Quality Scaling For TCP-Friendly Video Streaming ...

WSEAS TRANSACTIONS on COMMUNICATIONS Issue 5, Volume 5, May 2006 ISSN

1109-2742

http://www.wseas.org

Radiation Characteristics of Prime Focus Reflector Antennas System Based on Geometrical Optics Approximation A. Boualleg, N. Merabtine, M. Benslama

641

Equivalent Circuit, Transmission Line Model and Smith Chart Approach for Multilayer Periodic Spatial Filter Design P. T. Teo, X. F. Luo, C. K. Lee

647

Statistical Sampling Based Equalization Algorithms for Fading Channels and Multiuser Detection A. Olah, J. Levendovszky

656

New OFDM Channel Estimation Scheme with Two-Dimensional Hadamard Transform: Algorithm and Performance Qihong Ge, Liuguo Yin, Huazhong Yang

665

On Modeling Datagram Congestion Control Protocol and Random Early Detection using Fluid-Flow Approximation Hiroyuki Hisamatsu, Hiroyuki Ohsaki, Masayuki Murata

672

Quality Scaling for TCP-Friendly Video Streaming over AWGN Wireless Channel Ghaida A. Al-Suhail, Naoki Wakamiya

686

Convolutional Coding Data Transmission System Using Real-Valued Self-Orthogonal Finite-Length Sequences Jiong Le, Yoshihiro Tanada

694

Modeling and Simulation of the Downlink Transmission Power for WCDMA Networks in a Multiservice Environment Chie Dou, Liang-Luen Huang

702

A Stable Differentiated Service Enabled MAC and Routing Protocol for in Mobile Ad Hoc Networks Chenn-Jung Huang, Yi-Ta Chuang, Wei Kuang Lai, Sheng-Yu Hsiao

710

Scheduling the Transmission for Two-Hop Broadcast in Static Wireless Ad hoc Networks Chenguang Xu, Yinlong Xu, Minhong Lin, Guanjun Ma

718

A Dispatching Technique to Solve the Overloading Conditions of Web Cache Servers K. Y. Wong, K. H. Yeung

725

Analytical Study on Web Caching Systems using Closed Queuing Network Modeling K. Y. Wong, K. H. Yeung

732

Modeling and Performance Evaluation of Long-Term Evolution for 3G System Xing Zhang, Renshui Zhu, Shiming Liu, Wenbo Wang

738

Efficient Multiple OVSF code assignment strategy in UMTS Min-Xiou Chen

745

Study on Monitoring Approach of 3D Scanning and Measuring Machine Based on Internet/Intranet Wang Wen, Lu Keqing, Chen Zichen

753

Reliable Simulation of Communication Networks Tapio Frantti

758

A New Design "NT&SD Transducer" Used for Earthquake Detecting System Through Network Communication S. -S. Lin, Huay Chang, Cheng-Chien Kuo, Yong Su

766

Building/Extracting/Mapping/Visualizing an Individualized Learning Contents Based on Ontology Kyeungshun Kim, Cheol Min Kim, Seong Baeg Kim

774

Secure Group Communication in Multi-Agent Systems Zhu LieHuang, Cao YuanDa, Liao LeJian, Tan Yu'an, Muhanmad Hanif Durad, Wang DaZhen

781

Profiling Scheme for Mobile Learning Service Seong Baeg Kim, Kyoung Mi Yang, BongKyu Lee, Doo-Yeong Yang, Sun Young Kim, Cheol Min Kim

788

A Template-based Customized Learning System for Instructors Cheol Min Kim, Hye Sun Kim, Seong Baeg Kim

796

Simulation Based Analysis of a Reliable Datagram Protocol M. Lulling, J. Vaughan

803

An Image Steganography Systems Based on BPCS and IWT Silvia Torres-Maya, Mariko Nakano-Miyatake, Hector Perez-Meana

814

Analysis of Delay Distribution of Data call in CDMA Systems Supporting Voice and Delay-tolerant Data Calls Insoo Koo, Jeongrok Yang, Kiseon Kim

821

The Study on the Detection Methods of DSSS/QPSK Signal Based on a Slice of the Fourth-Order Moment Zhao Zhijin, Jie Tingting, Wu Jia, Shang Junna, Kong Xianzheng

826

An Adjustable Hot-video Broadcasting Scheme Hung-Chang Yang, Hsiang-Fu Yu, Li-Ming Tseng, Yi-Ming Chen

832

Efficient Request Scheduling Algorithms for All-Optical WDM Networks with Tunable ADMs Yinlong Xu, Wei Zhou

840

Low Complexity Bit-Flipping Based Decoding for Low-density Parity-check Codes Jin Sha, Minglun Gao, Zhongjin Zhang, Li Li, Zhongfeng Wang

