We consider the design and adaptation of a linear equalizer with a finite n
umber of coefficients in the contest of a classical linear intersymbol-inte
rference channel with Gaussian noise and a memoryless decision device. If t
he number of equalizer coefficients is sufficient, the popular minimum mean
-squared-error (MMSE) linear equalizer closely approximates the optimal lin
ear equalizer that directly minimizes bit-error rate (BER). However, when t
he number of equalizer coefficients is insufficient to approximate the chan
nel inverse, the minimum-BER equalizer ran outperform the MMSE equalizer by
as much as 16 dB in certain cases. We propose a simple stochastic adaptive
algorithm for realizing the minimum-BER equalizer. Compared to the least-m
ean-square algorithm, the proposed algorithm can provide a substantial redu
ction in BER with no increase in complexity.