Video Streaming in Wireless Environments Manoj Kumar C Advisor Prof. Sridhar Iyer Kanwal Rekhi School of Information Technology Indian Institute of Technology, Bombay Mumbai 1
Motivation Refers to real-time transmission of stored video Has stringent bandwidth, delay and loss requirements Approaches New protocols, router scheduling disciplines Adapt output rate of video to available bandwidth Rate Control Schemes - employ feedback ( loss, delay etc.,) Need to adapt rate control schemes to wireless environments What is specific to Video Streaming? Application layer QoS control Affects user-perceived presentation quality 2
HTTP based Streaming Architectures for Video Streaming Standard web servers used to deliver video content Guaranteed-delivery protocols (like HTTP, TCP etc.,) not optimal for continuous media Web Browser 1. HTTP request/response for meta file 2. Meta File Web Server Media Player 3. Audio/Video file requested and sent over HTTP Substantial fluctuations in delivery times of packets due to re-transmission, available bandwidth variations etc., 3
UDP/RTP based Streaming Streaming Server retrieves media components in a synchronous fashion Video sent over UDP using application-layer protocols tailored for video streaming (e.g., RTP) Streaming Server Client/Receiver Video Decoder Media Synchronization Audio Decoder Storage Device Raw Video Video Compression Compressed Video Application layer QoS control Application layer QoS control Raw Audio Audio Compression Compressed Audio Transport Protocols Transport Protocols Internet (Continuous Media Distribution Services) ( Figure from Wu et al, Streaming in the Internet : Approaches and Directions) 4
Transport Protocols (RTP/RTCP) RTP Provides end-end transport functions for supporting real-time applications Functions for media streaming like RTCP sequence numbering time-stamping payload identification Works in conjunction with RTP Designed to provide QoS feedback to participants 5
Application-Layer QoS Control : Rate Control Minimizes network congestion by adjusting the output rate of the video coder to estimated available bandwidth Classified into Source-Based Rate Control Receiver-Based Rate Control Source based rate control schemes may use Probe-Based Approach Example : AIMD, MIMD Algorithms etc., Model-Based Approach Example : TFRC, RAP Algorithms etc., 6
Rate Control in Wireless Environments Characteristics of Wireless Channels Limited Bandwidth High Error Rates Burst Errors Loss based rate control schemes may inaccurately estimate the available bandwidth AIMD based on packet loss fraction during each interval In TCP Friendly Rate Control (TFRC), λ = 1.22 MT U RT T p (1) Assuming MTU and RTT constant, λ 1 p (2) 7
The Problem During bad channel conditions, loss rate reported by receiver may be high Sender may inaccurately assume the network to be congested and decrease the output rate Hence, quality of video delivered to the receiver affected Streaming Server Storage Device Compressed Video Source based Rate Control Rate Shaper RTP/RTCP UDP/TCP Internet RTP Base Station RTCP MH 8
Prime reasons for the problem Solution Scheme(s) Inability of receiver to distinguish between congestion and wireless packet losses Sender estimates state of network using loss rate as principal feedback parameter Two Schemes proposed Report Only Congestion Losses (ROCL) Report Correlation of Loss and Delay (RCLD) 9
Report Only Congestion Losses (ROCL) Receiver enabled to report loss rate only due to congestion Uses heuristic proposed by Saad Biaz et al to discriminate congestion and wireless losses Heuristic based on inter-arrival times of packets at the receiver Sender T T T 3 2 1 3 2 1 3 2 1 BS Receiver Sender Router 2T 3 2 1 3 2 1 BS 3 2 1 Receiver Sender Router T 3 2 1 3 1 BS 3 1 Receiver 2 ( From Nitin Vaidya et al, Discriminating Congestion Losses and Wireless Losses Using Inter arrival times at the Receiver ) 10
