation of FGS is Progressive Fine Granularity Scalability (PFGS) [17] which
is developed by Microsoft Research. PFGS shares the good features of FGS, such as fine
granularity bit-rate scalability and error resilience. The essential difference between FGS
and PFGS is that FGS only uses the base layer as a reference for motion prediction while
PFGS uses multiple layers as references to reduce the prediction error, resulting in a
higher coding efficiency. PFGS is adopted in our SMDC approach.
Unequal Error Protection
To serve as an error resilient tool, layered coding must be paired with UEP in the
transport system, so that the base layer is protected more strongly, e.g., by assigning a
more reliable sub-channel, using stronger FEC codes [21] [53] , or allowing more
retransmissions.
Different networks may implement transport prioritization using different means. In
ATM networks, there is one bit (CLP) in the ATM cell header that signals the cell loss
32
priority. When traffic congestion occurs, a network node can choose to discard the cells
having low priority first. Transport prioritization can also be implemented by using
different levels of power to transmit the substreams in a wireless transmission
environment. Also, DiffServ will be well suitable for prioritized transmission.
Advantages
1) It is highly adaptable to unpredictable bandwidth fluctuation due to dynamic
changes in network conditions.
2) It provides the flexibility to combat both network heterogeneity and receiver
heterogeneity. The scalable source coding leaves the media servers and proxies
(or gateways) an opportunity to trim the video stream to appropriate bit rate
before transmission or relay.
3) It is applicable to both unicast and multicast.
4) It provides error resilience through UEP, which is a nice match for future
DiffServ networks.
5) There is no feedback channel requirement and therefore lower delay.
6) It is well suitable to combine with an encryption mechanism to support multiple
levels of security for intellectual property protection [32].
Disadvantages
1) It will lead to a disastrous effect in the decoded visual quality or even break
down if a loss is in the base layer or the channel of the base layer fails.
The current Internet can not support priority service and has to rely on the
conventional way (e.g., FEC or retransmission) to realize error- free transmission of the
base layer. However, FEC-based approaches often suffer from dynamic network
conditions. Retransmission-based approaches are not applicable in streaming applications
when a back-channel is not available or when the transmission delay is not acceptable. In
the case of multicast or broadcast, too much feedback creates a problem, called feedback
implosion [49]. Moreover, the sender can not afford to honor independent
retransmission requests from each receiver.
2) The significant improvement performance of a layered coder over a single- layer
coder in the presence of channel errors is at the cost of a coding overhead.
33
Generally, the four scalability modes in MPEGnamely, data partitioning, temporal
scalability, SNR scalability, and spatial scalabilityhave increasingly better error
robustness in that order, but also an increasing coding overhead. To be more precise, data
partitioning requires the least number of bits (only 1% more bits), while the spatial
scalability has a better reconstructed image when there exist significant losses in the
enhancement layer. SNR scalability and temporal scalability is in the middle on both
scales. Table 3-1 [20] summarizes the required ratio of the base layer to the total bit rate
and the highest packet loss rate at which the video quality is still considered visually
acceptable. These results are obtained by assuming that the base layer is always intact
during the transmission.
Table 3-1 Comparison of different scalability modes in MPEG-2
Coding Mode
Required base layer to total
bit rate ratio
The maximum sustainable
packet loss rate
One layer (MP@ML) 100% 10
-5

Data partitioning 50% 10
-4

*Temporal scalability <50% 10
-4
~10
-3

SNR scalability <20% 10
-3

Spatial scalability <20% 10
-3
~10
-2

* The data of temporal scalability came from theoretical analysis.
3.2 Multiple Description Coding
As described in Section 3.1, if a data network were able to provide a preferential
treatment to the packets of the base layer and transmit them in an error- free channel,
layered coding would be almost the best solution. The Internet, however, usually does
not look inside packets and discriminate; packets are dropped at random (e.g., drop-tail or
RED) when congestion occurs.
The typical way to handle lossless transmission in a lossy network is to invoke many
retransmissions or add a lot of redundancy via strong FEC. In jitter-sensitive streaming
applications, however, it may not be feasible or cost effective. A better alternative is to
34
make do with whatever arrives upon first transmission and combat the transmission error
from the source side. Multiple Description Coding is well competent for this challenge
due to its inherent diversity attribute.
Principle of multiple description coding
As with LC, multiple description coding also codes a source into several sub-
streams, known as descriptions, but the decomposition is such that the resulting
descriptions are correlated and have similar importance. Any single description should
provide a basic level of quality, and more descriptions together will provide improved
quality. Figure 3-2 shows the block diagram of a generic two- layer coding and transport
system.


Figure 3-2 Block diagram of MDC coding and decoding
For each description to provide a certain degree of quality, all the descriptions must
share some fundamental information about the source, and thus must be correlated. This
correlation enables the decoder to estimate a missing description from a received one, and
thus provide an acceptable quality level from any description. On the other hand, this
correlation is also the source of redundancy in MDC.
An advantage of MDC over LC is that it does not require special provisions in the
network to provide a reliable sub-channel. To accomplish their respective goals, LC uses
a hierarchical, decorrelating decomposition, whereas MDC uses a non-hierarchical,
correlating decomposition.
35
Implementation mechanisms
Some approaches [34] [49] that have been proposed for accomplishing such
decomposition include Multiple Description quantization, MDC with correlation
transform, transform domain subsampling, spatial domain subsampling, temporal domain
subsampling (e.g. , Multiple State Recovery (MSR) [46]), and interleaved spatial-
temporal sampling. The last approach is known as video redundancy coding (VRC) in
H.263+ [18]. MSR has excellent capability of error recovery in the presence of packet
loss on all descriptions and is adopted in our SMDC approach.
Advantages
1) Robustness to losses and bit errors
a. Path diversity
MDC directly attacks the problem of communicating the continuous-valued source.
MD coders can be designed with concern for every combination of received descriptions
with an appreciable probability. MDC assumes that there are several parallel channels
between the source and the destination and that each channel may be temporarily down or
suffering from long burst errors. Furthermore, the error events of different channels are
independent, so that the probability that all channels simultaneously experience losses is
small.
Diversity techniques have been studied for many years in the context of wireless
communication, e.g., frequency, time, and spatial diversity. However, the problem of path
diversity over a packet network has been largely unexplored. The recent work [26] adds
justification to our approach for path diversity: in comparing the performance of the
default path between two hosts on the Internet to that of alternative paths between those
two hosts, it is found that in 30-80% of the cases, there is an alternate path with
significantly superior quality. The quality is measured in terms of round-trip-time, loss
rate, and bandwidth. Therefore, diversity would also appear to be beneficial for
communication over the Internet. There are several ways to set up multiple paths or links
for a single virtual connection in a wireless network. In a single-hop wireless network, a
station would need to establish channels to multiple base stations instead of one. This is
already done in soft hand-off systems, during the hand-off phase. In a multiple- hop
adhoc networks, each station has router- like functionality to establish multiple disjoint
36
paths with another wireless station. The Internet Engineering Task Force (IETF) MANET
Working Group has been the main forum for research in this area. Most of the proposed
ad hoc routing protocols have the ability to discover multiple routes. In a CDMA system,
the multiple antenna technique can be employed with the MDC. Each description should
use one transceiver and one antenna in the BS and the MS.
b. Error recovery
Some MDC algorithms have an excellent capability of error recovery in the presence
of packet loss on all descriptions, such as multiple states and state recovery [46].
Due to robustness to losses and bit errors, it is well suitable for the imperfect and
unpredictable channels that have relatively high loss or failure rates, such as streaming
media channels over the current Internet or a wireless network. Also, it should be forward
compatible with the potential DiffServ network.
2) Enhanced quality
If a receiver receives multiple descriptions, it can combine them together to produce
a better reconstruction than that produced from any one of them.
3) Distributed storage
Distributed storage [49] of streaming media matches the MD framework well.
Consider a database of images stored at several locations with MD encoding. A typical
user would have fast access to the local image copies; for higher quality, one or more
remote copies could be retrieved and combined with the local copy.
Distributed storage is common in the use of edge servers for popular content. In
current implementations, identical data is stored at the servers, so there is no advantage in
receiving multiple copies. Storage can also be distributed to make the reliability of each
device less important; lowering reliability requirements can decrease costs.
4) No feedback channel requirement and therefore lower delay.
Disadvantages
1) Low coding efficiency
To guarantee an acceptable quality with a single description, each description must
carry sufficient information of the original signal. This implies that there will be overlap
in the information contained in different descriptions. Obviously, this will reduce the
coding efficiency.
37
The core issue in designing MD coder is the tradeoff between coding efficiency and
redundancy among the descriptions, such that the degradation in quality in the event of
failures is graceful. A mechanism [19] is required to adapt the amount of redundancy
added to the channel condition.
2) Unbalanced MD operation
The characteristics of each path in a packet network are different and time- varying,
therefore the available bandwidth in each path may differ. This results in the requirement
of unbalanced MD operation [26], where the bit rate of each description is adapted based
on the available bandwidth along its path. We will discuss this issue in the next section.
3.3 Distributed Multimedia Delivery Mobile Network
3.3.1 Concept of Content Delivery Network
Content Delivery Network (CDN) [43] was developed to overcome performance
problems, such as network congestion and server overload in the star-type network
topology, which arise when many users access popular content. CDN that is distributed
via a WAN generally consists of the origin server containing the content and a set of edge
servers. Each edge server is located closer to users and stores a subset of the content or
caches popular content. CDN provides a number of advantages.
1) It enhances server scalability and helps prevent server overload and network
congestion.
Conventionally, content is delivered by the central content server to the entire
network. The central server become a bottleneck of the networks and is lack of scalability
under the limitation of bandwidth, storage and computational complexity in the content
server. In the case of media multicast or broadcast, too much feedback creates a problem,
called feedback implosion.
The numerical investigations in [12] indicated that, for typical scenarios, the revenue
rate increases logarithmically with the cache space and linearly with the link bandwidth
connecting the cache space to the central server. Thus, it is beneficial to establish the
edge servers for caching before increasing the link bandwidth of the central server. With
38
the concept of distributed storage, CDN helps to prevent server overload and network
congestion, since the replicated content can be delivered to users from edge servers.
2) It reduces the content delivery time for services, the probability of packet loss,
and the total network resource occupation
As content is delivered from the closest edge server and not from the origin server,
the content is sent over a shorter network path, thus reducing the content delivery time,
the probability of packet loss, and the total network resource occupation.
3.3.2 Proposed Mobile Network Model
While CDN was originally intended for static web content, it can be extended to our
approach for the delivery of streaming media as well. With the comprehensive
consideration and integration of concepts of CDN and SMDC, a distributed multimedia
delivery mobile network is proposed for the introduction of media streaming services in
GSM/GPRS/UMTS. The proposed network model in the scenario of intra-RAN handoff
and inter-RAN handoff are illustrated in Figure 3-3 and Figure 3-4 respectively, where
DiffServ is employed to provision UEP for scalable video coding. The media streaming
services are pushed to the edge of CN so that the content delivery time for services is
significantly reduced by the media delivery network.
In each media delivery network, a set of complementary distributed Media
Description Servers (MDSs) interact and collaborate with each SGSN or MSC for media
delivery to mobile stations in the radio access network. Each MDS should keep one
complementary description of media streams that were originally downloaded from the
service provider during the streaming service publication.
In a real world of networks, such as GSM, the high- level streaming service control
functions (e.g., service registration, publication and discovery, service subscription and
binding, subscriber authentication and service allowance verification) can be peeled off
from the MSC and be implemented in VLRs. It is cost-effective because the MSC
becomes more efficient, does not waste cycles in processing new services, and simplifies
new service development.

39

Figure 3-3 Proposed network model of intra-RAN handoff (Data plane)
The proposed distributed multimedia delivery mobile network helps to deal with the
following problems:
1) Network congestion and server overload problems in the star-type network
topology.
2) The media streaming handoff problems in the mobile networks.
The streaming media is delivered from the closest edge server and not from the
origin server, the streaming media is sent over a shorter network path, thus reducing the
media service delivery time (end-to-end delay), the probability of packet loss, and the
total network resource occupation.
3) The high requirement of storage, reliability and load balancing among the
distributed media edge servers and thus high cost of network components in the
conventional CDN.
40


Figure 3-4 Proposed network model of inter-RAN handoff (Data plane)
One difference between CDN and D-MDMN which should be paid attention to is
that load balancing is even more important for media edge servers than for web servers,
since the resource commitment is typically larger and lasts longer.
Figure 3-5 shows a comparison between the single description (SD) and multiple
description (MD) approach.
During the construction phase of conventional CDN where the SD approach is used,
a redirection server (RS) should choose a set of media edge servers (ESs) that can achieve
the best end-user performance for each edge router (ER). The ER acts as an entry of a
wired access network. In Figure 3-5, ES 1 and ES 2 are supposed to be the best set for the
client.
Usually, each ER has one RS to monitor and balance the traffic load among the set
of ESs. Each edge server should keep one copy (SD) of media streams. Suppose ES 1 is
overloaded during a media delivery, then RS should direct ES 2 which is underloaded to
41
continue the delivery of the rest part of that media stream. This incurs the server-side
handoff problem [11] due to load balancing.
Similarly, during the phase of D-MDMN construction, each SGSN or MSC should
choose a set of MDS that can get the best end-user performance through the simulation of
system configuration. Each MDS should keep one description of media streams. In
Figure 3-5, MDS 1 and MDS 2 are supposed to be the best set for the client (SGSN or
MSC) and keep D1 and D2 of media streams, respectively.


