Decoding of non-systematic turbo codes for stationary memoryless and piecewise stationary memoryless sequences

Summary form only given. In this paper, we consider channel decoding of sequences generated by i.i.d. sources and piecewise stationary memoryless sources (PSMSs). We design a turbo decoder that utilizes compression techniques for PSMSs, and use codes that have the quick-look-in (QLI) property, where two parity sequences generated by one of the constituent recursive convolutional codes sum modulo-2 to the systematic message sequence. During the first decoding iterations, source statistics are obtained between iterations by segmentation of the PSMS data, estimation of the i.i.d. statistics in each segment, and then enhancement of these estimates by reversing the channel transition probabilities. Then, during later iterations, the extrinsic information can be used for the same algorithm without need to reverse the channel. To improve code performance, an interleaver, that diversifies the channel code from the PSMS structure, is included in the encoder between the source output and the channel code.