Introduction

MLSE is widely used in digital communication systems to estimate the transmitted data sequence observed over noisy ISI channels. For an unknown channel, many algorithms have been developed to estimate the sequence and identify the channel blindly. Some assume the channel is time-invariant and are not applicable for fast time-varying channels (e.g. [1,7]), while others are applicable for channels with fast temporal variation (e.g. Expectation-Maximization (EM) based [4,10]) but are block processing (off-line) and/or sensitive to initialization. Many adaptive MLSEs have also been developed. Some of these algorithms have difficulty in tracking fast time-varying ISI channels due to: tentative decision delay[3]; or error propagation introduced by a decision-feedback equalizer (DFE) embedded in the MLSE structure [5].

Raheli et al. proposed PSP (Per-Survivor Processing) -based MLSE [6] for tracking fast time-varying channels. PSP has been applied for sequence estimation and channel identification successfully. Adaptive algorithms such as LMS and RLS can be combined with PSP to identify the channel and further estimate the data sequence. A suboptimum PSP adaptive MLSE algorithm was proposed in [2], in which the adaptive MLSE consists of a channel estimator and an MLSE implemented by the Viterbi algorithm, and channel estimation is accomplished for each state in the Viterbi algorithm (VA) along the survivor path associated with each state. LMS and RLS can be applied for channel estimation. This adaptive MLSE has a capability of good tracking performance for fast time-varying ISI channels. However, the choice of the parameters of LMS and RLS in channel identification may still affect the performance of this method.

In order to reduce the performance degradation caused by the channel estimation, this paper introduces a channel statistics-aided MLSE method to estimate the transmitted data sequence on-line. This method marginalizes over the unknown channel coefficient distribution and thus obtains the probability of the received signal given the transmitted signal, which is a function of the given transmitted signal. Then the transmitted signal that maximizes the probability is obtained from this function. The Viterbi and PSP Viterbi algorithms are used to obtain optimum and suboptimum results, respectively. For a temporally independent Gaussian channel, an optimum Viterbi algorithm is derived. For a Gauss-Markov channel, we will use PSP and the generalized Viterbi algorithm (GVA), which retains a fixed number of survivors (L) per state instead of retaining one survivor trellis path per state as in the Viterbi algorithm. It is shown that the proposed method approaches that of ML exhaustive search for reasonably small values of L. This new algorithm is suitable for fast time-varying channels. Simulations prove the advantage of this algorithm over PSP LMS and RLS algorithms.

This paper is organized as follows. In section 2, the system model is introduced and the problem is formulated. Our proposed algorithm is presented in section 3 and section 4 for time-independent Gaussian channel and Gauss-Markov channel respectively. Section 4 will also present an adaptive PSP GVA to reduce the computational complexity. The simulation results are shown in section 5. This section also compares our algorithm with PSP LMS/RLS. Finally, we present our conclusions in section 6.