CROSS-LAYER OPTIMIZATION FOR STREAMING MPEG4

VIDEO OVER HSDPA NETWORKS

I. Abdeljaouad, T. Rachidi and H. Bouzekri

Al Akhawayn University in Ifrane, Morocco

Keywords: Cross layer Optimization, HSDPA, MPEG4, HARQ.

Abstract: A novel cross layer optimization technique for efficient streaming of MPEG4 VIDEO over a High Speed

Downlink Packet Access (HSDPA) network is proposed in this paper. The proposed technique uses the

types of frames produced by the MPEG encoder to optimize the performance of the Hybrid Automatic

Repeat reQuest (H-ARQ) protocol at the MAC layer. Our aim is to reduce the total power at the NodeB, and

to increase the overall system throughput, while maintaining satisfactory user-perceived Quality of Service

(QoS). The proposed technique consists in applying ARQ retransmission for MPEG4 I-frames (the most

critical frame of an MPEG4 stream) upon the reception of a negative acknowledgment (NACK) message

from the receiver (UE). Packet combining is then performed with the aid of the available I-frames at the

receiver side. Different packet combining strategies have been investigated to assess the performance of the

proposed cross-layer technique. We show that compared to the blind HARQ Chase Combining scheme

applied indiscriminately to all MPEG4 frames, our scheme allows for saving up to 11% of the power at the

NodeB, and up to 10% of the system bandwidth, while ensuring satisfactory video quality to users.

1 INTRODUCTION

HSDPA (also known as 3.5G and as WCDMA

Release 5) is a new release of UMTS networks. Its

new downlink transport channel HS-DSCH (High

Speed Downlink Shared Channel) provides greater

capacity –up to several Mbps-, as well as increases

the wireless system performance by supporting new

features. Among these features are: fast link

adaptation, Adaptive Modulation and Coding (AMC

- changing modulation and coding format according

to channel conditions), fast scheduling, and Hybrid

Automatic Repeat request (HARQ). The new

transport channel (HS-DSCH) uses available radio

frequency resources efficiently by sharing multiple

access codes, transmission power, and use of

infrastructure hardware between users.

AMC provides the possibility to match

modulation-coding scheme to the channel conditions

for each user. The power of the retransmitted signal

is kept constant over a frame interval while the

modulation and coding format changes to match the

current received signal quality.

Contrary to UMTS Rel’99, in HSDPA the

scheduler has been moved from the RNC (Radio

Network Controller) to the NodeB. Scheduling is

done based on information about the channel quality,

terminal capability, QoS class and power/code

availability. It is fast because it is close to the air

interface and a shorter frame length is used. The

most famous and widely used scheduling algorithms

are Round Robin (RR), Maximum Carrier-to

Interference ratio (MAX-CI) and the Proportional

Fairness (PF).

Moreover, HARQ handles retransmissions

requested by UEs (User Equipments) due to errors in

the radio packets. These requests are processed in

the current WCDMA networks by the RNC while in

HSDPA, they are processed at the NodeB to provide

the fastest response possible. HARQ has two

schemes: Chase Combining (CC) and Incremental

Redundancy (IR). CC keeps the erroneous packet,

and requests that the exact same packet be

retransmitted. Upon receipt of this latter, it uses soft-

combining to combine the erroneous and

retransmitted packets to increase the possibility of

successful decoding. IR, on the other hand,

retransmits the same packet but differently coded.

The receiver selects correctly transmitted bits from

the original transmission and the retransmission to

minimize the need for further repeat requests when

multiple errors occur in transmitted signals.

Abdeljaouad I., Rachidi T. and Bouzekri H. (2008).

CROSS-LAYER OPTIMIZATION FOR STREAMING MPEG4 VIDEO OVER HSDPA NETWORKS.

In Proceedings of the International Conference on Wireless Information Networks and Systems, pages 74-77

DOI: 10.5220/0002022800740077

 SciTePress

CC is simple to implement, while IR is more

powerful but complex, eventually adding extra delay

in the decoding.