848

A Trust Management System in Mobile Enterprise Networking Zheng Yan, Peng Zhang

854

To Slim Down Internet Response Time using SHA1 & RIPEMD-160 Methods in CDN S. Manikandan, A. Chitra

862

Efficient Cluster-based Routing Protocol for Wireless Sensor Network Yoon-Su Jeong, Yoon-Cheol Hwang, Sang-Ho Lee

868

Design of Monitoring System and Conference System on NTP VoIP Platform Whai-En Chen, Chai-Hien Gan, Yi-Bing Lin

877

Postgate: QoS-aware Bandwidth Management for Last-mile ADSL Broadband Services

884

Ming-I Hsieh, Eric Hsiao-Kuang Wu A Traffic Separation Mechanism (TSm) allowing the coexistence of CSMA and real-time traffic in Wireless 802.11e Networks Ricardo Moraes, Francisco Vasques, Paulo Portugal, Jose Alberto Fonseca

890

A New Optimal Adaptive Voice Smoother Based on Lagrangian Multiplier Method for VoIP Service Shyh-Fang Huang, Eric Hsiao-Kuang Wu, Pao-Chi Chang

898

Broadband TEM Horn Array for FOPEN Radar Sh. Norouzi, Ch. Ghobadi, J. Nourinia

906

A New External Sorting Algorithm with Selecting the Record List Location Iraj Hassanzadeh, M. H. S. Afsari, S. Hassanzadeh

909

Error Concealment Algorithms for MPEG-2 Video Yuk Ying Chung, Siqi Li, Xiaoming Chen, Changseok Bae

914

Power Efficient MAC Protocol Through Intelligent Transmission Mechanism Selection for IEEE 802.11 Based Infrastructure Basic Service Set Ye Ming Hua, Lau Chiew Tong, Benjamin Premkumar

922

Genetic Algorithm Assisted Channel Estimation for Multi-user Communication Systems Arash Ghasemmehdi, Vahid Reza Asghari, Mehrdad Ardebilipour

930

Investigating the Opinions of University Students on E-learning Hsieh-Hua Yang, Hung-Jen Yang

935

Amplification chain used on UHF transmitter Lahcène hadj Abderrahmane, M. Benyettou, M. Sweeting, J.R. Coocksley, Peter Garner

940

Magnetorquer Control for Orbital Manoeuvre of Low Earth Orbit Microsatellite A.M. Si Mohammed, M. Benyettou, S. Chouraqui, A. Boudjemai, Y. Hashida

944

The Quantum Cryptography - Solution To the Problem due to the Principle of Uncertainty of Heisenberg Aris Skander, Michel Planat, Malek Benslama

948

Quality Scaling For TCP-Friendly Video Streaming Over AWGN Wireless Channel GHAIDA A. AL-SUHAIL NAOKI WAKAMIYA Computer Engineering Department Graduate School of Information College of Engineering Science and Technology Basrah University Osaka University Basrah, IRAQ Osaka, JAPAN Email: [email protected], [email protected], [email protected] Abstract: - Although there are emerging needs of multimedia communications such as video streaming over wireless networks, the end-to-end quality of service for video streaming applications can severely be affected by frequent packet losses over wireless links. In this paper, video streaming over an AWGN (Additive White Gaussian Noise) wireless channel behind wired links is investigated. For this purpose, we propose VFR-TCP algorithm (Variable Frame Rate based on TCP-Friendly Rate Control) to evaluate the predicted frame rate of MPEG-4 video streaming. Quality of Service (QoS) is also evaluated by the predicted quantizer scale Q for the case that the network throughput is assumed to be equal to the required bandwidth. We analyzed influences of a wireless channel on the perceived video quality in terms of the frame rate and the quantizer scale, by varying the channel error rate or the channel SNR where AGWN and a BPSK scheme dominate. We also evaluated effects of the number of connections on the achievable throughput and the video quality. Key-Words: - TCP-Friendly, Video streaming, Wireless video, QoS, Quality Scaling.

1 Introduction Recently, wireless multimedia communications over networks has grown over the last decade involving real-time video applications, such as video conferencing, video phoney, and on-demand video streaming. The continuing growth in wireless communications such as 3G and 4G wireless systems has attracted considerable applications as well as researches to transmit video over wireless channels [1]. In practice, wireless video communications face several challenges such as high bit error rates, bandwidth variations and limitations as well limited power for multimedia services, and processing capability constraints on the handhold devices. Among these, wireless channels are afflicted by an additive white Gaussian noise AWGA, time-varying fading and interference conditions which may lead to packet corruption. To provide an acceptable end-to-end quality of service (QoS) for video applications (i.e. good video quality) at high loss rates over wireless links, there are several approaches have been pursued: adaptive rate control, passive error recovery, Forward-ErrorCorrection (FEC) for error control rates and recently adaptive modulation to improve throughput [1-6]. Usually, these techniques are applied separately and independently. However, other studies [1-3] have combined adaptive modulation and joint source

