Maximum likelihood decoding with reduced complexity

We design maximum-likelihood (ML) decoding, whose complexity has exponential order q/sup nR(1-R)/(1+R)/ for virtually all linear q-ary codes of length n and rate R used over any memoryless channel. This exponent reduces up to 3 times the known complexity exponent q/sup n/ /sup min(R,/ /sup 1-R)/ of ML decoding.