When an analog source such as an audio or video signal is transmitted over a noisy channel, typically the source is first quantized and compressed (source coded) and then encoded to make it robust to the noisy channel (channel coded). For a class of channels called ergodic channels, it is optimal to separately perform source and channel coding. This principle has served as an engineering guideline to design several modern communication systems. However non-ergodic channels are encountered in several practical communication systems, for example in broadcasting a source to many users. For this class of channels, the separation based approach is not necessarily optimal. The focus of this work in understanding and providing guidelines for performing source and channel coding for such channels.
Three separate but related problems in coding for non-ergodic channels are considered. The first is that of minimizing a measure of end-to-end distortion such as expected mean squared error in communicating a source over a non-ergodic noisy channel. The focus is on the design of analog and hybrid analog to digital joint source-channel coding techniques which can outperform conventional separation based digital approaches. Then the design of practical universal channel codes for non-ergodic channels with memory is considered and it is shown that coded decision feedback signal processing is a promising technique to solve these problems. Finally, the design of universal codes for distributed compression where the underlying correlation model is not perfectly known is considered.
Apart from standard educational activities, a key aspect of the broader impact is in creating avenues for our students to interact with researchers from other universities.