channel coding (JSCC) over fading wireless channels indicating significant performance advantages for worst channel conditions. In contrast, [7] explains quality scaling for streaming MPEG based adjusting FEC in packet-level application layer over VBR Internet traffic. The use of media scaling in multimedia servers would reduce the streaming data rate to match the channel capacity constraint (i.e. available channel bandwidth). Practically, the multimedia server adjusts the quantization level before transmission to reduce the streaming bit rate. With FEC, a multimedia application can increase the quantization level even higher to save capacity for the FEC overhead. On the other hand, the current TCP-Friendly flow is the dominate transport protocol in the wired and wireless Internet [1,2,6]. In this work, TCP-Friendly is, however, used over wireless link for several reasonable advantages such as introducing high reliable transmission due to being connectionoriented protocol and avoiding the network congestion collapses. Strictly speaking, one way to achieve the required QoS in wireless link, TCPfriendly rate control (TFRC) is suitable to the video applications where the video emission rate can be adjusted to the appropriate level according to the estimated network condition [6]. Therefore adaptive rate control can mitigate effects of channel impairments but becomes inefficient without some

form of FEC. Previously, we derived and developed a model of playable frame rate for streaming MPEG [8] that did not applied for wireless link. This paper extends the analytical model to characterize the wireless state condition at high bit error rates for random errors occurred during transmission. The end-to-end packet loss can consequently impact streaming video quality. Such packet loss needs an algorithm to search all possible combinations of opening TCP connection and scaling level required to obtain a required peak allocated bandwidth, which is constrained according to the predicted video quality factor as well frame rate both are based on TCP-Friendly. In this paper, the framework is focused on the bit error rate (BER) in term of channel SNR of the physical-layer for an additive white Gaussian noise (AWGN) wireless channel combined with wire link. The wireless channel is assumed in bad condition whereas the BPSK scheme is applied at high bit error rate. The emphasis is being placed on streaming video based TFRC (TCP-Friendly Rate Control) flow involving the network parameters such as end-to-end packet loss and the expected round trip time. In Section 2, the background describes briefly MPEG video standard and the quality scaling techniques. Section 3 is problem formulation which explains video streaming over a typical wireless channel model and the proposed VFR-TCP model (Variable -Frame Rate TCPFriendly) for packet loss is employed to predict the effective playable frame rate. Section 4 describes the problem solution for optimal frame rate and the strategy to reach the optimal throughput. The rest of paper involves the simulation results in Section 5 for multiple-opened TCP connections and the full utilization of the wireless channel with/without exceeding the upper bounded limit. Also, the QoS (Quality of Service) in term of SNR scalability is accounted for the predicted quantizer scale Q if the network throughput is assumed to be equal the required bandwidth. Finally, Section 6 introduces the conclusion.

2 Background 2.1 MPEG Basically, MPEG (Motion Picture Expert Group) is a popular standard for compression video streams. Fig. 1 illustrates a typical MPEG GOP (Group of Picture) structure. Each GOP structure consists of three types of frames: I-, P- and B-frames. I (Intra) frames are coded as still images and serve as reference for previous and future predicted frames.

Every P (Predicted) frame depends on the preceding I or P-frame. Finally, every B-frame (Bidirectionalcoded) depends on the surrounding reference frames (the closet two I and P or P and P frames that surround it). The loss of one P frame can render some of other P and B frames undecodable, and the loss of one I frame can result in the whole GOP being undecodable. This implies that I frames are more important than P frames, and P frame are more important than B frames [3-7]

Fig. 1

A typical MPEG Group of Pictures

2.2 Quality Scaling In MPEG, the video signal is quantized by dividing the DCT coefficient by an integer called the quantization value, and then rounding the result to the nearest integer. Thus when high quantization value is used, each MPEG frame is encoded with lower precision and in consequence transmitted with fewer bits. Hence, the quality scaling technique will reduce the bit rate of the video streaming. However, the user’s preference is related to video quality of service (QoS) parameters in terms of spatial resolution scalability, PSNR resolution scalability and timely scalability of MPEG videos. The spatial resolution of perceived video is described by the number of pixels in each frame and the user may chose 640x480, 320x240 or 160x120 pixels. The PSNR resolution scalability is introduced by adjusting the degree of quantization during the video coding process. The quantizer scale can be adjusted in MPEG coding algorithm whereas a specific quantizer scale is performed against each block of 16x16 pixels. For large quantizer scale, the quality of decoded block becomes poor. It means this scale leads to degraded PSNR values. On the other hand, a reasonable end-to-end video quality can be achieved in allocating the bandwidth to VBR traffic for MPEG coded video data. The bandwidth can equal to actual peak rate of video stream. As the timely resolution is degraded by dropping one or more frames of GOP, the resulting peak rate decrease is inversely proportional to the number of dropped frames in the peak rate