Figure 3-5 Single Description versus Multiple Description
With the employment of the MD approach, however, there is no or extremely less
requirement of load balancing, especially when the proposed SMDC is used to combat
the problem of unbalanced MD operation. MDS 1 and MDS 2 provide the delivery of
different description of a media stream simultaneously. If MDS 1 is overloaded during
media delivery, it can discard the appropriate enhancement layers to lower its load. If
MDS 1 remains overloaded even after layer dropping, D2 still can be presented to the
client at a tolerable quality.
MD approach also lowers the reliability requirement of MDS due to its loss
tolerance capability. Suppose the reliability of a single description is 90%, then that of
two descriptions will be 1-(1-90%)(1-90%) = 99%. Also, each SMDC description is
almost only 56~60 % of the amount of a SD copy [47], which decreases the storage
requirement of MDS. In addition, SMDC has a very strong capability of error recovery
during media transmission.
42
A summary of comparison between the SD and MD approaches is given in Table 3-2.
Table 3-2 Single Description versus Multiple Description
Single Description Multiple Description
Storage 100 % 56~60 %
Reliability 90 % 99 %
Error recovery Weak Strong
Load balancing Required Not required

3.3.3 MPEG-4 Video Streaming over IP DiffServ
UMTS QoS Classes and IP DiffServ Classes
The UMTS specifications define in [61] four QoS classes: conversational,
streaming, interactive, and background. The main distinguishing factor among these
classes is delay sensitivity. The conversational class is the most sensitive, while
background is the least sensitive. Conversational and streaming classes are intended for
real-time traffic. They both preserve time relation (variation) between information
elements of the stream, but conversational has stricter delay requirements. Example
applications are IP telephony for the former and streaming video for the latter. For the
interactive and background classes, transfer delay is not the major factor. Instead, they
both preserve the payload content. The interactive class follows a request-response
pattern and defines three priorities to differentiate bearer qualities, while it does not
provide explicit quality guarantees. The main characteristic of the background class is
that the destination does not expect the data within a certain time. Example applications
are FTP or Web traffic for interactive and download of emails for background.
On the other hand, the IETF has also defined DiffServ mechanisms in RFC 2475
[45] for IP based networks aiming at QoS provisioning by means of Class of Service
(CoS) approaches, which is well suitable for unequal error protection of video layered
coding.
43
DiffServ architecture defines a simple forwarding mechanism, i.e., per-hop behavior
(PHB) [39], at interior network nodes while pushes most of the complexity to network
boundaries. Differentiated services are realized in RFC 3290 [54] by the use of particular
packet classification and traffic conditioning mechanisms at boundaries coupled with the
concatenation of per-hop behaviors along the transit path of the traffic.
The traffic conditioner is decoupled from the network interior and consists of
marker, meter, shaper and policer (i.e., dropper). Marking is performed at the source host
or the first- hop router administrative domain by means of mapping the DiffServ
codepoint (DSCP) contained in the IP packet header to a PHB. A replacement header
field, called the DS field, is defined in RFC [44], which is intended to supersede the
existing definitions of the IPv4 TOS octet and the IPv6 Traffic Class octet. Six bits of the
DS field are used as a DSCP to select the PHB that a packet experiences at each node.
In the packet forwarding path, per-hop behaviors are defined to permit a reasonably
granular means of allocating buffer and bandwidth resources at each node among
competing traffic streams. PHBs are expected to be implemented by employing queue
management and scheduling on a network node's output interface queue. In DiffServ,
three main PHB have been defined:
l Expedited Forwarding
Expedited Forwarding (EF) [41] provides a low delay, a low loss and an assured
bandwidth similarly to CBR in ATM.
l Assured Forwarding
Assured Forwarding (AF) PHB group [39], similarly to nrt-VBR/ABR/GFR in
ATM, is a means for a provider DS domain (i.e., Media Delivery DiffServ Network in
Figure 3-4) to offer different levels of forwarding assurances for IP packets received from
a customer DS domain (e.g., a RAN A or B in Figure 3-4). N independent AF classes are
defined, where each AF class in each DS node is allocated a certain amount of
forwarding resources (buffer space and bandwidth). IP packets that wish to use the
services provided by the AF PHB group are assigned by the video provider in the Media
Delivery DiffServ Network into one or more of these AF classes according to the services
that the customer has subscribed to. All packets belonging to an AF class are admitted
into one AF queue to avoid out of order delivery. Within each AF class, IP packets are
44
marked (again by the video provider DS domain) with one of M different levels of drop
precedence. In case of congestion, the drop precedence of a packet determines the relative
importance of the packet within the AF class. A congested DS node tries to protect
packets with a lower drop precedence value from being lost by preferably discarding
packets with a higher drop precedence value.
An IP packet that belongs to an AF class i and has drop precedence j is marked with
the AF codepoint AFij, where 1 i N and 1 j M. Currently, four classes (N = 4)
with three levels of drop precedence in each class (M = 3) are defined in RFC 2597 for
general use. More AF classes or levels of drop precedence may be defined for local use.
The queue management and scheduling mechanisms of AF PHBs are illustrated in
Figure 3-6. The drop preferences within each class should be considered in the potential
approaches of queue management, such as WRED. The drop precedence queue
management can be implemented, for example, by using a leaky bucket traffic policer
with one token rate and two bucket size, which can be decided according to the service
level agreement (SLA).


Figure 3-6 Queue management and scheduling mechanisms of AF PHBs
l Best Effort
Best Effort (BE) provides no QoS guarantee.
It is challenging to provide QoS attribute translation and mapping between the IP
world and the UMTS world and to implement the IP differentiated services for the traffic
encapsulated and isolated by tunneling in UMTS. This issue will be addressed in Section
4.2.1 and Section 4.2.2.
45
IP DiffServ and GTP/FP Tunnels in UMTS
The protocol stack in the IP transport mode for non-video services in UMTS is
shown in Figure 3-7 (a) [47] [56] [62] [67] [69], while that for video services in the
proposed D-MDMN is shown in Figure 3-7 (b), where the Uu, Iub, Iu, Gn and Gi
interfaces are defined in 3GPP TS 25.401 V5.5.0 [74].


Figure 3-7 Protocol stack in UMTS and the Proposed D-MDMN (Data plane)
For non- video services, The uplink user- level packets are segmented by the user
equipment (UE) into Radio Network Layer (RNL) frames, called transport blocks. These
are carried over the Radio Frequency Layer (RFL), using W-CDMA access and
modulation techniques, to the Node Bs within reach of the mobile. Each Node B
46
encapsulates a set of transport blocks into a single frame of the RNL Frame relay
transport Protocol (FP) and forwards the frame to its RNC over the Iub interface.
The FP frames can be exchanged between the drifting and serving RNCs over the
Iur interface. The serving RNC of the host is responsible for frame selection among the
multiple received copies of the same transport block, processing the other sublayers of
the RNL, and finally reassembling the user- level packet.
To deploy the IP transport mode on the Iub interface, the FP frames are encapsulated
into IP packets. The destination addresses of these packets refer to the network components
(i.e., RNC or Node B) and not the users IP address. In fact, the hosts IP address is never
used for forwarding purposes in the UTRAN, the decisions being made on the basis of
RNL specific protocols. The Mobile Wireless Internet Forum has specified further details
[50] concerning the implementation of IP in the UTRAN in the transport mode.
In order to communicate with the data network, the mobile host needs to register
with the CN by performing a GPRS attach operation. This results in the creation of two
GPRS Tunneling Protocol (GTP) sessions, specific to that host: between the RNC and the
SGSN on the Iu interface, and between the SGSN and the GGSN on the Gn interface.
The user- level multi-protocol packets are allowed to be encapsulated into GTP frames
and be forwarded between the RNC and the GGSN. The GTP protocol is implemented
only by SGSNs and GGSNs. No other systems need to be aware of the GTP's presence.
Mobile hosts are connected to an SGSN without being aware of GTP.
Upon the GPRS attachment, a mapping is created at the RNC between the host
identity and the GTP session between the RNC and the SGSN. In addition, a record is
created at the GGSN, which contains the mapping between the hosts IP address and the
GTP session with the corresponding SGSN.
To deploy the IP transport mode on the Iu and Gn interfaces, the GTP frames are
encapsulated into IP packets. The destination addresses of these packets refer to the
network components (i.e., RNC, SGSN, or GGSN) and not the hosts IP address.
Forwarding decisions are based on the GTP mapping tables in those nodes.
For video services, the only UMTS network component needed to be enhanced is
the SGSN where a media module should be added to interconnect with IP-based media
delivery network in the IP native mode. The reason not adopting the IP transport mode is
47
based on the overhead of GTP, UDP and another IP layer which has to be inserted
between L2 and the original IP layer in all MDSs. The IP transport mode is kept,
however, for non- video services. The deployment of the IP native mode between SGSNs
media module and MDS leverages the evolution of UMTS towards its final all IP-based
phase. The protocol stack of UDP/RTP will be discussed in details in Section 3.4.
As discussed above, the video IP packets generated from the video provider are
supposed to pass two different tunnels in UMTS. One is GTP tunnel existing between
GGSN and RNC in the CN. The other one is FP tunnel started from RNC and terminated
at Node B in the RAN. As the video IP packets pass through the tunnel, there are
additional headers (i.e., GTP or FP tunnel header) inserted between the two IP headers.
The inner IP header is that of the original traffic with differentiated services; an outer IP
header is attached and detached at tunnel endpoints without differentiated processing. In
general, intermediate network nodes between tunnel endpoints operate solely on the outer
IP header, and hence DiffServ-capable intermediate nodes access and modify only the
DSCP field in the outer IP header. Thus, it is a challenging design issue [40] to
implement DiffServ in the mobile network for the traffic which is encapsulated by the
tunnels. This issue will be addressed in Section 4.2.3.
3.4 Protocol Stack of End Systems
One main design goal of the network-aware end systems is to extend the
applications of real time streaming protocols in the Internet to wireless networks. The
protocol stack of streaming server and client is shown in Figure 3-8.


Figure 3-8 Protocol stack of network-aware end systems
48
There are two communication planes required by remote media access: a Data Plane
for media transport, and a Control Plane for media session control. The MPEG-4
specification adopts out-of-band signaling, so that the Data and Control Planes can use
different transport protocols, i.e., User Datagram Protocol (UDP) and TCP respectively in
our approach. The process of transporting of MPEG-4 content over the Internet or
wireless networks can therefore be split into two parts, the MPEG-4 Media Transport and
MPEG-4 Media Control.
3.4.1 MPEG-4 Media Transport
The transport protocol family for media streaming includes UDP, TCP, Real Time
Protocol (RTP), and Real Time Control Protocol (RTCP) protocols. UDP and TCP
provide basic transport functions while RTP and RTCP run on top of UDP/TCP.
UDP and TCP
UDP/TCP protocols support such functions as multiplexing, error control,
congestion control, or flow control. Since TCP retransmission introduces delays that are
not acceptable for streaming applications with stringent delay requirements, UDP is
typically employed as the transport protocol for video streams, while TCP is for
streaming control. In addition, since UDP does not guarantee packet delivery, the receiver
needs to rely on the upper layer (i.e., RTP) to detect packet loss.
RTP and RTCP
The Real Time Protocol [15] is an Internet standard protocol designed to provide
end-to-end transport functions for supporting real-time applications. The Real Time
Control Protocol [15] is a companion protocol with RTP and is designed to provide QoS
feedback to the participants of an RTP session. In other words, RTP is a data transfer
protocol while RTCP is a control protocol.
RTP does not guarantee QoS or reliable delivery, but rather, provides the following
functions in support of media streaming: time-stamping, sequence numbering, payload
type identification, and source identification.
RTCP is the control protocol designed to work in conjunction with RTP. In an RTP
session, participants periodically send RTCP packets to convey feedback on quality of
49
data delivery and information of membership. Basically, RTCP provides the following
services:
1) QoS feedback: This is the primary function of RTCP. RTCP provides feedback to
an application regarding the quality of data distribution. The feedback is in the form of
sender reports (sent by the source) and receiver reports (sent by the receiver). The reports
can contain information on the quality of reception such as:
a) fraction of the lost RTP packets, since the last report;
b) cumulative number of lost packets, since the beginning of reception;
c) packet interarrival jitter;
d) delay since receiving the last senders report. The control information is useful to
the senders, the receivers, and third-party monitors.
Based on the feedback, the sender can adjust its transmission rate; the receivers can
determine whether congestion is local, regional, or global; and network managers can
evaluate the network performance for multicast distribution.
2) Participant identification
3) Control packets scaling: To scale the RTCP control packet transmission with the
number of participants, a control mechanism is designed as follows. The control
mechanism keeps the total control packets to 5% of the total session bandwidth. Among
the control packets, 25% are allocated to the sender reports and 75% to the receiver
reports. To prevent control packet starvation, at least one control packet is sent within 5 s
at the sender or receiver.
4) Inter-media synchronization
5) Minimal session control information
3.4.2 MPEG-4 Media Control
In the media control part, specific session control protocol should be used to define
the messages and procedures to control the delivery of the multimedia data during an
established session. The Real Time Streaming Protocol (RTSP) [14] is such a session
control protocol which has been recommended by 3GPP for packet-switched streaming
50
service (PSS) [9] and International Telecommunication Union (ITU) H.323 for
multimedia teleconferencing services [43].
One of the main functions of RTSP is to support video-cassette-recorder- like control
operations such as stop, pause/resume, fast forward, and fast backward. In addition,
RTSP also provides means for choosing delivery channels (e.g., UDP, multicast UDP, or
TCP), and delivery mechanisms based upon RTP. RTSP works for multicast as well as
unicast. Another main function of RTSP is to establish and control streams of continuous
audio and video media between the media servers and the clients.
However, RTSP is specially designed for the Internet. For the wireless applications,
it should reply on others mobility mechanisms, such as the GTP-based link- level
mobility mechanism in GPRS and UMTS or SIP-based application level mobility
mechanism. In our approach, an enhanced GTP-based handoff procedure is proposed for
handling user mobility in media streaming.
In this chapter, the concepts of layered coding and multiple description coding are
introduced in order to solve the bandwidth fluctuation, packet loss and heterogeneity
problem in the wireless environment. The system model is proposed as a distributed
multimedia delivery mobile network, followed by the discussion of the video streaming
over IP DiffServ. The protocol stacks of the proposed D-MDMN and the network-aware
end system are presented. In the next chapter, we propose solutions for video mobility
under this system model.
51
4 Proposed Solutions for Video Mobility
The problem of handoff in a wireless network is well-know; however, it is largely
unexplored in the applications of streaming media.
For the purpose of solving the handoff problems in media streaming as discussed in
Chapter 2, a combination of scalable multiple description coding with distributed video
storage in the DiffServ mobile network to support streaming video handoff is proposed. It
leverages the distributed multimedia delivery mobile network to provide path diversity to
combat outage due to handoff. If a feedback channel is available, receiver reports from both
the base station and mobile host will be employed to split the wired and wireless domain,
such that the wireless channel condition (e.g., packet loss) can be known by the sender.
In this chapter, the join design of layered coding and multiple description coding and
the proposed scalable multiple description coding will be presented. And then, the
MPEG-4 video streaming issues over the IP DiffServ mobile network and the proposed
handoff procedures will be discussed.
4.1 Joint Design of MDC and LC
There are almost no referenced works on the joint design of layered coding and
multiple description coding to complement their drawbacks. In our approach, a Scalable
Multiple Description Coding framework is proposed to leverage the distributed
multimedia delivery mobile network to provide path diversity to combat outage due to
handoff. The coded video stream consists of MDC components and LC components. In
the proposed multimedia delivery mobile network, MDC components enhance the
robustness to losses and bit errors of LC components through path diversity and error
recovery. MDC components also reduce the storage, reliability and load balancing
requirement among distributed media edge servers. At the same time, LC components not
only deal with the unbalanced MD operation at the server end, but also combat the
bandwidth frustrations of the time-varying wireless channels.
52
4.1.1 Architecture of Proposed SMDC Framework
The architecture of the proposed SMDC framework is depicted in Figure 4-1. It is an
object-based coding which jointly employs the PFGS and Multiple State Recovery (MSR)
on the condition that it is compatible with MPEG-4.


