Sequential Blind Detection using Bayesian Techniques

Receivers for a variety of communication channels must often be designed to operate in the absence of channel information. While a channel estimate may be generated from training data, known data then occupies available spectrum, which is often a scarce resource. We propose a blind maximum likelihood (ML) sequence detection and decoding technique for which the receiver requires no training data or estimate of the channel. The data is encoded via a tree or convolutional code, and the channel is assumed to have a set of unknown parameters that are drawn from a known probability distribution. The likelihood of a sequence of transmitted data is then computed via a Bayesian approach by averaging over the unknown parameters. ML sequence detection is performed via a tree search, allowing the receiver to accumulate information about the channel based on all previously transmitted data. We describe the application of the proposed Bayesian detector and decoder to the intersymbol interference (ISI) channel and show via simulation that it can achieve bit error rates (BER) within 0.25 dB of ML sequence detection for a known channel. We also show that the proposed Bayesian metric is asymptotically equivalent to a maximum a posteriori (MAP) metric that uses an entire sequence of data to generate a channel estimate. In addition, we describe the application of the Bayesian receiver to the binary symmetric channel (BSC) and discuss its relationship to the work presented in [1], which proves that the proposed metric is pairwise universal.