Joint data and channel estimation using blind trellis search techniques

A novel method of maximum likelihood sequence estimation is proposed for data that are transmitted over unknown linear channels. This procedure does not require a startup sequence for estimating the channel impulse response. Rather, the data and the channel are simultaneously estimated. It is implemented without any loss of optimality by a trellis search algorithm which searches for the best data sequence from among a number of hypothesized trellises which are constructed from the observed sequence. The number of states in each trellis and the number of trellises grow exponentially with the channel memory. A suboptimal trellis search algorithm is proposed whose complexity at best is slightly higher than that of the adaptive Viterbi algorithm operating with a known channel response. A simplified channel estimation algorithm when the number of data alphabets is greater than 2 is also proposed. Fast convergence of the algorithm in estimating the channel is demonstrated for binary pulse amplitude modulation over a variety of channels. Convergence over a wide range of SNR occurs within 100 symbols. The channel estimation algorithm for multi-level signals converges within 500-1000 symbols. We finally present an application of this algorithm to the problem of sequence estimation in the presence of rapidly time-varying intersymbol interference. The algorithm provides reliable zero-delay decisions which results in a better channel tracking algorithm when compared to previously proposed schemes