Yet, despite the high data rates and the

previously cited improvements offered by HSDPA

over UMTS Rel’99, its shared medium represents

still a challenge for the provisioning of QoS

(guaranteed bandwidth, delay and jitter) for delay

and/or error-sensitive applications such as MPEG4

video streaming applications. And although, the

radio protocol stack at the NodeB is designed to

operate under worst condition scenarios, it remains

generic, and does not factor in specific application

requirements (such as the differentiation in the

transmission/retransmission schemes to be used for

various application data/frames), yielding ineffective

use of spectrum.

For achieving optimal decision, and therefore

yielding efficient transmission subsystem, the

different layers of the end-2-end protocol stack need

to cooperate and exchange. Sharing knowledge/data

types among the different protocol layers (which is

the main idea behind Cross Layer Optimization -

CLO) helps achieve a higher adaptability to the

changing network conditions although this is

violating strict layering design rules.

2 PREVIOUS WORKS

Recognising the importance of CLO when streaming

MPEG4 video over wireless networks (and best

effort networks in general), many researchers have

looked into how the availability of application layer

information across the layers up until the MAC layer

can help achieve better performance. For instance,

the main idea explored in (Ahmed et. al., 2003) is to

add a cognitive layer able to change transport

parameters, bit rates and QoS mechanisms based on

the network conditions. Therein the proposed

architecture takes into consideration the

characteristics of MPEG4 and IP Diffserv to propose

techniques for media content analysis and network

control mechanisms for adaptive video streaming

over IP networks. In (Zheng, 2003) Zheng studies

the effect of the scheduler (MAX and PF) as well as

the error detection/protection techniques (HARQ

fitted mapping and 1% FER based mapping) on QoS

parameters for the case of streaming MPEG4 over

HSDPA. He also compares the performance of UDP

and UDP-Lite for streaming MPEG4 over UMTS-

like systems. In (Chen et. al., 2003), three techniques

are presented to tackle the changing conditions of

the wireless medium for multimedia delivery. These

are: swift-OFDM, low-latency packet-awareness

coder and adaptive noise filtering. Last, (Yufeng et.

al., 2002) proposes a set of end-to-end application

layer techniques for adaptive video streaming over

wireless networks. These techniques are:

Application layer packetization scheme, Class based

unequal error protection and finally a Priority based

ARQ scheme.

In wireless networks, video streaming

applications suffer the most from delay and jitter that

are introduced mainly by retransmitting erroneous or

lost packets. As for errors, such applications use

error concealment techniques to compensate for any

erroneous video frame. Also, we know that HSDPA

is known for delivering better QoS in terms of delay

and jitter values, as well as for its strong

retransmission strategy, namely HARQ, thanks to

the new added. Still, using HARQ will result in more

delay and jitter but better quality.

In this paper, we propose a new scheme based on

cross layer optimization for streaming MPEG4 video

over HSDPA. Our adaptive scheme uses interaction

between both the link and application layers (the link

layer being the one that knows about the changing

network conditions and the application layer being

the one that knows about the type of video frame) to

take a retransmission strategy based on the type of

video frame being retransmitted (I, P or B). To our

best knowledge, there has not been any published

research that combines HARQ retransmission

strategy with the importance of the video frame

being retransmitted over HSDPA.

The remaining of the paper is organized as

follows: section 3 presents the technique and the

underlying assumptions. Section 4 describes the

simulation setup, while the results are presented in

section 5.

3 PROPOSED TECHNIQUE

In this paper, we make information normally

available to the application layer (which is the type

of video frame) accessible to the MAC layer so that

this latter makes a retransmit decision based on the

type of the video frame. When the MAC-HS entity

receives an erroneous frame, it checks its type before

requesting a retransmission. If the frame is of type I,

it requests the retransmission. However, if the frame

is of any other type (P or B), it just discards it and

does not request retransmission. I-frames are the

ones that carry much information and that P (and B)

frames depend on the previous (and following) I-

frames for successful decoding. We also know that I-

frames are the ones that achieve the least

compression ratio while other types of frames

CROSS-LAYER OPTIMIZATION FOR STREAMING MPEG4 VIDEO OVER HSDPA NETWORKS

achieve the best compression ratio. Thus, if an I-

frame is lost or erroneously received and if we

choose to discard it, all successive P and successive

and previous B frames depending on this I-frame

will fail decoding and thus we would have lost

bandwidth and power by transmitting them because

they will be discarded at the receiver anyway. This is

why our scheme favors I-frames over the other types

of frames.