Figure 4-1 Proposed SMDC architecture
Similar to the human visual system mechanism, the smallest entity in the SMDC is
each object in a picture with its associated shape, texture in the interior of the shape, and
motion. The original video input to the encoder is segmented into a set of individual
video objects (VOs). Each VO then is separately compressed through shape encoding and
PFGS texture encoding, such that the shape and texture information can be split into four
different VOPs. For support of two descriptions, the encoder should store the last two
previously coded frames (instead of just the last one) and choose which previously coded
frame to be used as the reference for the cur rent prediction. After multiplexing, the four
different VOPs converge into one video stream with four different layers. This video
53
stream is further partitioned into two subsequences of frames: odd numbered video
frames (Description 1) and even numbered video frames (Description 2).
The different descriptions should be transmitted over different channels undergoing
independent error effects to minimize the chance that both video streams are corrupted at
the same time. As a matter of fact, the video stream can be partitioned into N
complementary frame subsequences if there are N different channels between the encoder
and the decoder. However, it also adds complexity of MD assembling at the decoder. It is
an open issue to determine how many descriptions should be used for encoding. For
presentation simplification, two descriptions and two corresponding channels are chosen
in the following discussion.
At the decoder, the processing procedures reverse in accordance. Similarly, the
decoder should alternate which previous erroneously decoded frame it uses as the
reference for the next prediction. Both MPEG-4 version 2 (NEWPRED mode) [8] [29]
[30] and H.263+ (RPS mode) [31] [21] support switching prediction among reference
frames. If both descriptions are received erroneously after parallel to serial converter
(MD assembler) and demultiplexing, then the shape and texture information of VOs are
restored from shape and PFGS texture decoder for final composition into reconstructed
video. If there is an error in a stream, the error propagation will happen in that stream due
to motion compensation and differential encoding.
The SMDC framework can employ any shape coding [13] [42], e.g., binary shape
coding or grayscale coding [63]. The texture coding techniques are still DCT-based
coding for arbitrary shaped objects. The concept of object based representation makes it
possible to exploit the content redundancy in addition to the data redundancy and
improve the coding efficiency for the very low bit-rate transmission.
For an illustration of the capability of error recovery, an MD approach is compared
with a conventional single description approach in Figure 4-2 [47]. For simplicity, B-
frames are not illustrated in the figure. There is an error when decoding P- frame 4 in SD
or P-frame 5 correspondingly in MD which is forward predicted by P-frame 3. In the SD
approach, P frame 4 is lost and the decoder has to freeze P-frame 3 (or perform other
error concealment) until I- frame 10. In the MD approach, however, P- frame 5 may be
recovered or concealed by using information from its previous P- frame 4 and future P-
54
frame 6 which are correctly decoded in the other description. Errors on both descriptions
in decoding P- frames 8 and 11 are recovered or concealed in the same manner as long as
both descriptions are not simultaneously lost.

Figure 4-2 Single Description versus Multiple Description
Note that, the MD approach should add extra bits for MD property; and the SD
approach should also add extra bits (e.g., I-frame 10) for intra coding or FEC. It has been
shown that the MD approach requires an increase of 12~20 % transmission bit rate as
compared with the SD approach but will result in much stronger capability of error
recovery than its counterpart [47].
4.1.2 Scalability Structure of SMDC Framework
The proposed SMDC scalability structure (shown in Figure 4-3) is as follows:
1) The Shape Base Layer that consists of shape information of VOs in the intra-
coded plane (I-VOP) or shape and motion information of VOs in the predictively
coded plane (P-VOP);
2) The Texture Base Layer that consists of basic texture information of VOs
contoured by the Shape Base Layer;
55
3) The Texture PFGS Layer that consists of texture information of SNR scalable
enhancement for the Texture Base Layer;
4) The Texture PFGST Layer that consists of motion-compensated residual frames
(i.e., motion vectors and bitplane-DCT residual signals) predicted from the
Texture Base Layer for temporal scalable enhancement. In comparison, the
motion-compensated PFGST frames in SMDC take the place of B- frames in the
multilayer FGS-temporal scalability structure presented in [6].


Figure 4-3 Proposed SMDC scalability structure
For illustration sake, shown in Figure 4-4, suppose the first three layers are
implemented and the Texture PFGST Layer is left as an option. Thus, playing only one
description with only the Shape Base Layer gets a black and white (or grayscale) video at
the half frame rate, shown in Figure 4-4 (a). Playing only one description with the Shape
Base Layer and the Texture Base Layer gets a color video in a basic quality at the half
frame rate, shown in Figure 4-4 (b). Playing only one description with all three layers
yields a color video in a better quality at the half frame rate, shown in Figure 4-4 (c). In
the same way, if both two descriptions with all the three layers can be decoded correctly,
it yields a color video in the best quality at the full frame rate, shown in Figure 4-4 (d).
However, the layering in SMDC is more flexible than that of illustration. The
Texture PFGS Layer needs not be discarded as a whole. The enhancement bit stream can
56
be truncated anywhere to achieve the target bit-rate. This benefit of achieving continuous
rate control comes from the bitplane coding in the PFGS encoder [17] for the
enhancement stream.



(a) One description with only the Shape Base Layer



(b) One description with the Shape Base Layer and the Texture Base Layer



(c) One description with all three layers


(d) Both two descriptions with all the three layers
Figure 4-4 Video illustration of SMDC layered structure
57
4.1.3 Advantages of Proposed LC Component
1. Wired Domain
1) Unbalanced MD Operation
As mentioned in Section 3.2, the requirement of unbalanced MD coding [26] is well-
known but largely unexplored in MD video coding. The characteristics of each path in a
packet network are different and time-varying, therefore the available bandwidth in each
path may differ. This results in the requirement of unbalanced MD operation, where the
bit rate of each description is adapted based on the available bandwidth along its path.
The proposed SMDC description is scalable along its path without any close- loop
feedback delay, which is well suitable for unbalanced MD operation.
The proposed SMDC approach is naturally balanced in both streams (i.e.,
descriptions) assuming that the even and odd frames have equal complexity. To achieve
unbalanced operation one can adapt the quantization, spatial resolution or frame rate.
However it is important to preserve approximately equal quality in each stream to prevent
an observer from perceiving a quality variation (flicker) at half the original frame rate
(particularly important for the case of no losses).
Rate control via coarser quantization may be used for small rate changes (e.g., 10%
rate reduction at a cost of 0.5 dB [26]); however, it may not be appropriate for large rate
changes. The potential flicker also suggests that changes in spatial resolution (i.e., spatial
subsampling) may be inappropriate.
Adapting the frame rate (i.e., temporal subsampling) [26] is a simple and effective
mechanism for reducing the required bit rate while preserving the quality per frame and
largely preserving the error recovery capability, as illustrated in Figure 4-5. However, if
the frame rate of one stream is decreased too much, the quality variation of that stream
can not be approximately preserved. Also, the unbalanced MD operation will fail if the
bit rate ratio of these two streams is larger than 2:1, which is illustrated in Figure 4-5.
Suppose that the bit rate of the upper stream is bigger than that of the lower stream in
Figure 4-5, where P
x
denotes the P-frame X. The balanced MD operation is shown in Figure
4-5 (a), where the damaged P-frame 5 can be recovered or concealed from P-frames 4 and 6,
and P-frame 11 is recovered or concealed from P-frames 10 and 12. In Figure 4-5 (b), the
frame rate of the lower stream has to be decreased by 50% for a bit rate ratio of 2:1. That is,
58
P-frames 4 and 8 have to be discarded. The damaged P-frame 5 can be recovered but only
from P-frame 6, and P-frame 11 can be recovered but only from P-frame 10. It is
straightforward that the error recovery capability of 2:1 unbalanced MD operation, illustrated
in Figure 4-5 (b), is lower that that of the balanced MD operation illustrated in Figure 4-5 (a).
In Figure 4-5 (c), the frame rate of the lower stream has to be decreased by 75% for a bit rate
ratio of 3:1. However, the damaged P-frames 5 and 11 in the high bit rate stream can not be
recovered from the low bit rate stream because their adjacent previous P-frames 4 and 10,
and their adjacent future P-frames 6 and 12 have to be discarded.


Figure 4-5 Balanced and Unbalanced MD Operation
59
In order to address the unbalanced problem, the concept of layered coding is
proposed for the MD video coding. The capability of error recovery of MDC and SMDC
are compared under the bit rate ratio of 3:1 in Figure 4-6.


Figure 4-6 Comparison of Unbalanced MDC with Unbalanced SMDC
As discussed above, the errors occurred in the high bit rate stream can not be
recovered or concealed from the low bit rate stream in the case that the bit rate ratio is
larger than 2:1. Suppose that the bit rate of the upper stream in Figure 4-6 is three units
and that of the lower stream is only one unit. Instead of temporal subsampling illustrated
in Figure 4-6 (a), the layered coding is introduced in Figure 4-6 (b) for the unbalanced
MD operation. As to the path of low bandwidth, part of the enhancement layers can be
dropped so that the original frame rate can be preserved. In Figure 4-6, P
x
denotes the
rest past of P-frame X after layer-dropping in order to adapt to the bandwidth limitation.
Thus, the damaged P-frames 5 and 11 in the upper stream still can be recovered from the
frame of the lower stream, shown in Figure 4-6 (b), in the same manner as the balanced
MD operation shown in Figure 4-5 (a). In other words, the unbalanced MD operation
using LC does not affect the error recovery capability of SMDC.
60
Table 4-1 summaries the methods of adapting bit rate of each description. LC
technique is suitable for the unbalanced MD operation, especially when the bit rate ratio
of both descriptions is larger than 2:1.
Table 4-1 Methods of adapting bit rate of each description
Performance without loss Performance with loss
Quantization
Good for small changes (0-
10 %), above which
possible flickers exist
Good for small changes (0-
10%), above which error
recovery capability reduces
Spatial Subsampling Potential flicker Potentially reduced
Temporal Subsampling
Good for middle range of
bit rate changes (bit rate
ratio less than 2:1)
Generally good recovery
(bit rate ratio less than 2:1)
Layered Coding in SMDC
Good for large range of bit
rate changes (bit rate ratio
larger than 2:1)
Generally good recovery
(bit rate ratio larger than
2:1)

2) Error Resilience Enhancement
To explore the error resilience and concealment tools in MPEG-4, there is a clear
advantage to distinguish not only different kinds of frames (referred to as Video Object
Plane or VOP in MPEG-4), but also different types of information, such as shape,
motion, and texture, within the same frame. Therefore, the shape, motion, and texture
information in the bit-stream of an object-based video is re-organized into different layers
in the proposed SMDC scheme to support the classification and priority assignment in the
DiffServ network, which is discussed in details in Section 3.3.3.
2. Wireless Domain
1) Open-loop Rate Control
Design issue 1: RTCP-based rate control is specifically designed for the Internet
A number of papers have considered how to control the transmission rate of non-TCP
flows. TCP-friendly model-based [66] and probe-based [36] rate control mechanisms
calculate their maximum transmission rate using a TCP throughput formula [24] [22] or
mimicking TCP behavior. To determine the transmission rate, these mechanisms require
feedback from the receiver to obtain packet loss rate and round trip time (RTT)
61
information. Some rate control mechanisms [36] utilize the RTCP to obtain feedback
information from receivers . The receiver of an RTP media stream sends back RTCP
receiver reports, which include packet loss and jitter information, so that the sender can
identify network congestion condition and control its transmission rate accordingly. Most
of the rate control mechanisms mentioned above are designed specially for the Internet and
assumed that packet loss, delay, and jitter are caused by network congestion.
In mobile networks, however, packet loss and jitter may also be caused by radio link
errors. Since radio links have much higher bit error rates, packets are frequently discarded
due to the presence of bit errors. When conventional rate control mechanisms are applied
to mobile networks, a sender cannot identify the network congestion condition correctly,
and this leads to inappropriate rate control. A typical symptom is that a sender reduces its
transmission rate even if the network is not congested [75].
Design issue 2: The relatively long RTCP transmission interval.
To scale the RTCP control packet transmission with the number of participants, a
control mechanism is designed as follows in RFC 1889 [15]. The control mechanism
keeps the total control packets to 5% of the total session bandwidth. Among the control
packets, 25% are allocated to the sender reports and 75% to the receiver reports. To
prevent control packet starvation, at least one control packet is sent within 5 seconds at
the sender or receiver.
RTCP makes no provision for timely feedback that would allow a sender to repair the
media stream immediately: through retransmissions [65], reactive FEC, or media-specific
mechanisms such as reference picture selection for some video codecs. Typically, the
feedback interval is constrained on the order of tens to hundreds of milliseconds [48].
The QoS maintenance or guarantees to multimedia streams using RTCP-based reports
is still under investigation, especially in the wireless environment. As a result, instead of
close- loop rate control mechanism, the concept of layered coding is adopted in our
approach to implement open- loop adaptive rate control mechanism (e.g., DiffServ-based
rate filtering) in the BS and other wired intermediate nodes to combat traffic congestion.
2) Resource Reservation and Heterogeneity
The concept of layered coding provides a very flexible and efficient solution to the
problem of resource reservation and receiver heterogeneity in the wireless domain.
62
First, there is no need to reserve bandwidth for the complete stream since typically
only the base layer needs QoS guarantee. As a result, CAC can be based only on the
requirement of the base layer and resources are reserved only for the base layer. Second,
the enhancement layers of one connection can share the leftover bandwidth with the
enhancement layers of other connections. The CAC algorithms for wireless channels are
out of the scope of the thesis.
In addition, the technique of the layered coder is a better alternative of trancoder in
the base stations to combat the network heterogeneity or receiver heterogeneity so that
there are no migration of transcoding state information required, which will be discussed
in Section 4.3.3.
4.2 MPEG-4 Video Streaming over IP DiffServ
4.2.1 QoS Mapping between UMTS and IP DiffServ
Table 4-2 Mapping of UMTS QoS classes to DiffServ PHB classes
UMTS
QoS classes
Conversational
Class
Streaming
Class
Interactive
Class
Background
Class
Application
VoIP/video
conferencing
Streaming
audio/video
Web browsing
Background
download
/E- mails
Delay jitter
Stringent
and low
Bounded Tolerable Unbounded
Features
BER Tolerable Tolerable Low Low
Maximum bit
rate
2Mbps 2Mbps 2Mbps 2Mbps
Guaranteed
bit rate
2Mbps 2Mbps - -
Defined
attributes
Transfer
delay
100ms 280ms - -
Mapping
DiffServ
PHB classes
DSCP = EF DSCP = AF DSCP = BE DSCP = BE

63
To provide an end-to-end QoS for IP based traffic over a UMTS network, one of the
most difficult issues [72] [73] is to provide QoS attribute translation and mapping
between the IP world and the UMTS world. The features and the attributes of UMTS
QoS classes defined in [61] are summarized in Table 4-2. Our mapping scheme between
the DiffServ and UMTS QoS classes is also presented in Table 4-2.
4.2.2 IP DiffServ MPEG-4 Video Marking Algorithm
Existing Internet video dissemination schemes usually do not support classification
of varying video information beyond the simple distinguishment of different types of
frames (I, P and B frames) [16]. Encoded video data are placed in the bit-stream
according to the temporal and spatial positions, i.e., block by block, macroblock by
macroblock, and frame by frame. Different types of information, such as shape, motion,
and texture, are interleaved together, although they have different levels of importance
during decoding. For example, the shape and motion information is more important than
the texture for a P frame in MPEG-4 [30]. If the shape and motion information is lost
during transmission, it is hard for the decoder to reconstruct the P frame successfully.
However, if partial texture information is lost, it is still possible to reconstruct the P frame
with somewhat acceptable quality using error concealment algorithms.
Aiming at robust transmission, MPEG-4 supports a set of error resilience and
concealment tools, such as video packet based resynchronization approach which
provides a flexible self-contained decoding unit, and data partitioning mode which
separates the shape, motion and texture data in VPs using DC Markers or Motion
Markers. Such tools are quite suitable for wireless transmissions where most errors are at
the bit- level, but they are not sufficient for Internet transmission where most errors are
caused by packet loss. This is because a video packet (or several video packets) is
usually encapsulated as packet payload directly, and consequently the information is still
interleaved together within an IP packet.
To explore the error resilience [28] [29] [30] and concealment [34] tools in MPEG-
4, there is a clear advantage to distinguish not only different kinds of frames (i.e., VOP in
MPEG-4), but also different types of information within the same frame. Usually, IP
64
video traffic is classified as one AF class with three different levels of drop precedence.
For example, an IP DiffServ video marking algorithm (DVMA) is proposed in [71] [72]
as follows:
If stream is video stream then
If base layer video stream then
Level 1 = minimum QoS
DSCP= AF Low Drop Precedence (e.g., AF21)
If enhanced layer video stream 1 then
Level 2 = medium QoS
DSCP= AF Medium Drop Precedence (e.g., AF22)
If enhanced layer video stream 2 then
Level 3 = maximum QoS
DSCP= AF High Drop Precedence (e.g., AF23)
Typically, (W)RED [4] or similar active queue management approach has to be
adopted to combine stochastic dropping of packets with IP Precedence. The WRED
gateway calculates the average queue size AvQ, using a low-pass filter with an
exponential weighted moving average. Given the minimum threshold TH
min
, a maximum
threshold TH
max
, the exponential weight factor ef and the mark probability denominator p,
the WRED algorithm is described in the following:
For each packet arrival, calculate the average queue size

