Reduced-search blind trellis decoders for frequency-selective Rayleigh fading channels

Breadth-first reduced-search trellis decoding algorithms are applied to blind maximum likelihood sequence estimation (BMLSE) for frequency-selective Rayleigh fading channels. The channel is modeled as a first-order autoregressive moving average (ARMA) process with known second-order statistics. The optimal BMLSE detector is shown to operate recursively on an expanding tree structure with a Kalman filter for each branch. The T algorithm and M algorithm are used to form practical trellis decoders for suboptimal BMLSE. Error performance and algorithm complexity are determined for each decoder using computer simulations. It is shown that these suboptimal BMLSE technique can offer an order-of-magnitude improvement in error rate over conventional decision-directed approaches