Adaptive Channel Memory Truncation for Maximum Likelihood Sequence Estimation

Forney 1 has recently proposed a receiver structure for a communication system operating over a known time-dispersive channel with little loss in performance due to intersymbol interference by using maximum likelihood sequence estimation, or the Viterbi algorithm (VA).2 This has resulted in much attention being given to practical methods of applying his results. Magee and Proakis 3 proposed the use of the VA directly in conjunction with a channel estimator. This approach can result in a receiver too complex for practical use because the complexity of the VA depends exponentially on the duration of the channel impulse response. In particular, if the impulse response of the channel has an effective duration of r seconds and if an L-level PAM system transmits 1/7 1 data symbols per second, the number of operations per received symbol is proportional to LTIT. For channels such as voiceband telephone channels, the bandwidth of which is used efficiently, typical values of T/T may be between about 20 and 200, making direct application of the VA infeasible. Thus, it seems clear that effective practical application of the VA or of related techniques involves a compromise between optimum performance and receiver complexity. The complexity-limiting approach we take here is to use a linear prefilter at the receiver to "condition" the overall sampled impulse response seen by the VA so t h a t it is significantly different from zero over only a small number of samples, and any remaining intersymbol interference is considered to be noise.