Standard video streaming applications use UDP,

RTP and RTCP as transport protocols. RTP runs on

top of UDP, packetizes and provides in-order

delivery of video frames. RTCP, when used,

operates as closed loop control mechanism for

informing the video source of the received video

quality. During the simulations, we assume no

interaction between the video client and server.

Thus, RTCP is not modeled. However, those

functions needed such as packetization, packet

sequence numbering and in-order delivery are

supported by the different tools in Evalvid (Klaue et.

al., 2007). For example, packetization is

implemented by the Video Sender at Evalvid.

Without loss of pertinence, we use MAX-CI as a

scheduler, because it serves users with good channel

quality increasing system throughput and providing

better QoS. We also use CC as our HARQ scheme to

minimize delays.

Also, since the primary goal of the simulation is

to investigate the impact of ARQ/HARQ schemes on

the quality of the MPEG4 video, we assume no

packet losses, errors or congestion occurring in

either the Internet or the UMTS core network. This

is a fair assumption when compared to air interface

generated errors. Moreover, we assume that ACKs

and NACKs coming from UEs to the BS do not

undergo any losses or errors. The delay introduced

by the Internet and UMTS core network is kept

constant and low throughout the simulation time.

Each link capacity was chosen so that the radio

channel in the connection bottleneck. Moreover, the

functionality of GGSN and SGSN was abstracted out

and modeled as traditional ns2 nodes since in

general, they are wired nodes and mimic the

behavior of IP routers. Last, we assume no header

compression at the PDCP (Packet Data Convergence

Protocol) layer.

4 EXPERIMENTAL SETTINGS

The network consists of one or several MPEG4

streaming servers that stream videos to one or more

user equipments. The packets sent by the servers

flow through the Core Network and the UTRAN to

arrive at the Node B. This latter uses the MAC-HS

entity to send the packets to the intended user

equipment. For each case of number of user

equipments/streaming servers (1, 5 and 10), we

change the ARQ scheme and collect and analyze the

data. We use three schemes: no ARQ, blind HARQ

with CC, and our adaptive scheme. The streaming

server streams a 10mn MPEG4 encoded video to the

UE. The core network and the UTRAN links have a

high data rate and a very low delay so that this part

of the network does not cause any delay or loss. We

would like to concentrate on the link between the

NodeB and the UEs. The simulated application is

H320 videophone (Halsall, 1996) with a 48 MB

play-out buffer at the receiver.

The simulations were performed on a Rayleigh

fading environment that conforms to the ITU-T

recommendations (Recommendation ITU-R M.1225,

2000).

5 RESULTS

Figure 1 shows the overall frame loss percentage and

the bandwidth delay product for the three techniques

as a function of increasing load/users. One can see

that CC gives the best frame loss percentage while

no ARQ gives the worst. Our adaptive scheme

comes in between but is closer in performance to no

ARQ because of the high number of P and B frames

that are discarded. As for the delay bandwidth

product, we clearly see that our adaptive scheme

incurs little deterioration compared to no ARQ

(which is the one that would give the best results

since no retransmissions take place). We also see

that the gap between CC and the two other schemes

gets bigger as we load the network with more users.

This means that our adaptive scheme lowers the

buffering requirements compared to CC that needs

larger buffers due to delays introduced by

retransmissions especially with high network load.

BW Delay Product & Frame Loss

1 user 5 users 10 users

Network Loa d

Frame Loss (%)

0.5

1.5

2.5

Bandwidth Delay Product (MB)

No ARQ CC Adaptive Scheme No ARQ CC Adaptive Scheme

Frame Loss Bandwidth Delay Product

Figure 1: Performance of various HARQ techniques.

