Adjusting Forward Error Correction for TCP Friendly Streaming

Adjusting Forward Error Correction for TCP- Friendly Streaming MPEG Huahui Wu, Mark Claypool & Robert Kinicki Computer Science Department Worcester Polytechnic Institute Worcester, MA, 01609 USA {flashine, claypool, rek}@cs. wpi. edu

TCP-Friendly Flows Unresponsive flows get unfair share of network bandwidth and AQM techniques will punish them. Streaming flows need to be TCP-Friendly. A TCP-Friendly flow’s bandwidth is no more than a conformant TCP flow running under comparable network conditions. [PFTK 98]

Forward Error Correction (FEC) Packet loss degrades video quality. FEC adds redundancy to recover from losses. With an MPEG frame made up of K ordinary packets, FEC adds N-K redundant packets to frame. If K or more of the N packets are received, the original frame can be successfully reconstructed. The successful frame transmission probability is

MPEG with FEC Prior Approaches were ad-hoc and did not consider TCP-Friendliness. Our Approach: Adjusting FEC and GOP Forces TCP-Friendly bandwidth constraint. Develops an analytic model to compute playable frame rate for given FEC and GOP pattern. Exhaustively searches for FEC and GOP pattern that optimizes playable frame rate.

Playable Frame Rate Model (1 of 2) GOP rate (Group of Pictures per second): Playable Frame Rate of I, P and B frames

Playable Frame Rate Model (2 of 2) Given values of loss rate and frame size, the total playable frame rate can be estimated with the FEC and GOP pattern. The best playable frame rate can be obtained by searching the FEC and GOP space.

Benefits from Adjusting FEC and/or GOP RTT: 50 ms s: 1 KB p(%): 0. 5, 1, 2, 3, 4, … 10 SI: 25 packets SP: 8 packets SB: 3 packets

Adjusted GOP and FEC pattern Adjusted GOP pattern Y-axis: Number of frames X-axis: Loss percentage Adjusted FEC pattern Y-axis: Number of packets X-axis: Loss percentage

Contribution Summary Contribution Built a model to estimate the playable frame rate with GOP and FEC pattern for MPEG. Used the model to study adjusting FEC and GOP • Adjusting FEC provides large benefit. • Adjusting GOP contributes little improvement. Ongoing Work Model verification with realistic network Adjustable FEC combined with Media Scaling.