'

0 ) 1 (
0 ) 1 (
Q if AvQ W
Q if Q W AvQ W
AvQ
m
q
q q

If TH
min
AvQ < TH
max

Drop the arriving packet with probability P
Else if TH
max
AvQ
Drop the arriving packet
where

( )
ef
q
W 2 1
,

( ) ) ( 1
min max min
TH TH TH AvQ p P
,
m = queue_idle_time / transmission_time.
65
(W)RED takes advantage of the TCP retransmission mechanism. However, as
discussed in Section 3.4, UDP is more suitable for video streaming. All the random
dropping packets will be considered as packet loss and will not be retransmitted. Because
of error propagation of streaming video, the effect of packet loss gets worse. Since MPEG-
4 video is predictive inter- frame coded and layered coded, artifacts due to random packet
dropping can persist for many frames or layers. For example, consider a 30 frame/s MPEG
video sequence with one I frame every 15 frames. If an error occurs while transmitting the I
frame, the effect persists for 15 frames, or 500 ms, which is quite noticeable to a viewer.
Jill and Gaglianello analyzed and presented the results of the relationship between the
packet loss rate and the frame error rate [35], shown in Figure 4-7, from a study of
streaming MPEG compressed video over the public Internet, using the RTP and UDP
transport protocols. Similarly, if an error occurs while transmitting the base layer, its
enhancement layers have to be discarded. It means that stochastically isolated single packet
loss or bit error is converted to burst packet loss or bit errors. Therefore, early random
packet dropping before congestion is not suitable for video or audio streaming.


Figure 4-7 Packet loss effect on frame error rate
A novel marking algorithm for MPEG-4 video encoded by SMDC is proposed in
Table 4-3 to support the priority assignment after the classification given in Table 4-2. In
the proposed SMDC scheme, we re-organize different types of information, such as
shape, motion, and texture, in the bit-stream of an object-based video into four different
layers. Moreover, different layers of information are packetized into three different
66
classes of Application Level Packets (ALP) with various priorities, so that more
important compressed information can be put into higher priority packets and less
important information into lower priority ones. In comparison with the DVMA solution
where there is only one queue with three different levels of precedence for video stream,
each AF class in the proposed algorithm has one separated queue. This algorithm can be
implemented by class-based Weighted Fair Queuing (WFQ). Details of the WFQ
algorithm can be found in [76]. WRED should be disabled in each class.
Table 4-3 Proposed IP DiffServ MPEG-4 video marking algorithm

Control
Information
Shape Base
Layer
Texture Base
Layer
Texture PFGS
Layer
Texture
PFGST Layer
OD &
BIFS
DSCP = EF
or AF11
- - - -
I-VOP -
DSCP = AF11
(Class I
stream)
DSCP = AF11
(Class I
stream)
DSCP = AF21
(Class II
stream)
-
P-VOP -
DSCP= AF11
(Class I
stream)
DSCP= AF21
(Class II
stream)
DSCP = AF31
(Class III
stream)
-
PFGST
VOP
- - - -
DSCP = AF31
(Class III
stream)

In addition, MPEG-4 introduces extra data control stream, such as the object
descriptor (OD) and scene description (BIFS). These signalling streams are very loss-
and jitter-sensitive and need to be protected and marked as EF or AF11 if EF PHB is not
available.
4.2.3 Evolution of System Model for Native IP DiffServ
Evolution of Mobile Network Model
In order to support DiffServ in UMTS, we propose to copy the DSCP value in the
inner IP header to the outer IP header at encapsulation and copy the outer header's DSCP
67
value to the inner IP header at decapsulation. This mechanism allows GTP/FP tunnels to
be configured without regard to DiffServ domain boundaries.


Figure 4-8 UMTS network model in IP native mode with 3G AR
A more efficient alternative, however, is to keep the native DiffServ processing
procedure in the mobile network rather than the modified one as above (i.e., copying
DSCP value). This requirement leverages the evolution of UMTS towards its final all IP-
based phase, which is depicted in Figure 4-8. Note that the IP-based CN has enlarged to
the edge of UTRANs compared with the network model shown in Figure 2-2. The
functions of the GGSN, SGSN and RNC are further combined and implemented at one
node, Access Router (AR). The GGSN and SGSN functions within the 3G AR provide all
the UMTS-specific accounting and security features. The rest of the CN consists of
regular routers and switches that forward packets on the basis of the user-level IP
addresses. The Border Router (BR) denotes the functionality to avoid unwanted traffic
between GPRS CN and the Internet. One or more BRs are served as gateways to the
public Internet.
This network architecture provides a solution to implement the IP native mode
forwarding in a larger portion of the operators domain independent of any given access
technology, and hence can be used by the operator to support heterogeneous access
68
networks. As the coverage of the IP native mode increases, the wireless-specific
protocols are pushed farther toward the access segment. The operator may share the
domain with other access techniques by just using a specialized AR. For example, an
IEEE 802.11 AR may coexist with a 3G AR, using the same CN.


Figure 4-9 D-MDMN network model in IP native mode with 3G AR
Along with the enlargement of the IP-based CN, the introduction of D-MDMN in
the 3GPP network is depicted in Figure 4-9. The IP-based Media Delivery Network can
also be enlarged around ARs towards the whole IP-based CN, so that the media streaming
services can be pushed further to the edge of UTRANs.
Conceivably, the IP native mode coverage can be extended into the UTRAN by
implementing an AR with collocated Node B, RNC, SGSN and GGSN functions. Some
equipment vendors are adopting this approach by building what are known as intelligent
base stations with varying combined functionalities.
Evolution of Protocol Stack
In order to further analyse the implementation of DifferServ in Figure 4-9, the
protocol stack in the IP native mode for non-video services is shown in Figure 4-10 (a),
while that for video services is shown in Figure 4-10 (b).
69

Figure 4-10 Evolution of Protocol stack (Data plane)
As the coverage of the IP native mode increases, the stack becomes more efficient,
and the whole CN uses regular IP forwarding based on the end-users IP address instead
of the tunnel ID. FP frames are transported to and from the ARs over an IP network in the
transport mode. If the ARs are the next hop of Node Bs, the tunnels are short enough that
the ingresses (i.e., ARs) can execute DiffServ PHB based-on the inner IP header before
the it is encapsulated by the addition of the outer FP tunnel header, and the egresses can
also perform DiffServ PHB based-on the inner IP header after it is decapsulated by the
removal of the outer FP tunnel header. Thus, the native DiffServ processing procedure in
the mobile network is implemented rather than copying DSCP value from the inner IP
header to the outer IP header.
70
4.3 Proposed Handoff Procedures for Video Streaming
4.3.1 Proposed intra-RAN Handoff Procedure
Under the proposed intra-RAN network model of the distributed multimedia
delivery mobile network, shown in Figure 3-3, and the same assumptions given in
Section 2.3.2, the intra-RAN handoff procedure for media streaming are proposed as
follows, which also consists of three phases. The control plane of handoff procedure is
shown in Figure 4-11, while the data plane of these three phases is shown in Figure 3-3.
Only the differences are described in comparison with that in UMTS Release 4.


Figure 4-11 Propose intra-RAN handoff procedure (Control plane)
71
l Phase I: Preparation of RNS handoff and resource allocation
The control plane of the proposed handoff phase I is almost the same as that in
UMTS Release 4, shown in Figure 2-4, except that the current position (offset) of the
received media stream should go along with measurement report given by the MS. It is
the only state information required for session migration which is small enough to be
hidden inside the handoff signalling and be relayed to the SGSN.
On the data plane of handoff phase I, at the end of the preparation phase, the sRNS
stops transmitting downlink data to the MS but will not store all downlink data which
continue to arrive from the SGSN to the sRNC as no data forwarding is required.
l Phase II: Moving the Serving RNS role to target RNS
The most important difference between the proposed handoff phase II in the control
plane and that in UMTS Release 4 in Figure 2-4, is that the Stream Re-establishing takes
the place of the Media Stream Forwarding. There are no buffered data required to be
forwarded. As soon as the GTP tunnel is created between the tRNS and the SGSN, the
SGSN initiates the MD-request message (signal # 7) and the uplink flow is switched from
the old path to the new path. Upon receiving the MD-request, the set of MDSs
surrounding the SGSN starts the downlink media delivery from the offset point of the
same stream at the handoff decision according to subscriber-service bindings in VLR
(i.e., how many descriptions and layers the MS subscribes). In other words, the media
stream is re-established. The MD-request message contains the offset information at the
handoff decision point.
l Phase III: Releasing resource reservation in the old path
The most important difference between the proposed handoff phase III and that in
UMTS Release 4, shown in Figure 2-4, is that there is no buffer requirement in the BSs
for data forwarding and resequencing. Only a smaller buffer is needed in the BSs for
absorbing the delay jitter of a video stream and for re-ording due to changes in routing
paths. The functionality of multiple description assembly is implemented in the MSs.
4.3.2 Proposed inter-RAN Handoff Procedure
Under the proposed inter-RAN network model of the distributed multimedia
delivery mobile network, shown in Figure 3-4, and the same assumptions given in
72
Section 2.3.2, the proposed inter-RAN handoff procedure for media streaming also
consists of three phases. Here the control plane of handoff procedure is briefly presented
in Figure 4-12, while the data plane of these three phases is shown in Figure 3-4.
l Phase I: Preparation of RNS handoff and resource allocation
Note that there is no GTP tunnel required between SGSNs compared to UMTS
Release 4, since no data forwarding is required.
l Phase II: Moving the Serving RNS role to target RNS
l Phase III: Releasing resource reservation in the old path


Figure 4-12 Proposed inter-RAN handoff procedure (Control plane)
73
4.3.3 Handoff Enhancement for Streaming Services
The advantages of the proposed handoff approach for media streaming are
summarized as follows:
1. Due to the replacement of stream re-establishing with data forwarding, the
handoff latency can be reduced, which also reduce the buffer size in the BSs.
In the UMTS Rel 4 handoff procedures, for downlink media streams, there are two
possible situations when media stream gap or overlapping may happen:
(a) The media stream overlap/gap may be introduced when the tRNS takes the
Serving RNS role and starts to produce the downlink data from forwarded
GTP-PDUs. In this case the estimated gap/overlap for hard handoff is equal
to the delay of the GTP tunnel used for data forwarding. This first instance of
media stream overlap coincides with radio hard handover.
(b) The additional media stream gap may be introduced when the CN transport is
optimized. In this case the gap will exist only if the delay via the optimized
route is larger than the delay via the forwarding route.
In comparison, for downlink media streams during the proposed handoff procedures,
there are only one possible situation when media stream gap (but no overlapping) may
happen. That is, the media stream overlap/gap may be introduced when the tRNS takes
the serving RNS role and starts to request a new set of MDSs for the rest of video
delivery. In this case the estimated gap for hard handoff is equal to the delay of stream re-
establishment for media delivery. This coincides with radio hard handoff.
If the transport bearer delay difference is smaller than the air interface Transmission
Time Interval (TTI) (10, 20, 40 or 80 ms depending on the service), the amount of gap is
most likely not existent.
In addition, as discussed previously, there is no buffer requirement in the BSs for
data forwarding and resequencing in case of stream re-establishig. Only a smaller buffer
is needed in the BSs for absorbing the delay jitter of a video stream and for re-ording due
to changes in routing paths. The relatively small queue in the BSs also reduces the
handoff latency.

74
2. It has relatively low packet loss (or frame error), and end-to-end delay.
Since the media streaming services are pushed to the edge of core network and the
streaming media can be delivered over a shorter network path, the transfer delay and
delay jitter of media service delivery, the probability of packet loss, and the total network
resource occupation will be reduced.
Furthermore, with the employment of stream re-establishig, the relatively small
queue in the BSs reduces the end-to-end delay further.
3. There is no extra handoff latency introduced due to session migration.
Since the technique of layered coding, e.g., SMDC, is employed as an alternative of
trancoding to combat the network heterogeneity or receiver heterogeneity, there are no
migration of transcoding state information required. Also, the migration of session
description is not necessary because of the principle of stream re-establishing instead of
data forwarding. Thus, only the migration of session parameters at the handoff decision
point is required. The amount of state information is thereby small enough to be hidden
inside the handoff signalling, so that there are no extra handoff latency introduced due to
session migration.
4. It has relatively consistent QoS in all scenarios (Handoff scalability
enhancement).
In UMTS Release 4, the values of handoff latency (i.e., delay jitter) vary with the
lengths of data- forwarding path in different handoff scenarios. Also, the end-to-end delay
varies with different delivery paths and different locations of the media providers, which
is outside the CN and far from the mobile hosts.
However, due to the introduction of the distributed multimedia delivery mobile
network, the values of handoff latency and end-to-end delay in different handoff
scenarios depend mainly on the length of media delivery path from MDs to SGSN/MSC,
and then to MS. Usually, the SGSN/MSC and the MDs are neighbor nodes. It is
straightforward that the length of media delivery path from the MDs to the MS is
relatively consistent in different handoff scenarios.
5. The amount of signalling traffic is slightly reduced during inter-RAN handoff.
A brief summary of above comparison between our proposal and that of UMTS is
given in Table 4-4.
75
The proposed procedure in the scenario of inter-cell, intra-RNS handoff can be
found in Appendix. Note that, in the scenario of intra-cell handoff, it is not necessary to
re-establish media stream and the corresponding handoff procedures in UMTS R99 may
be extended to support media streaming services.
Table 4-4 Summary of handoff solution comparison
UMTS D-MDMN
Principle Data forwarding Stream re-establishing
Queue size in the BSs Large Small
Packet loss (Frame error) High Low
End-to-end delay High Low
Handoff latency
(gap or overlapping)
High
(For downlink, two
instances of stream gap/
overlapping may occur)
Low
(For uplink, only one
instance of stream gap
may occur)
Consistency of QoS in all
scenarios
(i.e., Handoff scalability issue)
Poor Good
intra-RAN handoff 12
13 Signalling
traffic
inter-RAN handoff 18
16