based bandwidth. Fig. 2 illustrates an example of GOP(1,2) IBBPBB whereas empty time slots of B frame are dropping and hence the transmission rate of I and P frames can be reduced. The resulting GOP becomes I BP B or I P only. The original frame rate, in consequence, is reduced by two or one thirds; i.e. from 30fps to 20fps or 10fps, respectively. By displaying frame repeatedly, the empty frame time rate

rate

p is being the steady end-to-end packet loss

allocated bandwidth

time

probability (loss rate), S is the packet size (in byte), t RTT is the round-trip time, t RTO is the TCP retransmit time out value.

time

Frame dropping

Fig. 2 Frame dropping and allocated bandwidth

should be filled and then the expected video quality is not highly degraded by dropping B frames. Moreover, there is a common tendency in the relationship between QoS parameters and the required bandwidth independently of the video content. The relation between the required bandwidth BWQ and the quantizer scale Q is expressed as in [5] for BW10 ( i.e. Q  10 ). Therefore, it is found that the required bandwidth BW(R, Q, F ) in [bps] to guarantee the preferred video quality can be estimated in terms of spatial resolution ( R [pixels]), PSNR resolution ( Q ) and the timely resolution ( F [fps]) as R

) 1 log4 ( 9.707 4.314 F BWR,Q,F  ( ) 640x480 (0.151  2 ) BWBase 3.1 Q Q 30

streaming applications [1,6,7] but effectively for video streaming over wireless the packets corrupted due to high bit errors at physical layer can not be distinguished from that caused due to buffer overflow. Thus the streaming video rate of TCPFriendly traffic is denoted by T as an equation based rate control on the path S (1) T 2p 3p t RTT  t RTO (3 ) p(1  32 p 2 ) 3 8

(2)

BWbase indicates to the peak based bit rate in [bps].

3 Problem Formulation Transmission of real time video over wireless networks is challenging because of the low bandwidth constraints involved and the negative impact of channel errors on the perceptual quality of video at the decoder. To achieve good video quality (Quality of service) guarantee at the decoder it requires a robust transmission scheme over wireless link. Therefore the network infrastructure or the transport protocol may be either changed or modified. Using an ARQ error control scheme for the wireless channels (e.g. mobiles), the effective channel bandwidth will become variable due to the retransmission involved when the channel condition is in poor state. Also, there are basically advantages for using TCP-Friendly in both wired and wireless

3.1 Wireless Channel Model A typical wireless channel model for video streaming can be considered as shown in Fig. 3. The video server s in the wired network is streaming video to a receiver r in the wireless link. The wireless link is assumed to have available bandwidth Bw and packet loss rate p w caused by wireless channel error and assuming no congestion at node 2, pc  0 , due to overflow buffer in wireless link.

Fig. 3

Typical wired/wireless video streaming model.

Here, the wireless channel is assumed as underutilized if the streaming throughput is less than the maximum possible throughput over the wireless link,i.e. T (1  p )  B w (1  p w ) . This condition indicates definitely the under utilization of the bandwidth, whereas the streaming throughput is less than the maximum possible throughput over wireless link. Hence, a following brief scenario can be applied as there are no cross-traffics at either node 1 or node 2: 1. The wireless link is assumed to be bottleneck by meaning no congestion at node 1. 2. Achieving pc  0 (no congestion) at node 2 and in consequence t RTT  t RTT min (i.e. minimum RTT) if and only if T  B w at node 1. 3.

Bw and pw are constants ( pw is assumed to be random and stationary causing packet loss in wireless link) [6].

4. The Backward route in model is assumed to be congestion-free but not-error free (i.e. with BER). In general, a given scenario can carry out two aims: the streaming rate should not cause any network instability, i.e. congestion collapse. Additionally, the optimal performance should result in highest possible throughput and lowest packet loss rate. In consequence, without modifying in the network infrastructure and protocol, whereas TFRC flow of (1) is employed as a sending rate for video streaming with overall packet loss p and with non delayed Acknowledgement.Thus p can be combined packet loss of two independently pw and pc .

p  pw  (1  pw ) pc

(3)

pw is a lower-bound for p if p c  0 and in consequence the upper-bound for the network throughput becomes S T  T (4) t RTT min

