Distributed Source Channel Coding of Correlated Sources

In their 1973 paper Slepian and Wolf described the achievable rate region for two correlated sources transmitting to a common destination over noiseless channels. The important case of distributed source coding, i.e where the encoding is done separately and the decoding is done jointly, is considered. In this thesis, we extend this scenario to the case where the channels are noisy. Distributed coding schemes are proposed for the transmission of both discrete valued and continuous valued sources over noisy channels. The schemes rely on iterative decoding at the receiver. First the case of two correlated continuous valued sources transmitting to a common destination is discussed. The encoding scheme consists of a variable length encoder at one source and a convolutional encoder at the other source. An iterative decoder is derived for this encoding scheme which is similar to the decoder of a parallel concatenated code. Two different correlation models are considered: a binary symmetric channel correlation model and a Gaussian correlation model. The proposed scheme is compared to an alternate scheme employing convolutional codes at each source in terms of symbol error rate and mean square error. In both cases, the proposed scheme is seen to give better performance for sources exhibiting low correlation. Next, the transmission of a binary source over a noisy channel when a correlated source is available at the receiver as side information is considered. A coding scheme which is a concatenation of an outer irregular repeat code with an inner convolutional code is proposed. A novel EXIT chart based optimization criterion is proposed to design the code parameters. The performance of codes designed using the proposed approach is seen to be close to theoretical bounds.