This chapter presents the details of the proposed solutions for video mobility under
the system model defined in Chapter 3. In Section 4.1, the Scalable Multiple Description
Coding framework is proposed to explore the joint design of layered coding and multiple
description coding. Section 4.2 describes a novel IP DiffServ video marking algorithm to
support the UEP of SMDC, which re-organizes the shape, motion, and texture
information of video stream into different layers in order to implement the DiffServ in
UMTS. Finally, the corresponding intra-RAN handoff and inter-RAN handoff procedures
in D-MDMN are studied in Section 4.3 with the employment of the principle of video
stream re-establishing for seamless handoff.
76
5 Simulations
5.1 Simulation Models
The simulation model of UMTS intra-RAN and inter-RAN handoff are shown in
Figure 5-1 and Figure 5-3, respectively. Correspondingly, the simulation model of D-
MDMN intra-RAN and inter-RAN handoff are proposed as illustrated in Figure 5-2 and
Figure 5-4, respectively. The parameters and configuration attributes of the simulation
model can be chosen for different simulation scenarios.
The difference between UMTS and MDMN model is that the central Video Provider
outside the CN in the UMTS is distributed and pushed to the edge of the RAN in the
MDMN. For the sake of simplicity, one practical topology of multimedia delivery
networks in the real world is that each distributed media server is simplified as a
multimedia database into each SGSN. IP DiffServ is implemented in each node within
both UMTS and MDMN.
Note that since the bandwidth fluctuations and limitation of the wireless channels
are what we are more concerned with, we set up the system such that no congestion
happens at wired nodes, except for the BSs.
A brief description of the network node and link models and their roles is presented
below.
Radio Access Network
RAN is modeled as RNS and Wireless AP in our simulation. RNS node model is
shown in Figure 5-5. It consists of RNS (data plane), RNS (control plane), AN router and
IP-based AN. The radio functionality of Base Station (Node B) is implemented in the
Wireless AP. The data functionalities of BS and RNC are implemented in the RNS (data
plane); the control functionalities of BS and RNC are in the RNS (control plane). The
protocol stack is illustrated in Figure 3-7. The scale of the radio access network is
dependent on the attributes (e.g., packet latency and packet loss ratio) of IP-based AN
77
cloud model. In our simulation, the packet latency of IP-based AN is configured as
exponential distributed with mean value of 15 ms; the packet loss ratio is zero.
The BS acts as the extension of mobile clients. It handles the difference between
wireless and wired networks. Each BS is generally responsible for wireless connection
setup, handoff support, and medium access control in its service area. For streaming
video applications, it also has the responsibility of QoS control, such as rate filtering,
scheduling and ARQ, for media streaming.
The Packet Analyzer is used for OPNET ACE Tools to capture packet traces, which
will not affect the simulation results.
SGSN/GGSN and Video Provider
SGSN/GGSN node model in UMTS is shown in Figure 5-6. It is composed of
SGSN/GGSN (data plane), SGSN/GGSN (control plane) and Access Router. Figure 5-7
shows the SGSN/GGSN node model in MDMN, where Video Providers are distributed as
multimedia databases. The protocol stacks of SGSN/GGSN and Video Provider are
illustrated in Figure 3-7.
IP backbone core network
The scale of the CN is dependent on the attributes (e.g., packet latency and packet
loss ratio) of IP backbone CN cloud model. The packet latency of IP backbone CN is
configured as exponential distribution with mean value of 20 ms; the packet loss ratio is
zero.
Mobile Station
MS node model is shown in Figure 5-8, which is implemented according to the
protocol stack illustrated in Figure 3-7. The RFL, RNL, ip, ip_encap, tcp, udp, rtp and
application layer processors are taken from OPNET Modelers library. Each MS supports
three types of services: video streaming service, voice service and web service. The voice
and web services are configured as background traffic which are established between IP
Phone User and Voice User, Web Server and Web User, respectively.
78

Figure 5-1 UMTS simulation model (Intra-RAN Handoff)

Figure 5-2 MDMN simulation model (Intra-RAN Handoff)
79

Figure 5-3 UMTS simulation model (Inter-RAN Handoff)

Figure 5-4 MDMN simulation model (Inter-RAN Handoff)
80

Figure 5-5 Node Model of RNS

Figure 5-6 Node Model of SGSN (or GGSN)

Figure 5-7 Node Model of SGSN in the proposed MDMN
81

Figure 5-8 Node Model of MS
Wired Links
The PPP point-to-point links at OC3 data rate (155Mbps) are used between nodes
with serial interfaces (e.g., routers with PPP ports) within the RAN and CN. The video
provider, IP phone user or web server is connected to a mobile network access point (e.g.,
GGSN) by using a 100BaseT duplex link at 100 Mbps.
Radio Links
Unlike point-to-point links, radio links are not statically represented; that is, they
cannot be seen in the network model. Instead, radio links are dynamically established
during simulation. Radio links exist between any radio transmitter-receiver channel pair,
but establishing a link depends on many physical characteristics of the components
involved, as well as time-varying parameters. During simulation, parameters such as
frequency band, modulation type, transmitter power, distance, and antenna directionality
are common factors that determine whether a radio link exists at a particular time or can
ever exist.
OPNET uses the radio transceiver pipeline model, shown in Figure 5-9, to model
wireless transmission of packets. It consists of thirteen stages. The attributes of each stage
are configured as shown in Figure 5-10. The stage 10 to stage 13 in the radio pipeline
model which are related to wireless BER will be discussed in Section 5.2. For more
details of each stage in the radio pipeline model, please refer to the OPNET Modeler 9.0
documents [77].
82

Figure 5-9 Radio Transceiver Pipeline Model

Figure 5-10 Radio Transceiver Attributes for Specifying Pipeline Stages
83
5.2 System Setup and Test Conditions
5.2.1 Rate Shaping (Filtering) and Packetization
The purpose of rate shaper (filter) [1] [2] [3] [23] includes: optimization of
bandwidth usage, adoption of filter for handling client heterogeneity, network
heterogeneity, and optimizations in the retrieval of stored media.


Figure 5-11 Level of rate shaping
In Figure 5-11, the points at which rate shaping can be performed on the compressed
bit-stream are illustrated.
Region A is the uncompressed raw video where rate shaping can be preformed
relatively simply, e.g., resizing/stretching, but the amount of data that has to be processed
is very large due to its uncompressed nature.
In region B, the data is the same size as the raw data, but if the bit-stream is being
decompressed in order to accomplish a particular rate shaping and then recompressed (e.g.,
re-quantization filtering), performing filtering at this point saves completing the
computationally intensively functions of forward-DCT (FDCT) and inverse-DCT (IDCT).
In region C, the many zeros produced by the FDCT have been removed and the data
is considerably smaller, the functions of frequency filtering that are feasible at this point
84
include: low-pass filtering, color-reduction filtering, color to monochrome conversion
and simple coder-conversions. The current transcoder which is designed for speed
therefore operates in this region.
Compared with region A, B and C, rate shaping (e.g., frame dropping, layer dropping)
on the fully compressed data in region D is standard specific and relatively simple. We
employ the techniques of rate shaping in the sender directly to video distribution services as
network filtering used in the BSs. For simplicity of simulation, we packetize each frame of
every layer into one packet in order to mimic the principle of a frame dropper in the BSs.
5.2.2 BER over Rayleigh Fading Channels
As mentioned previous, the BER on wireless channels is computed at the BER stage
(stage 11) in the radio pipeline model. In general, the bit error rate provided by this stage
is a function of the type of modulation used for the transmitted signal. This stage
evaluates BER based on the previously computed average SNR and also accounts for
processing gain at the receiver.
The SNR (in dB) is given by
SNR = 10 log [ P
r
/

(N
b
+N
i
) ]
where P
r
= received power (Watts), which is computed in stage 7;
N
b
= background noise power (Watts), which is computed in stage 8;
N
i
= interference noise power (Watts), which is computed in stage 9.
The SNR value is added to the processing gain (also in dB) to obtain the effective
SNR. This effective SNR is also written as E
b
/ N
0
where E
b
is the received energy per bit
(in Joules) and N
0
is the noise power spectral density (in Watts/hertz). The bit error rate is
derived from the effective SNR based on the QPSK (downlink) / BPSK (uplink)
modulation curve assigned to the receiver in a Rayleigh fading channel. The probability
of bit error P
b
(i.e., BER) is given by [52]
( )
1
1
]
1

+

0
0
0
1
1
2
1
N E
N E
N E P
b
b
b b

Figure 5-12 shows the BER curves for QPSK (downlink) and BPSK (uplink) in a
Rayleigh fading channel and an AWGN channel.
85

Figure 5-12 BER performance over flat Rayleigh fading channels
Following the BER stage, the error allocation stage (stage 12) translates the given
BER computed in stage 11 into an actual set of bit errors for each valid packet which is
received. The approach taken in stage 12 is to avoid sequencing through all the bits in the
packet and, therefore, not to generate positional information about bit errors, but to still
accurately compute the overall number of bit errors. This should be done with the
minimum number of computations for simulation efficiency considerations.
The error allocation algorithm is based on the expression for the probability of
exactly k bit errors occurring in a packet segment of length N. This probability is denoted
as P
k
. Under the assumption of independent bit errors, given the wireless channel bit error
probability P
b
(i.e., BER), P
k
can be expressed by
r P
k
N
P P P
N
k
k
k N
b
k
b k

,
_

0
) 1 (

The algorithm first generates a random number r between 0 and 1, to provide a value
against which the probability of occurrences of different numbers of bit errors will be
86
tested. Next the algorithm begins iterations. First the probability that 0 bit error occurs is
computed according to the formula above, and compared to r. If r is lower than this
probability, then 0 is the number of bit errors in the packet. Else, the probability of
occurrence of 1 or fewer bit errors is computed. If r is less than this probability, yet
higher than the probability of occurrence of the previous number of bit errors, then 1 is
the number of bit errors allocated to the packet. The algorithm continues iterating in this
manner until a value k is found for which the probability of k or fewer errors occurring is
greater than r. Then the number of errors assigned to the packet is k.
The purpose of error collection stage (stage 13) is to determine whether or not the
arriving packet can be accepted in the destination node. This is usually dependent upon
whether the packet has experienced collisions, the result computed in the error allocation
stage, and the capability of the receiver to correct the errors affecting the packet. In our
simulation, there are no error collection techniques implemented except for ARQ.
The following three cases in Table 5-1 are tested for the evaluation of the
relationship between BER and frame error rate.
Table 5-1 Test Cases of BER and FER in wireless channels
Test Case Channel Model Modulation Multiple Access BER
Rayleigh Fading QPSK (DL), BPSK(UL) DS-CDMA [10
-5
, 2 10
-5
]
Rayleigh Fading QPSK (DL), BPSK(UL) DS-CDMA [10
-4
, 2 10
-4
]

Rayleigh Fading QPSK (DL), BPSK(UL) DS-CDMA [.6 10
-3
, 10
-3
]
5.2.3 Delay-constrained ARQ
To enhance the video quality in the presence of unavoidable packet loss or bit errors,
error control mechanisms have been proposed. Basically, error control approaches can be
broadly categorized as open- loop error control (e.g., SMDC, error resilience tools in
MPEG-4, error concealment) and close- loop error control, (e.g., delay-constrained ARQ).
However, because of the complexity of open- loop error control, multiple description
coding, error resilience tools in MPEG-4 and error concealment can not be emulated in
this simulation. Instead, delay-constrained ARQ is under consideration.
87
According to the decision- making points, delay-constrained retransmission can be
sender-based, receiver-driven, or hybrid. Since the computational complexity is limited at
the mobile handset, the sender-based control at the BS is chosen to suppress
retransmission of packets that will miss their display time at the mobile user.
Given the maximum transfer delay D
max
in UMTS, the wired link delay D
wired
and
the wireless link RTT
wireless
, the maximum number of retransmissions N
ARQ
allowed for a
video packet is approximately given by

'

>
1
]
1

wired
wired
wireless
wired
ARQ
D D if
D D if
RTT
D D
N
max
max
max
, 0
,

Typically, the wireless link RTT
wireless
is about several milliseconds. The BS can
calculates the wired link delay D
wired
from Video Provider to the BS by recording the
time T when a RTCP sender report (SR) is received, and then subtracting the value of
NTP timestamp field in SR to obtain D
wired
= (T - NTP).
5.2.4 Traffic Profile
In order to evaluate the effectiveness of the proposed IP DiffServ MPEG-4 video
marking algorithm and the streaming video handoff procedures under MDMN, two
groups of traffic parameters have been set up and are listed in Table 5-2 and Table 5-4,
respectively.
Traffic Profile for Evaluation of Video Marking Algorithm
The video traffic, shown in Figure 5-14, is made up of Class I, Class II and Class III
layer-coded video streams, shown in Figure 5-13, which are classified through the
proposed IP DiffServ MPEG-4 video marking algorithm. The video traffic which is
generated by OPNET has to be subjected to the requirements of UMTS bearer service
attributes of streaming class [61], listed in Table 5-3.
The background traffic in each cell is composed of one video client, 25 voice users,
and web users, shown in Figure 5-15 ~ Figure 5-17, respectively. Under the condition of
background traffic, the serious congestion will occur in the BSs due to the limitation and
fluctuation of wireless bandwidth.
88
Table 5-2 Traffic profile of each cell for evaluation of video marking algorithm
Traffic Type User Number Sending Rate (mean) Packet Length (mean) Standard
Video Stream* 2 240 kbps per client 1 Kbytes MPEG-4
Voice Stream 25 16 kbps per user 200 Bytes G.728
Web Traffic - 400 kbps 1 Kbytes HTTP
* Frame Rate (mean) = 20 frame/s, Frame Length (mean) = 1 Kbytes.
Table 5-3 UMTS bearer service attributes of streaming class
Maximum bit rate (outdoor) 384 kbps
Maximum SDU size 1500 Bytes
SDU error ratio (i.e., Frame error rate) 10

%
Transfer delay 280 ms
Delay jitter 50 ms (Frame Rate = 20 frame/s)

Traffic Profile for Evaluation of Streaming Video Handoff
Table 5-4 Traffic profile of each cell for evaluation of streaming video handoff
Traffic Type User Number Sending Rate (mean) Packet Length (mean) Standard
Video Stream* 2 240 kbps per client 1 Kbytes MPEG-4
Voice Stream 5 16 kbps per user 200 Bytes G.728
Web Traffic - 400 kbps 1 Kbytes HTTP
* Frame Rate (mean) = 20 frame/s, Frame Length (mean) = 1 Kbytes.
The video traffic, shown in Figure 5-18, is also made up of Class I, Class II and
Class III layer-coded video streams and has to be subjected to the requirements of UMTS
bearer service attributes of streaming class [61], listed in Table 5-3. The hard handoff will
occur 28 times during the testing period of 5 minutes.
The background traffic in each cell is composed of one video client, 5 voice users,
and web users, shown in Figure 5-19 ~ Figure 5-21, respectively. With the decrease of
background traffic, no congestion happens in the BSs during the handoff tests.
89

Figure 5-13 Layered Video Traffic (Video Client1)

Figure 5-14 Video Traffic (Video Client1)
90

Figure 5-15 Background Traffic (Video Client2)

Figure 5-16 Background Traffic (25 Voice Users)

Figure 5-17 Background Traffic (Web Users)
91

Figure 5-18 Video Traffic (Handoff occurs 28 times)

Figure 5-19 Background Traffic (Video Client s)

Figure 5-20 Background Traffic (5 Voice Users)