2 pw 3 pw 2  t RTO (3 ) pw (1  32 pw ) 3 8

b

Tb is being the upper-bound of sending rate that can be achieves as a required bandwidth (i.e. maximum network throughput). Hence, for under-utilized channel the condition of (4) becomes, Tb  Bw for only one-TFRC connection. For full-utilized channel, a strategy of one application involves a number of connections that can be opened to reach the optimal performance (i.e. optimal throughput). This can be achieved if the total throughput is not exceeding Bw (1- p w ). Given Bw , p w and packet size S for each connection then nc  nopt , optimal number of connections must satisfy Bw nopt.S   tRTTmin 2Pw  tRTO(3 3Pw )Pw(132Pw2 ) 1 3 3

(5)

where nopt  B w / Tb is being to represent fullutilization channel state, and nopt .Tb should not exceed the available bandwidth of wireless channel

Bw . In wireless link, however, BPSK scheme is considered in the case of poor state channel condition. Here a channel is assumed at high bit error rate which is due to additive WG noise ignoring fading effect. Thus firstly it is required to specify the range of channel SNR per bit allowed and in consequence the corruption-based packet loss

can take place towards achieving the effective network throughput based on TFRC flow for video streaming. The performance basically depends on using a typical GOP structure for MPEG-4 video streaming in order to perform the compatibility requirement between the low bit rates of MPEG-4 and bandwidth constraints for high bit errors over AWGN wireless channel. Thus for an ideal assumption any bit corruption which occurs in any certain packet over AWGN wireless channel may cause that packet is dropped (i.e. lost) completely. Hence the packet loss performance of BPSK can be achieved by setting the channel bit error rate equal to packet loss p w . As a result, corruption-based loss leads an altering in the sending rate described in TFRC scenario in Section 2.1. The assumption depends on BER performance of uncoded BPSK scheme [3] Pe  Q (  )  Q (

2Eb ) No

(6)

Here Eb is the bit energy, N o is the noise power and   2 Eb N o represents total channel SNR for BPSK channel and the Gaussian cumulative distribution function is being Q (.) .

3.2 Analytical Packet-Loss Model This section provides the details of the analytical packet loss model. It can be used for both wired and wireless networks if the packet loss rates are considered random and stationary. The model is called VFR-TCP (Variable Frame Rate TCPFriendly) model for VBR traffic [8]. The model can introduce the corrupted packet loss over wireless link in order to evaluate QoS parameter of the playable frame rate for VBR traffic. The TFRC in Section 2.1 is considered to control the network throughput according to the corrupted-based loss due to the bit errors over wireless channel, which is considered under estimated bandwidth. Thus a frame-dropping mechanism for MPEG VBR service is proposed to compensate the varying TCP-Friendly bandwidth. By this mechanism, the GOP pattern will be adjusted to improve the transmission rate (network throughput) when there is a packet loss. For simplicity, the parameters related to TCPfriendly video streaming for MPEG-4 are identified by GOP structure and frame type dependencies during decoding GOP at the receiver. The proposed model considers F as the frame rate and the quality of a frame (i.e. certain PSNR) falls below a certain threshold, then the frame is considered dropped (i.e. "lost"). The quality of the delivered video depends on: (i) PSNR (i.e. Peak SNR) and (ii) the frame rate

at which frames whose PSNR is above some PSNRthreshold is played at the receiver. As packet loss p is appeared, the picture quality threshold increases and eventually the number of acceptable frames in sequences (frame rate) will decrease. However, here the analytical model depend on the Bernoulli packet loss rate [9] and frame rate degrades roughly as for some constants α and c (7) F   (1  p ) C To understand how packet loss affects frame rate, a relationship between p and the observed frame rate F can be expressed as

F  f o (1   )

Table 1: Frame targets states for full-motion video

(8)

where  is the "frame drop rate", the fraction of frames dropped, and f o is the frame rate of the original bit stream in [fps] (e.g. 30 fps). If f o is assumed not constant by using VBR traffic such as TCP-friendly traffic, then f o is replaced with f r . Since GOP consists of number of frames and GOP rate represents number of GOPs per second, the total frame rate can be expressed as, (9) F  G.SGOPsize (1   ) where f r  G .S GOPsize , denotes as variable frame rate depending on GOP structure ignoring the frame drop rate (i.e.   0). Whilst, SGOPsize1 NP  NB denotes the size of GOP and G is defined as GOP rate [GOP per second], where, G  (T / S ) ( S I  N P S P  N B S B ) . The parameters for a streaming MPEG flow are defines as: N P , the number of P frames in a GOP; NB , the number of B frames in an interval of I and P frames, N B  (1  N P ) N BP ; and SI , SP and SB are the sizes of I, P and B frames (in packets), respectively. In this model, the frame rate control depends on the variability of TCP-Friendly over wireless link due to high bit error characteristic if and only if the throughput is not exceeding the wireless bandwidth (available bandwidth). In addition, the framedropping probabilities introduce the packet loss in a prior estimation for video transmission. Therefore, the model can be used to adjust GOP pattern to obtain the expected playable frame rate. The frame drop rate  is defined as a sum of conditional probabilities

   P( f i ).P( F \ f i )

"useless" because it falls below a certain quality threshold. f i is the event that the corresponding frame of type i . The a priori probability P( f i ) can be determined directly from the fractions of bit stream data of each frame type [7-9]. The conditional probabilities for each frame type f i can be expressed under the assumption that if even one packet within a frame is lost (or the effect of one packet loss from a reference frame are seen), that the frame is rendered useless.

