Interative decoding based on dominant error events

Abstract

An iterative decoding system for intersymbol interference (ISI) channels has a module for extracting bit reliabilities from a partial response (PR) channel, an iterative decoder, and a module for updating the bit reliabilities. A transmitter parses a data sequence into blocks that are encoded to generate a sequence of codewords. By encoding, a correlation among the bits of each codeword output to the PR channel is created. A maximum likelihood sequence detector (MLSD) in the receiver produces estimates of transmitted bits from samples of the output from the PR channel. The MLSD detector has a priori knowledge of typical error events that can occur during transmission through the channel. Along with the bit estimates, at each time instant the MLSD detector generates set of error event likelihoods. These error event likelihoods are then converted into bit reliabilities that, together with estimates for the transmitted bits, are used to recalculate the bit reliabilities using the knowledge of the relation between bits within a codeword. The iterative decoder uses this soft input information (bit reliabilities and bit estimates) for each iteration of decoding to improve i) the estimate of the bit reliabilities, ii) the decisions of what bit has been transmitted, and iii) calculations for the error event likelihoods for the next iteration. These error event likelihoods are then converted into bit reliabilities that, together with estimates for the transmitted bits, are used by the iterative decoder to recalculate the bit reliabilities using the knowledge of correlation among bits within the codeword. The error event likelihoods may be updated using the updated bit reliabilities, and the updated error event likelihoods are then converted to new bit reliabilities for the next iteration. In an iterative manner, increasing those bit reliabilities that tend to show increasing confidence for corresponding decoded bits (i.e., corresponding Viterbi decisions) between iterations, while decreasing those reliabilities that tend to show decreasing confidence for corresponding decoded bits, tends to drive the iterative decoding scheme to fewer iterations while maintaining a predetermined probability of error.