Figure 5-21 Background Traffic (Web Users)
92
5.2.5 Test Cases
AF Queue Size
To implement the proposed IP DiffServ video marking algorithm in each node, we
use WFQ discipline to classify and schedule the incoming packet into and out of EF,
AF1, AF2, AF3 or BE queue. The WFQ profile for the proposed IP DiffServ video
marking algorithm in each node is listed in Table 5-5. The selection of buffer size is
crucial to video over IP network. An optimum buffer size has to be found which balances
both end-to-end delay and packet loss ratio to tolerable levels. If the buffer is set too low,
some packets may be lost; if set too high, higher delays result.
Table 5-5 WFQ profile for proposed IP DiffServ video marking algorithm
Queue Scheduling
Queue
Classification
Queue Size
(1 unit = 100 ms)
Normalized
Bandwidth
BE Queue 15 4.5 %
AF3 Queue 1~9 9.1 %
AF2 Queue 1~9 13.7 %
AF1 Queue 1~9 22.7 %
WFQ
EF Queue 15 50 %

There are eighteen cases, listed in Table 5-6, under test in order to evaluate the effect
of AF queue size. The effect of AF queue size is analyzed in Section 5.2.5. After the
evaluation of the effect of AF queue size, we select 500 ms as the optimum queue size of
the AF1, AF2 and AF3 queue.
Video Marking Algorithm
In order to compare the proposed IP DiffServ video marking algorithm with DVMA,
three scenarios are experimented. The first one is the best effort model using a drop-tail
BE queue in each node. The second one is the DiffServ model where DVMA is used with
WRED AF queue. The third one is the DiffServ model where the proposed IP DiffServ
video marking algorithm is used with the drop-tail AF1, AF2 and AF3 queues. Table 5-7
lists the configuration of queue scheduling in these three scenarios. Table 5-8 gives the
93
WRED profile for AF Queue in the DVMA scenario. The test cases of video marking
algorithm are listed in Table 5-9.
Streaming Video Handoff
The performance of the proposed streaming video handoff in D-MDMN is examined
and compared with that in UMTS model under the scenario of the proposed IP DiffServ
video marking algorithm. The test cases of streaming video handoff are listed in Table 5-10.
Table 5-6 Test Cases of Effect of Queue Size
Test Case Physical Characteristics Rayleigh BER QoS Mechanism AF Queue Size
1 QPFK(DL),DS-CDMA 10
-5
DiffServ, WFQ 100 ms
2 QPFK(DL),DS-CDMA 10
-5
DiffServ, WFQ 200 ms
3 QPFK(DL),DS-CDMA 10
-5
DiffServ, WFQ 300 ms
4 QPFK(DL),DS-CDMA 10
-5
DiffServ, WFQ 400 ms
5 QPFK(DL),DS-CDMA 10
-5
DiffServ, WFQ 500 ms
6 QPFK(DL),DS-CDMA 10
-5
DiffServ, WFQ 600 ms
7 QPFK(DL),DS-CDMA 10
-5
DiffServ, WFQ 700 ms
8 QPFK(DL),DS-CDMA 10
-5
DiffServ, WFQ 800 ms
9 QPFK(DL),DS-CDMA 10
-5
DiffServ, WFQ 900 ms
(1) QPFK(DL),DS-CDMA 10
-4
DiffServ, WFQ 100 ms
(2) QPFK(DL),DS-CDMA 10
-4
DiffServ, WFQ 200 ms
(3) QPFK(DL),DS-CDMA 10
-4
DiffServ, WFQ 300 ms
(4) QPFK(DL),DS-CDMA 10
-4
DiffServ, WFQ 400 ms
(5) QPFK(DL),DS-CDMA 10
-4
DiffServ, WFQ 500 ms
(6) QPFK(DL),DS-CDMA 10
-4
DiffServ, WFQ 600 ms
(7) QPFK(DL),DS-CDMA 10
-4
DiffServ, WFQ 700 ms
(8) QPFK(DL),DS-CDMA 10
-4
DiffServ, WFQ 800 ms
(9) QPFK(DL),DS-CDMA 10
-4
DiffServ, WFQ 900 ms
94
Table 5-7 Queue scheduling configuration
Solution
Queue
Scheduling
Queue
Classification
Queue Size
(1 unit = 100ms)
Normalized
Bandwidth
Best Effort FIFO BE Queue 45 100 %
BE Queue 15 4.5 %
AF Queue 15 45.5 %
DVMA WRED+CBQ
EF Queue 15 50 %
BE Queue 15 4.5 %
AF3 Queue 5 9.1 %
AF2 Queue 5 13.7 %
AF1 Queue 5 22.7 %
Proposed WFQ
EF Queue 15 50 %
Table 5-8 WRED profile for AF queue in DVMA
Video Stream
Class
Exponential
Weight Factor
Min Threshold
(1 unit = 100 ms)
Max Threshold
(1 unit = 100 ms)
Mark Probability
Denominator
AF31 9 1 5 5
AF21 9 2 5 5
AF11 9 3 5 5
Table 5-9 Test Cases of Video Marking Algorithm
Test Case Physical Characteristics
Rayleigh
BER
QoS
Mechanism
Queue
Scheduling
Marking
Algorithm
A QPFK(DL), DS-CDMA 10
-5
Best Effort IP FIFO BE
B QPFK(DL), DS-CDMA 10
-5
IP DiffServ WRED DVMA
C QPFK(DL), DS-CDMA 10
-5
IP DiffServ WFQ Proposed
a QPFK(DL), DS-CDMA 10
-4
Best Effort IP FIFO BE
b QPFK(DL), DS-CDMA 10
-4
IP DiffServ WRED DVMA
c QPFK(DL), DS-CDMA 10
-4
IP DiffServ WFQ Proposed
95
Table 5-10 Test Cases of Handoff Procedures
Test
Case
Modulation
Rayleigh
BER
Network
Model
QoS Mechanism Handoff
I QPSK(DL), BPSK(UL) 10
-5
UMTS DiffServ, WFQ Intra-RAN
II QPSK(DL), BPSK(UL) 10
-5
MDMN DiffServ, WFQ Intra-RAN
III QPSK(DL), BPSK(UL) 10
-5
UMTS DiffServ, WFQ Inter-RAN
IV QPSK(DL), BPSK(UL) 10
-5
MDMN DiffServ, WFQ Inter-RAN
i QPSK(DL), BPSK(UL)
10
-4
UMTS DiffServ, WFQ Intra-RAN
ii QPSK(DL), BPSK(UL)
10
-4
MDMN DiffServ, WFQ Intra-RAN
iii QPSK(DL), BPSK(UL)
10
-4
UMTS DiffServ, WFQ Inter-RAN
iv QPSK(DL), BPSK(UL)
10
-4
MDMN DiffServ, WFQ Inter-RAN
5.3 Simulation Results and Analysis
5.3.1 Effect of BER on FER over Wireless Channels
Figure 5-22 ~ Figure 5-27 show the effect of BER on FER over the Rayleigh fading
channel. The FER measure indicates the difficulty of sending video stream over the
wireless channel. A small BER translates into a much higher FER. For example, the BER
mean value of 1.3610
-4
in Figure 5-23 can translate into the FER mean value of 3.16%
in Figure 5-26. In the test case with a BER mean value of 1.5710
-5
, there are no
frame errors showed in Figure 5-25. That is because we use the delay-constrained ARQ
to combat Rayleigh BER.
Given that the maximum frame error rate allowed in UMTS is 10%, the video delivery
in UMTS can tolerate Rayleigh BER up to 10
-4
with delay-constrained ARQ. That is reason
that we choose BER in the range of [10
-5
, 2 10
-5
]

and [10
-4
, 2 10
-4
] as our test conditions.
However, from Table 1-1, the BER of wireless video can be as high as 10
-3
. It is
desirable to employ the proposed scalable multiple description coding as a diversity
technique to combat the high BER in the wireless channel.
96

Figure 5-22 Bit Error Rate over the Rayleigh fading channel (Test Case )

Figure 5-23 Bit Error Rate over the Rayleigh fading channel (Test Case )

Figure 5-24 Bit Error Rate over the Rayleigh fading channel (Test Case )
97

Figure 5-25 FER over the Rayleigh fading channel (Test Case )

Figure 5-26 FER over the Rayleigh fading channel (Test Case )

Figure 5-27 FER over the Rayleigh fading channel (Test Case )
98
5.3.2 Effect of AF Queue Size
Figure 5-28 ~ Figure 5-31 illustrate the effect of the AF queue size on packet end-to-
end delay and packet loss ratio of AF queue at the BSs. The video traffic begins at time
25 s and the background traffic starts at time 36 sec. From time 25 s to 36 s, the packet
arriving rate at BS is lower than the queue service rate and there is no congestion in the
AF queues. At time 36 s, the packet arriving rate at BS suddenly increases due to the
background traffic and exceeds the queue service rate. The AF queues continue growing
until time 50 s and start to drop frames.
As the AF queue size increases, the packet end-to-end delay increases, but the
packet loss ratio of AF queue decreases. Compared the results of total nine test cases, we
choose queue size 500 ms as the optimum value which balances both end-to-end delay
and packet loss ratio to tolerable levels for further simulation setting.
5.3.3 Effect of Video Marking Algorithm
Performance of Frame Error Rate
MPEG-4 video stream in UMTS can tolerance SDU error rate (i.e., FER) up to 10%.
The effects of different video marking algorithms on FER are shown in Figure 5-32 ~
Figure 5-37.
In the scenario of BER = 10
-5
, the FERs are caused by the packet loss at the BSs due
to congestion. In the scenario of BER = 10
-4
, the FERs are caused by both the packet loss
at the BSs due to congestion and the wireless channel bit errors. Due to the employment
of WRED for proactive packet-dropping in DVMA, the packet loss in DVMA begins
earlier than that in the proposed solution. However, the packet loss in the tail-dropping
BE solution occurrs even earlier than that in DVMA. That is because the background
traffic and the video traffic enter into the same queue, and cause the congestion happened
earlier. Note that the background traffic and the video traffic are separated in different
queues in DiffServ-based solutions.
Since we re-organize the shape, motion, and texture video information into different
layers, UEP can be introduced and results in differentiated service. If a bit error or a
99
packet loss occurs in the MPEG-4 Class I stream, the corresponding bits or packets in the
MPEG-4 Class II and Class III streams have to be considered erroneous or lost. Similarly,
if a bit error or a packet loss occurs in the MPEG-4 Class II stream, the corresponding
bits or packets in the MPEG-4 Class III stream also have to be considered erroneous or
lost. Some packets may arrive late and will also be considered lost. If the higher priority
traffic is protected, less packet loss (i.e., FER) will occur.
Compared with BE and DVMA, the protection of both voice stream and MPEG-4
Class I stream in the proposed solution is the best. This also results in the least FER
among all the scenarios. In addition, the protection of EF traffic in DVMA is better than
that in BE. However, this is at the cost of high FER of AF traffic in DVMA.
A brief comparison of QoS (e.g., FER) guarantee in the three solutions can be
summarized in Table 5-11. Note that: Unacceptable QoS means that the performance of
FER can not meet the QoS requirement all the time; Unpredictable QoS is a typical
characteristic of best-effort traffic; Unwarranted QoS means that the performance of
FER meets the QoS requirement sometimes and is predictable; Guaranteed QoS means
that the performance of FER meets the QoS requirement all the time and is predictable.
Table 5-11 Comparison of QoS (e.g., FER) guarantee in three solutions
Traffic BE DVMA Proposed
Voice Stream Unacceptable QoS Unwarranted QoS Guaranteed QoS
MPEG-4 Class I Unpredictable QoS Unwarranted QoS Guaranteed QoS
MPEG-4 Class II Unpredictable QoS Unwarranted QoS Unwarranted QoS
MPEG-4 Class III Unpredictable QoS Unwarranted QoS Unwarranted QoS

Performance of End-to-end Delay and Delay Jitter
The maximum end-to-end delay and delay jitter of video streaming allowed in the
simulation are 280 ms and 50 ms, respectively, under the test conditions. The effects of
different video marking algorithms on end-to-end delay and delay jitter are presented in
Figure 5-38 ~ Figure 5-49.
In the BE solution and the DVMA solution, the end-to-end delay and delay jitter of
video traffic in different classes can not be differentiated. That is because different
100
classes video streams go through the same queue (e.g., BE queue or AF queue). As we
expect in the proposed solution, the performance of MPEG-4 Class I stream is better than
Class II; and Class II is better than Class III.
In the scenario of BER = 10
-4
, with a sudden increase of the background traffic at
time 36 s, the end-to-end delay and delay jitter of video streams jump sharply up to a
higher level. The proposed solution delays 10 s the start point of performance degradation
compared with BE and DVMA. After the BSs begin to drop packets, the performance of
end-to-end delay and delay jitter turns better. In DVMA, the video delay jitters of all
three classes are not acceptable, though Class III and Class II streams in DMVA are
better than those in the proposed solution. In comparison, the MPEG-4 Class III and
Class II stream in the proposed solution are sacrificed in order to guarantee the QoS of
the Class I stream.
The performance tendency is similar in the scenario of BER = 10
-5
. Note that only
MPEG-4 Class III stream in the proposed solution suffers from high delay and delay jitter
in order to preserve the Class I and Class II streams. However, in DVMA, all three
classes are not acceptable until about time 1m40s. The results in the scenario of BER =
10
-5
is better than their counterparts in the scenario of BER = 10
-4
. This is because the
numbers of ARQ in BER = 10
-5
is less than that in BER = 10
-4
.
In both BER scenarios, the performance of delay and delay jitter in BE seems better
than the other two solutions. However, this is at the cost of unacceptable FER in the voice
stream and unpredictable quality (e.g., FER) of streaming video service.
5.3.4 Effect of Streaming Video Handoff
As discussed previously, the maximum end-to-end delay and delay jitter of video
streaming allowed in both UMTS and MDMN are 280 ms and 50 ms, respectively, under
the test conditions. The handoff latency also should keep below 50 ms as a delay jitter.
Figure 5-50 ~ Figure 5-57 show the performance of end-to-end delay and delay jitter
in the scenario of BER = 10
-5
. Figure 5-58 ~ Figure 5-65 illustrate the performance of
end-to-end delay and delay jitter in the scenario of BER = 10
-4
.
In UMTS, only the performance of maximum end-to-end delay in test case iii (Inter-
RAN Handoff, BER = 10
-4
) exceeds 280 ms. Furthermore, in all the UMTS test cases, the
101
handoff latency can not satisfy the 50 ms delay bound. Note that, because the distance
between the Video Provider and GGSN is unknown, we choose the best case in the
UMTS simulation model. That is, the distance between them is only one hop. If the
Video Provider is far from the UMTS core network, the performance of maximum end-
to-end delay in UMTS may not meet the QoS requirement of UMTS any more.
In comparison, with the proposed stream re-establishing handoff solution, all the test
cases in MDMN meet the performance requirement of the maximum end-to-end delay
and handoff latency. Because the Video Provider is distributed as the media databases
inside the core network, the handoff performance keeps stable in MDMN and does not
have the distance problem as mentioned above in UMTS.
A brief comparison of handoff performance in UMTS and MDMN is summarized in
Table 5-12. This comparison shows that the improvement of handoff performance
ascends with the increase of the scale of the mobile core network. For example, the
handoff latency improvement in the intra-RAN scale is 26 ms or 45 ms under different
conditions of wireless BER, but in the inter-RAN scale it is 38 ms or 57 ms. Furthermore,
the proposed stream re-establishing handoff performance in MDMN is relatively
consistent in all scenarios. This validates the enhancement of handoff scalability in
MDMN.
Table 5-12 Comparison of handoff performance in UMTS and MDMN
Test Case
End-to-end Delay
(mean)
Delay
Handoff Latency
(mean)
Handoff
Latency
I (UMTS, Intra-RAN) 74 ms 68 ms
II (MDMN, Intra-RAN) 42 ms
32 ms
42 ms
26 ms
III (UMTS, Inter-RAN) 75 ms 91 ms
IV (MDMN, Inter-RAN) 45 ms
30 ms
46 ms
45 ms
i (UMTS, Intra-RAN) 79 ms 82 ms
ii (MDMN, Intra-RAN) 45 ms
34 ms
44 ms
38 ms
iii (UMTS, Inter-RAN) 131 ms 104 ms
iv (MDMN, Inter-RAN) 56 ms
75 ms
47 ms
57 ms
102
In this chapter, the simulation models, system setup, test conditions, and simulation
results are presented and analyzed. Section 5.1 describes the UMTS and the proposed D-
MDMN simulation model, and the simulation pipeline stage of the radio transceiver. In
Section 5.2, we discuss the simulation and design issues, such as rate shaping and
packetization, BER over Rayleigh fading channels, delay-constrained ARQ, followed by
the description of traffic profile and test cases. Section 5.2.5 first evaluates the effect of
BER on FER over wireless channels and the effect of AF queue size for optimization of
system setup. Then performance evaluation of the proposed IP DiffServ video marking
algorithm is undertaken to show that it is more suitable for video streaming in IP mobile
networks compared with DVMA solution. Finally, the simulation analysis is concluded
by the handoff performance comparison of UMTS versus D-MDMN, indicating that the
proposed handoff procedures in D-MDMN have better performance in terms of handoff
latency, end-to-end delay and handoff scalability than that in UMTS.
103

