On Packet Loss Prediction for Real-time Packet Audio

The Internet measurement experiments continue to show that bandwidth, delay and packet loss on the Internet vary considerably, posing a challenge for real-time audio applications. Out of these, effect of packet loss is critical to the audio quality. Also, packet-retransmission is not a viable error recovery option for audio. In absence of such a guarantee of reliability, our objective is to predict packet loss, that is, to express the probability of loss of the next packet in terms of available bandwidth, delay variation and trend, and network history, enabling us to take proactive loss recovery actions. Our preliminary results for this work show intuitive and efficient algorithms based on insights from Internet experiments conducted from DePaul University.