The present letter deals with the design of suboptimal data detectors for b
inary transmission over frequency selective channels. A detector consisting
of a prefilter followed by a Viterbi processor with a number of states low
er than that needed for maximum likelihood sequence estimation is considere
d here. Often the parameters of the receiver are optimized according to the
minimum mean square error criterion. Here a stochastic approximation algor
ithm that optimizes the parameters of the receiver according to a measure o
f error probability is introduced. Simulation results show that the propose
d design gives substantial benefits at moderate to high signal to noise rat
io.