Figure 5-28 Average of End-to-End Delay (Test Case 1~9: BER = 10
-5
)

Figure 5-29 Average of Packet Loss Ratio (Test Case 1~9: BER = 10
-5
)
104

Figure 5-30 Average of End-to-End Delay (Test Case (1)~(9): BER = 10
-4
)

Figure 5-31 Average of Packet Loss Ratio (Test Case (1)~(9): BER = 10
-4
)
105

Figure 5-32 FER (Test Case A: Best Effort, FIFO, BER = 10
-5
)

Figure 5-33 FER (Test Case B: DiffServ, WRED, BER = 10
-5
)
106

Figure 5-34 FER (Test Case C: DiffServ, WFQ, BER = 10
-5
)

Figure 5-35 FER (Test Case a: Best Effort, FIFO, BER = 10
-4
)
107

Figure 5-36 FER (Test Case b: DiffServ, WRED, BER = 10
-4
)

Figure 5-37 FER (Test Case c: DiffServ, WFQ, BER = 10
-4
)
108

Figure 5-38 End-to-end Delay (Test Case A: Best Effort, FIFO, BER = 10
-5
)

Figure 5-39 End-to-end Delay (Test Case B: DiffServ, WRED, BER = 10
-5
)

Figure 5-40 End-to-end Delay (Test Case C: DiffServ, WFQ, BER = 10
-5
)
109

Figure 5-41 End-to-end Delay (Test Case a: Best Effort, FIFO, BER = 10
-4
)

Figure 5-42 End-to-end Delay (Test Case b: DiffServ, WRED, BER = 10
-4
)

Figure 5-43 End-to-end Delay (Test Case c: DiffServ, WFQ, BER = 10
-4
)
110

Figure 5-44 Delay Jitter (Test Case A: Best Effort, FIFO, BER = 10
-5
)

Figure 5-45 Delay Jitter (Test Case B: DiffServ, WRED, BER = 10
-5
)

Figure 5-46 Delay Jitter (Test Case C: DiffServ, WFQ, BER = 10
-5
)
111

Figure 5-47 Delay Jitter (Test Case a: Best Effort, FIFO, BER = 10
-4
)

Figure 5-48 Delay Jitter (Test Case b: DiffServ, WRED, BER = 10
-4
)

Figure 5-49 Delay Jitter (Test Case c: DiffServ, WFQ, BER = 10
-4
)
112

Figure 5-50 End-to-End Delay (Test Case I: UMTS, Intra-RAN, BER = 10
-5
)

Figure 5-51 Delay Jitter (Test Case I: UMTS, Intra-RAN, BER = 10
-5
)
113

Figure 5-52 End-to-End Delay (Test Case II: MDMN, Intra-RAN, BER = 10
-5
)

Figure 5-53 Delay Jitter (Test Case II: MDMN, Intra-RAN, BER = 10
-5
)
114

Figure 5-54 End-to-End Delay (Test Case III: UMTS, Inter-RAN, BER = 10
-5
)

Figure 5-55 Delay Jitter (Test Case III: UMTS, Inter-RAN, BER = 10
-5
)
115

Figure 5-56 End-to-End Delay (Test Case IV: MDMN, Inter-RAN, BER = 10
-5
)

Figure 5-57 Delay Jitter (Test Case IV: MDMN, Inter-RAN, BER = 10
-5
)
116

Figure 5-58 End-to-End Delay (Test Case i: UMTS, Intra-RAN, BER = 10
-4
)

Figure 5-59 Delay Jitter (Test Case i: UMTS, Intra-RAN, BER = 10
-4
)
117

Figure 5-60 End-to-End Delay (Test Case ii: MDMN, Intra-RAN, BER = 10
-4
)

Figure 5-61 Delay Jitter (Test Case ii: MDMN, Intra-RAN, BER = 10
-4
)
118

Figure 5-62 End-to-End Delay (Test Case iii: UMTS, Inter-RAN, BER = 10
-4
)

Figure 5-63 Delay Jitter (Test Case iii: UMTS, Inter-RAN, BER = 10
-4
)
119

Figure 5-64 End-to-End Delay (Test Case iv: MDMN, Inter-RAN, BER = 10
-4
)

Figure 5-65 Delay Jitter (Test Case iv: MDMN, Inter-RAN, BER = 10
-4
)
120
6 Conclusions and Future Work
To address the handoff problems in video streaming, as well as the bandwidth
fluctuation, packet loss and heterogeneity problems in the wireless networks, and to
further enhance the error resilience tools in MPEG-4, the 3G mobile network architecture
and the handoff procedures for video delivery in UMTS are studied.
The contributions of this thesis are:
1) A Scalable Multiple Description Coding framework is proposed to explore the
joint design of layered coding and multiple description coding.
Under the SMDC framework, MDC components enhance the robustness to losses
and bit errors of LC components through path diversity and error recovery. MDC
components also reduce the storage, reliability and load balancing requirement among
distributed media edge servers. At the same time, LC components not only deal with the
unbalanced MD operation at the server end, but also combat the bandwidth frustrations of
the time- varying wireless channel. Furthermore, SMDC leverages the distributed
multimedia delivery mobile network to provide path diversity to combat video streaming
outage due to handoff.
2) A Distributed Multimedia Delivery Mobile Network is proposed for the UMTS
core network.
D-MDMN introduces and combines the concepts of CDN and SMDC into the
UMTS network in order to solve the video handoff problem and meet the stringent QoS
requirements of video streaming in 3GPP. Since the media streaming services are pushed
to the edge of core network, it also reduces the media service delivery time, the
probability of packet loss, and the total network resource occupation with relatively
consistent QoS in all scenarios.
3) Handoff procedures of video streaming in D-MDMN are proposed.
The proposed handoff procedures employ the principle of video stream re-
establishing to replace the principle of data forwarding in UMTS. The intra-RAN handoff
and inter-RAN handoff procedures are studied in details.
121
4) A novel IP DiffServ video marking algorithm is proposed to support the unequal
error protection of LC components of SMDC.
The proposed algorithm re-organizes the shape, motion, and texture information of
video stream into different layers in the proposed SMDC scheme to implement the
DiffServ mobile network in UMTS. Furthermore, it spurs the evolution of UMTS toward
its final all-IP phase for the purpose of addressing the DiffServ tunneling issue in UMTS.
The above proposed schemes have been validated through the simulation presented
in Chapter 5, except that the verification of MDC components of SMDC can not be
undertaken because of technical complexity and time limitation.
The limitations and cost of the proposed schemes also can be summarized as
follows.
1) The significant performance improvement of SMDC is achieved at the cost of a
coding overhead.
2) The D-MDMN network solution is feasible as a client-server solution for video
streaming delivery service, but not for end-to-end real- time video conversation.
3) For video streaming delivery, the video descriptions should be distributed in
advance into all complementary MDSs at the edge of the RANs according to the service
subscription of video mobile users, which adds the complexity to UMTS. A potential
video adaptive deployment solution can be given as follow.
Firstly, during the CAC, the video service subscribed by an MS are distributed to the
MDSs in the current local RAN and all its neighbor RANs. Secondly, if the MS moves to
another RAN, this video service should be simultaneously distributed to the MDSs in all
its new neighbor RANs after the handoff successfully takes place.
Further work will include the verification of the proposed SMDC under the object-
based MPEG-4 video stream, especially the MDC components over the Rayleigh fading
channel, the study of the effect of SMDC coding overhead and the soft handoff
procedures in D-MDMN. In addition, the proposed solutions of video streaming may be
applicable to audio streaming. The joint design of audio LC and MDC, the distributed
audio delivery network, the IP DiffServ audio marking algorithm, and the media
synchronization between video stream and audio stream should be studied further.
122
Appendix
Proposed inter-cell, intra-RNS handoff procedure
In the scenario of inter-cell, intra-RNS handoff, the proposed handoff procedures,
shown in Figure 0-1, consists of three phase:
l Phase I: Preparation of BTS handoff and resource allocation
l Phase II: Moving the Serving BTS role to target BTS
l Phase III: Releasing resource reservation in the old path