(10)

i

where i runs over the three possible frame types (I, P and B), and F represents the event that a frame is

However, Section 5 introduces the simulation results for VFR-TCP model basing on the frame rate state. Table 1 describes the frame rate states which is affected the full-motion of the video application [7] and consequently it is clear that the required video quality is eventually based on the expected frame rate over network

4 Problem Solution To investigate the effects of adjusted GOP, two cases are considered in VFR-TCP model according to the frame dropping rate  and assuming the upper bound frame rate is Fo  30fps . Thus the following steps are suggested to find the optimal playable frame rate for QoS requirements: 1. Given a specified channel SNR per bit is based to determine the corresponding pw , if pc  0 (no congestion at wired link node). 2. Determine the range of channel SNR/bit (in dB) in term of BER, where p w  BER . 3. TFRC throughput is determined if and only if the under utilized bandwidth condition is verified by (4). 4. Compute the playable frame rate using adjusted GOP patterns. 5. For all the selected combinations of results obtained, it is compared to find maximum rates required. 6. The frame drop rate  is also estimated for VFR-TCP model. 7. To control only a quantizer scale Q, for example, if the base stream BW10 is known, which is coded with Q=10, R=640x480 and

40

35

30

25

20

15

10

5

0 -6

-4

-2

0 Channel SNR/bit (in dB)

2

4

6

(a)

1000

900

32-connecti ons 16-connecti ons

800

700 8-connecti ons

600

500 4-connecti ons

400 2-connecti ons

300

1-connection

200

100

0

0

5

10 15 Wireless channel error rate %

20

25

(b) 2 No extra connection One extra connection T wo extra connections Five extra connections T en extra connections

BPSK, No Fading Bw=1Mbps No congestion

1.8

3P 2P  4 (3 ) P (1  32 P 2 ) 8 3

5 Simulation Results

Bw=1Mbps (wireless channel bandwidth) No congestion effect

45

1.6

1.4 Round trip time (sec)

t RTT 

50

Max. Number of Opened TCP Connections

A strategy to reach the optimal throughput performance emphasizes the condition Tb  Bw which may not be satisfied to allow reasonable performance for one application. Actually, one application may use multiple simultaneous connections and therefore the total throughput of the application is expected to increase with the number of connections until it reaches the hard limit of Bw (1- p w ). Nevertheless, for fixed p w , the throughput increases with the number of connections up to a point, after which there is a saturation effect. Opening new connections can be achieved with out exceeding the wireless bandwidth constraint, but any extra connection or too many connections maximizes the throughput and results either higher packet loss rate p c or higher t RTT than t RTT min . This can be performed if the product condition nc .Tb  Bw is carried out. Thus (11) S / Tb

Table 2: Network and MPEG video settings over 1xRTT CDMA.

Throughput (Kbps)

F=30, the required bandwidth BWQ of the stream becomes as a peak rate of the system, say 144kbps or 128kbps. Then during a video session, the system estimates a TCP-friendly rate through observations on the network conditions, and consequently, Q can be derived numerically to obtain TCP-friendly rate with respect to BW10 . Hence, the video session becomes TCP-friendly [8].

1.2

1

0.8

0.6

0.4

0.2 RTT min=168msec

0

0

5

10 Wireless channel error rate %

15

20

(c) 0.25 BPSK , No Fading AWGN channel

10 additinal new connections opened 0.2 End-to-End Packet loss End-to-End Packet loss rate

To verify the performance of TFRC described in above model, simulations are based on a network and MPEG pattern settings for a typical 1xRTT CDMA wireless data network in Table 2 [6]. Meantime, the predicted playable frame rate of VFR-TCP model is also based on the frame targets described in Table 1. The results of maximum possible number of TFRC connections as a function of wireless channel SNR per bit, the throughput for 2, 4, 8, 16 and 32 connections and the expected round-trip time for different TFRC connections such as 1, 2, 5 and 10 as a function of wireless channel SNR per bit are shown in Fig.4. It is noticed that for pw  0.043, the the expected channel SNR is 1.68dB can also give the optimal number of connection around 4 or 5 for maximum throughput as in [6].

