FEC Performance in Large File Transfer over Bursty Channels Shuichiro Senda, Hiroyuki Masuyama, Shoji Kasahara and Yutaka Takahashi Graduate School of Informatics, Kyoto University, Kyoto 606-8501, Japan {senda, masuyama}@sys.i.kyoto-u.ac.jp, {shoji, takahashi}@i.kyoto-u.ac.jp Abstract Frequent packet loss and large transmission delay cause the degradation in the quality of service (QoS) for real-time applications over the Internet, such as video streaming and voice over IP. In par- ticular, it is well known that a packet loss process with bursty nature significantly deteriorates the QoS of real-time applications in more extent than that with randomness. In this paper, we consider the impact of the burst loss process on the QoS of real-time applications over the Internet. Focusing on forward error correction (FEC) for QoS guarantee mechanism, we analyze its performance. For a burst-loss model, we consider a Gilbert model which is a two-state discrete-time Markov chain with good and bad states. We derive the block loss probability that original data eventually fails in transmission, and develop its efficient computational algorithm. For numerical examples, we consider the transmission of uncompressed high definition television (HDTV) as a real-time appli- cation, and show how the packet loss process affects the frame-level QoS of uncompressed HDTV. Numerical results show that the block loss probability increases as the packet loss process becomes bursty. They also demonstrate how the redundancy of FEC is effective in recovering lost packets for the frame-level QoS guarantee of uncompressed HDTV. Key words: Forward error correction, Gibert model, discrete-time Markov chain, performance eval- uation, uncompressed HDTV. 1 Introduction The recent advancement of the Internet makes the transmission speed faster and the data size larger than ever before, resulting in the wide spread of multimedia contents such as voices, images, and videos. However, today’s Internet is supported by best effort services at the IP layer, and the quality of service (QoS) for multimedia applications is not strictly guaranteed. Therefore recovery techniques for packet loss are indispensable for the QoS guarantee of multimedia applications. The packet-loss recovery schemes are classified into retransmission-based recovery schemes and coding-based ones. In retransmission-based recovery schemes, lost packets are retransmitted by the sender host. Transmission control protocol (TCP) has a retransmission-based loss recovery mechanism. However, TCP retransmission mechanism is activated by receiving duplicate acknowledgement (ACK) packets or timer time-out, causing a large end-to-end delay. This large delay is not suitable for real-time applications such as streaming media and video conference. A typical coding-based error recovery scheme is forward error correction (FEC). In FEC, redundant data is generated from original data packets, and the sender host transmits both the original data and redundant one. In this paper, we focus on packet-level FEC scheme [5]. When h redundant data packets are generated from k original data, FEC can recover the lost data completely if the number of lost packets is less than or equal to h [2]. FEC is eligible for real-time applications because packet-loss recovery is performed at the receiver host, and hence the end-to-end delay for FEC is smaller than that for retransmission-based loss recovery schemes. If the number of lost packets are larger than that of redundant packets, however, FEC fails to recover the lost packets. On the other hand, adding a large number of redundant packets, increases the offered load. Therefore, it is important to find the optimal redundancy of FEC such that the QoS of real-time applications is highly guaranteed, keeping the offered load as small as possible. P07/1