Figure 0-1 Proposed inter-cell, intra-RNS handoff procedure (Control plane)
123
Acronyms
3GPP Third Generation Partnership Project
ABR Available Bit Rate
AF Assured Forwarding
ALP Application Level Packets
AMR Adaptive Multi Rate
AR Access Router
ARQ Automatic Repeat reQuest
ATM Asynchronous Transfer Mode
BE Best Effort
BER Bit Error Rate
B-frame Bi-directionally predicted frame
BIFS Binary Format for Scene
BPSK Binary Phase Shift Keying
BR Border Router
BS Base Station
BSC Base Station Controller
BSS Base Station Subsystem
BTS Base Transceiver Station
CAC Call Admission Control
CBQ Class Based Queuing
CBR Constant Bit Rate
CCIR Consultative Committee for International Radiocommunication
CDMA Code Division Multiple Access
CDN Content Delivery Network
CIF Common Interleaved Frame
CLR Cell Loss Rate
CN Core Network
124
CoS Class of Service
CS Circuit Switched
DC Direct Current
DCT Discrete Cosine Transform
DiffServ Differentiated Services
DL Down Link
D-MDMN Distributed Multimedia Delivery Mobile Network
DS DiffServ
DSCP DiffServ CodePoint
DVMA DiffServ video marking algorithm
EDGE Enhanced Data rates for GSM Evolution
EF Expedited Forwarding
ER Edge Router
FDCT Forward-DCT
FDD Frequency Division Duplex
FEC Forward Error Correction
FER Frame Erasure Rates
FGS Fine granularity scalability
FP Frame relay transport Protocol
GERAN GSM/EDGE radio access network
GFR Guaranteed Frame Rate
GGSN Gateway GPRS Support Node
GOB Group Of Blocks
GOP Group Of Pictures
GPRS General Packet Radio Service
GSM Global System for Mobile communications
GTP GPRS Tunneling Protocol
HLR Home Location Register
IDCT Inverse-DCT
IEEE Institute of Electrical and Electronics Engineers
IETF Internet Engineering Task Force
125
I-frame Intra-coded frame
IP Internet Protocol
ISDN Integrated Services Digital Network
ITU International Telecommunication Union
I-VOP Intra-coded Video Object Plane
LAN Local Area Network
LC Layered Coding
MC-CMDA Multiple Carrier CDMA
MD Multiple Description
MDC Multiple Description Coding
MDS Media Description Server
MPEG Moving Pictures Experts Group
MS Mobile Station
MSC Mobile Switching Center
nrt-VBR non real-time Variable Bit Rate
NSS Network Subsystem
OD Object Descriptor
OFDM Orthogonal Frequency Division Multiplexing
PDU Protocol Data Unit
PFGS Progressive Fine Granularity Scalability
PFGST PFGS Temporal
P-frame Predictively coded frame
PHB Per-Hop Behavior
PS Packet Switched
PSS Packet-switched Streaming Service
PSTN Public Switched Telephone Network
P-VOP Predictively coded Video Object Plane
QCIF Quarter Common Interleaved Frame
QoS Quality of Service
QPSK Quadrature Phase Shift Keying
RAN Radio Access Network
126
RED Random Early Detection
RFC Request For Comments
RFL Radio Frequency Layer
RNC Radio Network Controller
RNL Radio Network Layer
RNS Radio Network Subsystems
RPS Reference Picture Selection
RS Redirection Server
RTCP Real Time Control Protocol
RTP Real Time Protocol
RTSP Real Time Streaming Protocol
RTT Round Trip Time
SD Single Description
SDP Session Description Protocol
SDU Service Data Unit
SGSN Serving GPRS Support Node
SIP Initiation Protocol
SLA Service Level Agreement
SMDC Scalable Multiple Description Coding
SNR Signal- to-Noise Ratio
sRNS source RNS
sSGSN source SGSN
TCP Transmission Control Protocol
TD-CDMA Time Division-CDMA
TDD Time Division Duplex
tRNS target RNS
tSGSN target SGSN
TTI Transmission Time Interval
UDP User Datagram Protocol
UE User Equipment
UEP Unequal Error Protection
127
UL Up Link
UMTS Universal Mobile Telecommunications System
URL Uniform Resource Locator
UTRAN UMTS Terrestrial Radio Access Network
VLR Visitor Location Register
VO Video Object
VoIP Voice over IP
VOP Video Object Planes
VP Video Packet
WAN Wide Area Network
WCDMA Wideband-CDMA
WFQ Weighted Fair Queuing
WRED Weighted Random Early Detection
128
References
[1] Dapeng Wu, Y. T. Hou, Wenwu Zhu, Ya-Qin Zhang, Jon M. Peha, Streaming
Video over the Internet: Approaches and Directions, IEEE Trans. on Circuits
and Systems for Video Technology, Vol. 11 No. 3 , pp. 282 -300, Mar. 2001
[2] Dapeng Wu, Y. T. Hou, Ya-Qin Zhang, Transporting Real- Time Video over the
Internet: Challenges and Approaches, Proc. of the IEEE , Vol. 88, No. 12 , pp.
1855 -1877, Dec. 2000
[3] Frank G. Lin, Transparent MPEG Video Filtering for Wireless Networks,
Masters thesis, University of Waterloo, Canada, 2000.
[4] S. Floyd, and V. Jacobson, Random Early Detection Gateways for Congestion
Avoidance, IEEE/ACM Trans. on Networking, V.1 N.4, pp. 397-413,Aug. 1993.
[5] G. Blakowski, R. Steinmetz, A Media Synchronization Survey: Reference Model,
Specification, and Case Studies, IEEE Journal on Selected Areas in
Communications, Vol. 14, No. 1, pp. 5 -35, Jan. 1996
[6] H.M. Radha, M. van der Schaar, Yingwei Chen, The MPEG-4 Fine- grained
Scalable Video Coding Method for Multimedia Streaming over IP, IEEE Trans.
on Multimedia, Vol. 3, No. 1 , pp. 53 -68, Mar. 2001
[7] Weiping Li, Overview of Fine Granularity Scalability in MPEG-4 Video
Standard, IEEE Trans. on Circuits and Systems for Video Technology, Vol. 11,
No. 3 , pp. 301 -317, Mar. 2001
[8] Weiping Li, Fan Ling, Xuemin Chen, Fine Granularity Scalability in MPEG-4
for Streaming Video, Proc. IEEE Int. Symp. on Circuits and Systems, Vol. 1, pp.
299 -302 , 2000
[9] Transparent End-to-end Packet Switched Streaming Service (PSS); Protocols
and codecs, 3GPP TS 26.234 V5.2.0, Sept. 2002
[10] Sumit Roy, Bo Shen, Vijay Sundaram, Application Level Hand-off Support for
Mobile Media Transcoding Sessions, Pro. of the 12th Int. Workshop on Network
and Operating Systems Support for Digital Audio and Video, May 2002
129
[11] Roger Karrer, Thomas Gross, Dynamic Handoff of Multimedia Streams, Proc.
of the 11th Int. Workshop on Network and Operating Systems Support for Digital
Audio and Video, Jan. 2001
[12] J. Kangasharju, F. Hartanto, M. Reisslein, K.W. Ross, Distributing Layered
Encoded Video through Caches, IEEE Trans. on Computers, Vol. 51, No. 6, pp.
622 -636, June 2002
[13] A.K. Katsaggelos, L.P. Kondi, F.W. Meier, J. Ostermann, G.M. Schuster,
MPEG-4 and Rate-distortion-based Shape-coding Techniques, Proc. of the
IEEE , Vol. 86, No. 6, pp. 1126 -1154, June 1998
[14] H. Schulzrinne, A. Rao, R. Lanphier, Real Time Streaming Protocol (RTSP) ,
IETF RFC 2326, Apr. 1998.
[15] H. Schulzrinne, S. Casner, R. Frederick, V. Jacobson, RTP: A Transport Protocol
for Real-Time Applications, IETF RFC 1889, Jan. 1996.
[16] Jiangchuan Liu, Huai-Rong Shao, Bo Li, Wenwu Zhu, Ya-Qin Zhang, A
Scalable Bit-stream Packetization and Adaptive Rate Control Framework for
Object-based Video Multicast, Proc. of IEEE Global Telecommunications Conf.,
Vol. 3, pp. 2020 -2025, 2001
[17] Feng Wu, Shipeng Li, Ya-Qin Zhang, A Framework for Efficient Progressive
Fine Granularity Scalable Video Coding, IEEE Trans. on Circuits and Systems
for Video Technology, Vol. 11, No. 3, pp. 332 -344, Mar. 2001
[18] S. Wenger, G.D. Knorr, J. Ott, F. Kossentini, Error Resilience Support in
H.263+, IEEE Trans. on Circuits and Systems for Video Technology, Vol. 8, No.
7, pp. 867 -877, Nov. 1998
[19] Dapeng Wu, Y.T. Hou, Ya-Qin Zhang, Scalable Video Transport over Wireless
IP Networks, Proc. of IEEE Int. Symp. on Indoor and Mobile Radio
Communications, Vol. 2, pp. 1185 -1191, 2000
[20] R. Aravind, M.R. Civanlar, A.R. Reibman, Packet Loss Resilience of MPEG-2
Scalable Video Coding Algorithms, IEEE Trans. on Circuits and Systems for
Video Technology, Vol. 6, No. 5, pp. 426 -435, Oct. 1996
130
[21] A. Ortego, K. Ramchandran, Rate-distortion Methods for Image and Video
Compression, IEEE Signal Processing Magazine, Vol. 15, No. 6, pp. 23 -50,
Nov. 1998
[22] J. Padhye, V. Firoiu, D.F. Towsley, J.F. Kurose, Modeling TCP Reno
Performance: A Simple Model and its Empirical Validation, IEEE Trans. on
Networking, Vol. 8, No. 2, pp. 133 -145, Apr. 2000
[23] Qian Zhang, Wenwu Zhu, Ya-Qin Zhang, Network-adaptive Scalable Video
Streaming over 3G Wireless Network, Proc. of IEEE Int. Conf. on Image
Processing, Vol. 2, pp. 579 -582, 2001
[24] S. Floyd, K.Fall, Promoting the Use of End-to-end Congestion Control in the
Internet, IEEE Trans. on Networking, Vol. 7, No. 4, pp. 458 -472, Aug. 1999
[25] W.E. Leland, M.S. Taqqu, W. Willinger, D.V. Wilson, On the Self-similar
Nature of Ethernet Traffic, IEEE Trans. on Networking, Vol. 2, No. 1, pp. 1 -15,
Feb. 1994
[26] John G. Apostolopoulos, Susie J. Wee, Unbalanced Multiple Description Video
Communication Using Path Diversity, Proc. of Int. Conf. on Image Processing,
Oct. 2001
[27] M. Gallant, F. Kossentini, Rate-distortion Optimized Layered Coding With
Unequal Error Protection for Robust Internet Video, IEEE Trans. on Circuits
and Systems for Video Technology, Vol. 11, No. 3, pp. 357 -372, Mar. 2001
[28] Yao Wang, L. Karam, G.P. Abousleman, T. Key, B. Razzouk, Error Resilient
Video Coding Techniques, IEEE Signal Processing Magazine, Vol. 17, No. 4, pp.
61 -82, July 2000
[29] I. Moccagatta, S. Soudagar, J. Liang, H. Chen, Error-resilient Coding in JPEG-
2000 and MPEG-4, IEEE Journal on Selected Areas in Communications, Vol. 18,
No. 6, pp. 899 -914, June 2000
[30] R. Talluri, Error-resilient Video Coding in the ISO MPEG-4 Standard, IEEE
Communications Magazine, Vol. 36, No. 6, pp. 112 -119, June 1998
[31] G. Cote, B. Erol, M. Gallant, F. Kossentini, H.263+: Video Coding at Low Bit
Rates, IEEE Trans. on Circuits and Systems for Video Technology, Vol. 8, No. 7,
pp. 849 -866, Nov. 1998
131
[32] A.S. Tosun, W.C. Feng, Efficient Multi- layer Coding and Encryption MPEG
Video Stream, Proc. of IEEE Int. Conf. on Multimedia and Expo, Vol. 1, pp. 119
-122, 2000
[33] S. Dogan, S. Eminsoy, A.H. Sadka, A.M. Kondoz, Personalised Multimedia
Services for Real-time Video over 3G Mobile Networks, Proc. of Third Int.
Conf. on 3G Mobile Communication Technologies, pp. 366 -370, 2002
[34] Yao Wang, Qin-Fan Zhu, Error Control and Concealment for Video
Communication: A Review, Proc. of the IEEE , Vol. 86, No. 5, pp. 974 -997,
May 1998
[35] M. B. Jill, Robert D. Gaglianello, Packet Loss Effects on MPEG Video Sent
Over the Public Internet, ACM Multimedia 98 - Electronic Proc., Sept. 1998
[36] Dapeng Wu, Y.T. Hou, Wenwu Zhu, Hung-Ju Lee, Tihao Chiang, Ya-Qin Zhang,
H.J. Chao, On End-to-end Architecture for Transporting MPEG-4 Video over the
Internet, IEEE Trans. on Circuits and Systems for Video Technology, Vol. 10,
No. 6, pp. 923 -941, Sept. 2000
[37] Chung-Ming Huang, Ming-Yuhe Jang, Handoff Architectures and Protocols for
Transmitting Compressed Multimedia Information in Mobile PCSs, IEEE Trans.
on PCSs, Consumer Electronics, Vol. 43, No. 3, pp. 784 -794, Aug. 1997
[38] Y.Wang, Jrn Ostermann, Ya-Qin Zhang, Video Processing and Communications,
Prentice Hall, 2002
[39] J. Heinanen, F. Baker, W. Weiss, J. Wroclawski, Assured Forwarding PHB
Group, IETF RFC 2597, June 1999.
[40] D. Black, Differentiated Services and Tunnels, IETF RFC 2983, Oct. 2000.
[41] B. Davie , A. Charny, J.C.R. Bennett, K. Benson, J.Y. Le Boudec, W. Courtney,
et al, An Expedited Forwarding PHB (Per-Hop Behavior) , IETF RFC 3246,
Mar. 2002.
[42] N. Brady, MPEG-4 Standardized Methods for the Compression of Arbitrarily
Shaped Video Objects, IEEE Trans. on Circuits and Systems for Video
Technology, Vol. 9, No. 8, pp. 1170 -1189, Dec. 1999
[43] J. Apostolopoulos, T. Wong, Wai-tian Tan, S. Wee, On Multiple Description
Streaming with Content Delivery Networks, Proc. of Twenty-First Annual Joint
132
Conf. of the IEEE Computer and Communications Societies, Vol. 3, pp. 1736 -
1745, 2002
[44] K. Nichols, S. Blake, F. Baker, D. Black, Definition of the Differentiated
Services Field (DS Field) in the IPv4 and IPv6 Headers, IETF RFC 2474, Dec.
1998.
[45] S. Blake, D. Black, M. Carlson, E. Davies, Z. Wang, W. Weiss, An Architecture
for Differentiated Services, IETF RFC 2475, Dec.1998.
[46] J. Apostolopoulos, Error-Resilient Video Compression Through the Use of
Multiple States, Proc. of Conf. on Image Processing, Sept. 2000
[47] J. Apostolopoulos, Reliable Video Communication over Lossy Packet Networks
Using Multiple State Encoding and Path Diversity, Proc. of Visual
Communications and Image Processing, pp. 392-409, Jan. 2001
[48] S. Nanda, K. Balachandran, S. Kumar, Adaptation Techniques in Wireless
Packet Data Services, IEEE Commun. Mag., vol. 38, pp.5464, Jan. 2000
[49] V.K. Goyal, Multiple Description Coding: Compression Meets the Network,
IEEE Signal Processing Magazine, Vol. 18, No. 5, pp. 74 -93, Sept. 2001
[50] IP in the RAN as a transport option in 3
rd
generation mobile systems, Mobile
Wireless Internet Forum MTR-006 v2.0.0, June 2001
[51] General Packet Radio Service (GPRS); GPRS Tunnelling Protocol (GTP) across
the Gn and Gp Interface, 3GPP TS 29.060 V5.4.0, Dec. 2002
[52] John W. Mark and Weihua Zhuang, Wireless Communications and Networking,
Prentice Hall, New Jersey, 2002
[53] Qian Zhang, Wenwu Zhu, Ya-Qin Zhang, Resource Allocation for Multimedia
Streaming over the Internet, IEEE Trans. on Multimedia, Vol. 3, No. 3 , pp. 339
-355, Sept. 2001
[54] Y. Bernet, S. Blake, D. Grossman, A. Smith, An Informal Management Model
for Diffserv Routers, IETF RFC 3290, May 2002
[55] J. Schiller, Mobile Communications, Addison-Wesley, 2000
[56] Combined GSM and Mobile IP Mobility Handling in UMTS IP CN, 3G TR
23.923 V.3.0.0, May 2000
133
[57] Handovers for real-time services from PS domain, 3GPP TR 25.936 V4.0.1,
Dec. 2001
[58] Handover procedures, ETSI TS 100 527 V7.0.0, Aug. 1999
[59] Handover procedures, 3GPP TS 23.009 V5.3.0, Dec. 2002
[60] Network architecture, 3GPP TS 23.002 V5.9.0, Dec. 2002
[61] Quality of Service (QoS) concept and architecture, 3GPP TS 23.107 V5.7.0,
Dec. 2002
[62] J. Wiljakka, Transition to IPv6 in GPRS and WCDMA Mobile Networks, IEEE
Communications Magazine, Vol. 40, No. 4, pp. 134 -140, Apr. 2002
[63] Wei-Ge Chen, Ming-Chieh Lee, -channel Compression in Video Coding,
Proc. of IEEE Int. Conf. on Image Processing, Vol. 1, pp. 500 -503, 1997
[64] J. De Vriendt, P. Laine, C. Lerouge, Xiaofeng Xu, Mobile Network Evolution: A
Revolution on the Move, IEEE Communications Magazine, Vol. 40, No. 4, pp.
104 -111, Apr. 2002
[65] David Leon, RTP Retransmission Framework, IETF Internet draft, Mar. 2002
[66] J. Widmer, R. Denda, M. Mauve, A Survey on TCP-friendly Congestion
Control, IEEE Network, Vol. 15, No. 3, pp. 28 -37, May-June 2001
[67] General Packet Radio Service (GPRS); Service description; Stage 2, 3GPP TS
23.060 V5.4.0, Dec. 2002
[68] S. Das, A. Misra, P. Agrawal, TeleMIP: Telecommunications- Enhanced Mobile
IP Architecture for Fast Intradomain Mobility, IEEE Personal Communications,
Vol. 7, No. 4, pp. 50 -58, Aug. 2000
[69] F.M. Chiussi, D.A. Khotimsky, S. Krishnan, Mobility Management in Third-
Generation All-IP Networks, IEEE Communications Magazine, Vol. 40, No. 9,
pp. 124 -135, Sept. 2002
[70] F.A. Chiussi, D.A. Khotimsky, S. Krishnan, A Network Architecture for MPLS-
based Micro-mobility, Proc. of Wireless Communications and Networking Conf.,
Vol. 2, pp. 549 -555, 2002
[71] T. Ahmed, A. Mehaoua, G. Buridant, Implementing MPEG-4 Video on Demand
over IP Differentiated Services, Proc. of IEEE Global Telecommunications Conf.,
Vol. 4, pp. 2489 -2493, 2001
134
[72] T. Ahmed, G. Buridant, A. Mehaoua, Encapsulation and Marking of MPEG-4
Video over IP Differentiated Services, Proc. of IEEE Symp. on Computers and
Communications, pp. 346 -352, 2001
[73] Jitae Shin, Jong Won Kim, C.C.J. Kuo, Quality-of-service Mapping Mechanism
for Packet Video in Differentiated Services Network, IEEE Trans. on
Multimedia, Vol. 3, No. 2, pp. 219 -231, June 2001
[74] UTRAN Overall Description, 3GPP TS 25.401 V5.5.0, Dec. 2002
[75] T. Yoshimura, T. Ohya, T. Kawahara, M. Etoh, Rate and Robustness Control
with RTP Monitoring Agent for Mobile Multimedia Streaming, Proc. of IEEE
Int. Conf. on Communications, Vol. 4, pp. 2513 -2517, 2002
[76] R.Bennett and H.Zhang, WF
2
Q: Worst-case Fair Weighted Fair Queuing, Proc.
of IEEE INFOCOMM '96, pp. 120-128, March 1996.
[77] OPNET Modeler 9.0 Documentation, OPNET Technologies, Inc., 2002.

You might also like