Maximum-likelihood sequence estimation from subbands

In a paper by Rao and Pearlman (see Proc. IEEE-SP International Symposium on Time-Frequency and Time-Scale Analysis, p.69-73, Victoria, Canada, 1992), it has been shown that subband decomposition results in a reduction of memory in the subbands. Subband Decomposition is a technique in which the source spectrum is bandpass filtered and subsampled to obtain a time-frequency decomposition. Furthermore, the total prediction error power from the subbands is less than the fullband prediction error, for finite orders of prediction. In this paper we apply these results to equalizer design using subbands. It is shown that maximum-Likelihood (ML) sequence estimation in the subbands offers a gain in terms of estimation error. In addition, there is a considerable reduction in the computational complexity. It is also demonstrated that by working in the subband domain, it is possible to avoid the problems associated with the presence of nulls in the channel frequency response