0.15

wireless error rate 0.1

Packet loss due to Congestion 0.05

0

0

5

10 15 Wireless channel error rate %

20

25

(d) Fig. 4 TFRC performance over wireless link (a) Maximum number of TCP connections (b) Throughput (c) Round-trip time and (d) End-to-End Packet loss rate for 10 extra connections.

1000

w/o congestion with congestion

900

BPSK, NO fading Bw=1Mbps

800

Only one additional new connections opened

Congestion effect

600

500

400

300

700

600

500

400

300

200

200

100

100

0

0

5

10 15 End-to-End Packet loss rate %

5 Additional new TCP connections opened

800

700

20

w/o congestion with congestion

BPSK, NO fading Bw=1Mbps

900

Throuhgput of Each Connection (kbps)

Throuhgput of Each Connection (kbps)

1000

0

25

Congestion effect

0

5

10 15 End-to-End Packet loss rate %

20

25

(a) (b) Fig. 5 The throughput of each connection versus end-to-end packet loss 18

In addition, throughput should not exceed the hard limit of bandwidth Bw  1Mbps and consequently for

connections. Therefore, the end-to-end packet loss rate depicted in Fig. 4 as a function of packet loss rate, p w , explains the predicted network congestion due to extra TCP connections, for example around 10 connections, as well as bit error rate. It is found that the packet congestion due to extra connections decreases as p w increases and that is related to round-trip time feedback which should decrease but not be less than 168msec for high bit error rate values. The congestion effect due to extra TFRC connections is also depicted in Fig.5. It is found that as a number of extra TFRC connections (i.e. exceeding nc,max) the range of end-to-end packet loss increases, for example, the range of end-to-end packet loss becomes starting from 5% when 5 extra TFRC connections are opened. Thus the range of throughput (i.e. performance) for each connection is limited with the corresponding end-to-end packet loss. However, Fig. 6 explains the searching schemes of the predicted playable frame rate versus wireless error rate for 5 and 10 extra opened connections. Adjusting the number of extra connection opened from the video server (i.e. the end-host) leads an improvement in the performance but the degradation in the expected playable frame rate is clearly accompanied when more than one connection is simultaneously opened. As a result,

Frame rate (Frame per sec)

decreases as bit error rate p w increases until it be closely to t RTT  168 ms for single connection. In other words, such cases illustrate too many connections exceeding nc ,max by 1, 2, 5 and 10 extra

14

12

10

8 VFR-TCP w/o frame dropping

6

4 VFR-TCP with frame dropping 2

0

0

5

10 15 Wireless channel error rate %

20

25

(a) 18 Only one T CP 10 Extra connections opened Only one T CP 10 Extra connections opened

BPSK , AWGN channel No Fading Bw=1Mbps GOP(1,2) IBBPBB

16

14

Frame rate (Frame per sec)

fixed p w the expected t RTT increases as additional TFRC connections increases, meanwhile, such t RTT

Only one T CP 5 extra connections opened Only one T CP connection 5 extra connections opened

BPSK , AWGN channel No Fading Bw=1Mbps GOP(1,2) IBBPBB

16

12

10

8

6

VFR-TCP w/o frame dropping

4 VFR-TCP with frame dropping 2

0

0

5

10 15 Wireless channel error rate %

20

25

(b) Fig. 6 Predicted playable frame rate versus wireless error rate for adjusting the opened extra connection.

the predicted performance decreases as the number of extra connections is increased due to the end-toend packet loss rate. In general, it is evident that the performance of VFR-TCP model can be controlled not only by adjusting the number of extra connections opened but also GOP pattern quality compared with frame rates in [7]. Table (3) explains the effect of packet loss 2%, 3% and 4% using VFRTCP model over wireless link for typical GOP(4,2) and upper bound clip rate 30fps. It is evident that when the original sending rate is fixed at 30fps and

independent on VBR traffic during transmission then the expected frame rate is being much higher. That’s due to ignoring the latency of TCP-Friendly parameters such as round-trip and timeout. Moreover, the video quality of TCP-Friendly can also be evaluated in term of Q factor in Fig. 7 using the inverse equation of bandwidth required in [5] whereas the predictable frame rate for GOP(1,2) pattern is examined as an example. Table 3: The predicted frame rate for VFR-TCP model for a typical GOP(4,2) compared with [7].

