Turbo Source Coding: A Noise-Robust Approach to Data Compression

All traditional data compression techniques, such as Huffman coding, the Lempel-Ziv algorithm, run-length limited coding, Tunstall coding and arithmetic coding are highly susceptible to residual channel errors and noise. We have recently proposed the use of parallel concatenated codes and iterative decoding for fixed-length to fixed-length source coding, i.e., turbo coding for data compression purposes. The work presented here extends these results and also considers the case when decompression must be done from compressed data corrupted by additive white Gaussian noise. Both constituent encoders are based on finite state machines and are designed to maximize memory and output symbol usage. Each encoder is described using two matrices: an input state transition matrix whose (i,j) entry corresponds to the k-symbol input sequence when the encoder makes a transition from state i to state j, and an output state transition matrix whose (i,j) entry corresponds to the n-symbol output sequence for the described transition. These matrices are designed using Latin squares with