Report Correlation of Loss and Delay (RCLD) Based on general patterns of throughput and response time as a function of load Besides loss rate, sender reports correlation between the packet loss and delay curve Throughput Load Round Trip Delay Load During congestion, delay curve increases with loss curve hence will have positive correlation If loss rate high, sender decreases rate only if correlation is positive 11
Simulation Experiments Network Simulator ns from UC Berkeley (version 2.1b8a) Simulation Model MPEG Traffic RTP Agent 3 Mobile Host RTP Agent 1 BW1, D1 BW4, D4 R BW3, D3 BS BW2, D2 BW4, D4 2 4 CBR Cross Traffic Sink 12 Mobile Host
Experiment 1 Network set in an uncongested state so that only wireless losses occur Simulation parameters Experiment 2 BW 1 = BW 2 = 1Mbps, D 1 = D 2 = 2ms BW 3 = 256kbps, D 3 = 10ms BW 4 = 64kbps, D 4 = 1ms Network set in a congested state using cross traffic generated from Traffic/Expo Simulation parameters BW 1 = BW 2 = 128kbps, D 1 = D 2 = 2ms BW 3 = 80kbps, D 3 = 10ms BW 4 = 64kbps, D 4 = 1ms 13
Results Experiment 1 14
120 Congestion Losses Loss Rate Max Rate 100 80 60 40 20 0 0 50 100 150 200 250 300 Report Number Figure 1: Original scheme without proposed modification 15
250 Congestion Losses Loss Rate Max Rate 200 150 100 50 0 0 50 100 150 200 250 300 Report Number Figure 2: Report Only Congestion Losses (ROCL) Scheme 16
180 160 Congestion Losses Loss Rate Max Rate 140 120 100 80 60 40 20 0 0 50 100 150 200 250 300 Report Number Figure 3: Report Correlation of Loss and Delay (RCLD) Scheme 17
Results Experiment 2 18
120 Congestion Losses Loss Rate Max Rate 100 80 60 40 20 0 0 50 100 150 200 250 300 Report Number Figure 4: Original scheme without proposed modification 19
120 Congestion Losses Loss Rate Max Rate 100 80 60 40 20 0 0 50 100 150 200 250 300 Report Number Figure 5: Report Only Congestion Losses (ROCL) Scheme 20
120 Congestion Losses Loss Rate Max Rate 100 80 60 40 20 0 0 50 100 150 200 250 300 Report Number Figure 6: Report Correlation of Loss and Delay (RCLD) Scheme 21
Related Work Elan Amir et al proposed Application Level Video Gateway Employs split-connection approach Transcodes video stream from server to lower bandwidth Mobile Host Base Station Wired Network Video Gateway Problems Increase in end-end delay due to transcoding Transcoding difficult when packets are encrypted 22
Conclusion & Future Work ROCL and RCLD try to decrease the output rate only in response to congestion Simulation experiments using ROCL and RCLD show significant increase in the output rate of video during bad channel conditions Future Work Investigate appropriate functions to replace loss event rate p in model-based schemes Maintain network state at the sender to aid in making adaptation decisions 23
References [1] Elan Amir, Steve McCanne, and Hui Zhang. An application level video gateway. In Proc. ACM Multimedia 95, San Francisco, CA, 1995. [2] S. Biaz and N. Vaidya. Discriminating congestion losses from wireless losses using inter-arrival times at the receiver. IEEE Symposium ASSET 99, Richardson, TX, USA, 1999. [3] Jean-Chrysostome Bolot and Thierry Turletti. A rate control mechanism for packet video in the internet. In INFOCOM (3), pages 1216 1223, 1994. [4] Sally Floyd, Mark Handley, Jitendra Padhye, and Jorg Widmer. Equation-based congestion control for unicast applications. In SIGCOMM 2000, pages 43 56, Stockholm, Sweden, Auguest 2000. [5] Dapeng Wu and et al. Streaming video over the internet: Approaches and directions. IEEE Transactions on Circuits and Systems for Video Technology, 11:282, 2001. 24