to the packet loss rate. In other words, the required peak allocation bandwidth increases as packet loss decreases and consequently in order to obtain the average rate transmission for good and reasonable quality it is preferred to maintain the peak allocated bandwidth close to average rate between 50kbps 99kbps. For example, with Q=3.97 then packet loss is 1.5% for both GOP(1,2) and GOP(4,2) video patterns, also with Q=5.75 the packet loss is 2.7% for peak allocated bandwidth 19kbps to 50kbps. Hence, it can be concluded that the Q factor is independent on video GOP pattern but has significant effect on the peak allocated bandwidth to provide the effective bandwidth for transmission using TCP-Friendly traffic. Table 4: Peak Allocated Bandwidth required with respect to peak bandwidth 144kbps for R(640x480), 30fps, Q=10. (a) GOP(1,2) (b) GOP(4,2)

20

10 VFR-TCP w ith frame dropping BPSK scheme, No fading Bw =1Mbps RTT=168msec GOP(1,2)

Total Playable Frame Rate (fps)

16

VFR-TCP w /o frame dropping

8 6

14

4

12

2

10

0

8

-2

6

-4

4

-6

2

-8

0

0

5

10

15

20 25 30 Quantizer Scale Q

35

40

45

Channel SNR/bit (dB)

18

-10 50

Fig. 7 Video Quality of one TCP-Friendly connection versus the predicted frame rate and channel wireless error rate with respect to peak bandwidth required 144kbps for R (640x480), 30fps, Q=10.

The results are based on video characteristic stream of 144kbps peak bandwidth required, R (640x480), 30fps and Q=10. It is noticed that the required Q for the corresponding playable frame rate is increased for low values of frame rates and in contrast decreases for high channel SNR per bit, hence Q is allowed to be less than 10 to achieve a reasonable quality. However, Table 4 summaries the peak allocated bandwidth required and the corresponding quality scale (quantizer scale) based on the variability of TCP-Friendly throughput on the network for GOP(1,2) and GOP(4,2). In consequence, the required peak allocated bandwidth depends on the predicted quality scale with respect

6 Conclusion This paper has dealt with a variable frame rate model based TCP-Friendly rate control for under and full utilized bandwidth over wireless channel. The model investigates the effects of high packet loss due to high bit errors over a simple AWGN wireless link. The proposed work estimates QoS for the video streaming in terms of frame rate as well as the quality factor (Quantizer factor Q). However, the effects of adjusting GOP pattern on the predicted frame rate based TFRC video stream are pursued in Table 4 which explains the relationships among wireless channel state (in term of channel SNR per bit in dB), packet loss, quality scale and the required allocated bandwidth. Simulation results show that the proposed model introduces a good performance for the playable frame rate and peak allocated bandwidth over wireless link.

As a result, an improvement in the performance can be achieved in terms of QoS when system settings parameters of MPEG, network conditions state and the source encoding parameters (i.e. quantizer scale and PSNR of video image) are adjusted very well. In particular, the adjusting of quality scaling can consequently has a significant role to achieve a reasonable video quality (with low distortion in the decoded video) with respect to the wireless channel state. Further work can also be proposed to improve the performance using FEC schemes for more robust transmission based TFRC flow.

ACKNOWLEDGMENT G. A. AL-Suhail thanks RSISE, ANU, Canberra, in Australia for their free offer and help in using their Laboratory as well as to Prof. R.S. Fyath, AlNahrain University, Baghdad, Iraq for his advices. References: [1] F. Zahi, “ Optimal Cross-layer resource allocation for real-time video transmission over packet lossy networks", PhD thesis, June 2004. [2] T.C. Wang, H. Fang and L.G. Chen, “ LowDelay and Error-Robust Wireless Video Transmission for Video Communications ”, IEEE Tran. On Circuits and Systems for Video Technology, Vol. 12, No.12, p.p.1049-1058, Dec. 2002. [3] Y. Pei and J. W. Modestino, "Muli-Layered Video Transmission over Wireless Channels using an Adaptive Modulation and Coding Scheme”, Proc. IEEE ICIP, 2001, pp.1009-1012. [4] N. C. Ericsson, “Adaptive Modulation and Scheduling for Fading Channels”, GlobeCom ’99 Rio de Janeiro, 1999. [5] K. Fukuda, N. Wakamiya, M. Murata and H. Miyahara: "QoS Mapping between user's preference and bandwidth control for video transport", IFIP, 1997. [6] M. Chen and A. Zakhor, "Rate Control for Streaming Video over Wireless", Proceeding of INFOCOM 2004, Chain, March 2004. [7] H. Wu, M. Claypool, R. Kinicki, “ Adjusting Forward Error Correction with Quality scaling for Streaming MPEG”, ACM, 2005. [8] G. A. AL-Suhail and R.S. Fyath, “ Analytical Model For Packet Loss Recovery with TCPFriendly Bandwidth Based MPEG”, accepted to 6th JIEEEC 14-16 March, 2006. [9] N. Feamster, “ Adaptive delivery of Real-time streaming Video”, Master thesis, May 2001.