WINSYS 2008 - International Conference on Wireless Information Networks and Systems

Cumulative Jitter

-1.4

-1.2

-1

-0.8

-0.6

-0.4

-0.2

network Load

Min. Cum ulative Jitter (s)

1 user 5 users 10 users

Max. Cumulative Jitter (s)

No ARQ CC Adaptive Scheme No ARQ CC Adaptive Scheme

Min. Cumulative Jitter Max. Cumulative Jitter

Figure 2: Performance of various HARQ techniques.

Figure 2 shows the cumulative jitter values for the

three techniques and for different number of users on

the network. It is clear from this graph that our

adaptive scheme performs much better than CC in a

way that it has a maximum cumulative jitter near the

no ARQ scheme. We also notice that overloading the

network does not increase the gap between our

adaptive scheme and no ARQ as opposed to CC

where the gap grows bigger. As for the minimum

cumulative jitter, we notice that our adaptive scheme

outperforms CC and gives results that are close to no

ARQ. This means less variations and hence better

quality of service.

We have also conducted MOS (Mean Opinion

Score) for the assessment of user-perceived quality

of received media under the three schemes. MOS

provides a numerical indication of the perceived

quality of received media and is expressed as a

single number in the range 1 to 5, 1 being the lowest

perceived quality, and 5 being the highest perceived

quality. As shown in table 1 which summarizes

MOS values for our simulations, the proposed

adaptive scheme brings a clear improvement of the

perceived quality with little additional use of

network resources.

Table 1: MOS analysis (5 independent viewers).

Network Load

ARQ technique

1 user 5 users 10 users

No ARQ 3.2 2.5 0.8

Adaptive Scheme 4.2 3.2 1.5

CC 4.9 4.3 2.6

6 CONCLUSIONS

In this paper, cross-layer optimization has been used

to improve the QoS of streaming MPEG4 video over

the HSDPA network. We made information on the

type of video frame (normally known to the

application layer) available to the MAC-HS layer so

that this latter retransmit erroneous I frames only in

order to minimize delays and jitter. This scheme is

simple to implement since it requires only breaking

the layered architecture and have the MAC-HS layer

access the application payload and get the type of

frame. Overall, the proposed adaptive scheme is able

to provide better QoS and gain of bandwidth at the

expense of a slight degradation in video quality.

Finally, bandwidth gain simply means that the

system is able to support more users; the bandwidth

gain results show that we can gain up to 10% using

our adaptive scheme, meaning that this 10% can be

used by other applications and to support more users.

REFERENCES

T. Ahmed et all. “Adaptive Packet Video Streaming Over

IP Networks: A cross Layer

Approach”, IEEE journal on selected areas in

communications, vol. 23, n. 2, feb 2005.

H. Zheng, “Optimizing Wireless Multimedia Transmission

Through Cross Layer Design”, ICME 2003, 185-187.

J. Chen et all. “Joint Cross Layer Design for Wireless QoS

Video Delivery”, ICME 2003 197-200.

S. Yufeng et all., “Cross Layer Techniques for Adaptive

Video Streaming Over Wireless Networks”, IEEE

C2002 277-280.

J. Klaue et. all, “EvalVid - A Framework for Video

Transmission and Quality Evaluation”, http://

www.tkn.tu-berlin.de/publications/papers/evalvid.pdf,

last accessed: Sept. 10, 2007

F. Halsall, “Data Communications, Computer Networks

and Open Systems”, 4

edition, Addison-Wesley 1996.

“Recommendation ITU-R M.1225: Guidelines for

Evaluation of Radio Transmission Technologies for

IMT-2000”, available via: http://www.itu.int

3gpp specification 3GPP TS 25.308, “3rd Generation

Partnership Project; Technical Specification Group

Radio Access Network; High Speed Downlink Packet

Access (HSDPA); Overall description; Stage 2”

(Release 5)

3gpp specification, 3GPP TS 125 213, “Universal Mobile

Telecommunications System (UMTS); Spreading and

modulation (FDD)”

CROSS-LAYER OPTIMIZATION FOR STREAMING MPEG4 VIDEO OVER HSDPA NETWORKS