Performance comparison of detection methods derived from maximum-likelihood sequence estimation

The performance of various detection methods which are derived from full maximum-likelihood sequence estimation (MLSE) was investigated for channels introducing linear distortion and additive noise. The Viterbi algorithm may be applied to full MLSE, to maximum-likelihood soft-decision decoding of convolutional codes, or to the combined operations of detection in the presence of intersymbol interference and soft-decision decoding. In the latter case it was found that the free Euclidean distance of the coding and modulation scheme may be reduced, causing a decrease in coding gain. The pure equalisation techniques investigated include a prefilter which truncates the effective memory of a channel, a heuristic reduced-state Viterbi algorithm, and two hybrid receivers which incorporate a decision feedback equaliser into the simplified MLSE process. The performance of these equalisation techniques and one receiver which used a decision feedback equaliser and a soft-decision decoder for a convolutional code recommended by the CCITT, were investigated by computer simulation for a variety of channels. It was found that for all but the most severely distorted channels, significant coding gain could be achieved. For such channels the reduced-state method provided the best performance, while the other techniques were marred by severe error burst effects.