Patterns of Loss

With simple FEC you could lose every other packet and still not be missing content, but if pairs of packets are lost then you lose content.

To understand temporal patterns of speech, various models have been developed, see for example [Kim and Kleijn 2003].

Slide Notes

[Kim and Kleijn 2003] Moo Young Kim and W.B. Kleijn, ‘Rate-distortion comparisons between FEC and MDC based on Gilbert channel model’, presented at The 11th IEEE International Conference on Networks, 2003. ICON2003, Sydney, Australia, 2003, pp. 495–500 [Online]. DOI: 10.1109/ICON.2003.1266239

Transcript

[slide88] Now, Kim and Kleijn, and Kim was a student in the course the first time I taught it, together with Professor Kleijn, they did something which some of you might have heard of. They built the first open source voice CODEC, and it's actually the voice CODEC that was initially used in Skype. But they looked at temporal patterns. What happens, what's the probability that I lose not one isolated packet, but I lose both of these packets, or some particular pattern of them. Can I do coding to protect against that? And the answer is yes, I can. And so with clever coding, but an increase in delay, I can produce a system that will tolerate even burst errors. Which is very cool. Right? I pay a delay penalty, but I can now tolerate more packet loss.