Bayesian generation of dependent erasures for frequency-hop communications and fading channels

The use of block coding and errors-and-erasures decoding can enhance performance in frequency-hop (FH) communication systems, provided that a good scheme is employed to determine which symbols to erase. The problem of making erasure decisions from collections of receiver outputs is investigated in this paper. Methods to determine which received symbols to erase are derived from Bayesian decision theory. Decision rules are developed for a system with M-ary orthogonal signaling, noncoherent demodulation, and frequency-selective fading. One result is a class of Bayesian schemes in which erasure decisions are made independently from symbol to symbol. Within this class is a rule that uses signal amplitude estimates for improved performance. A second result is a Bayesian technique in which erasure decisions are mutually dependent and are made collectively for each codeword. These techniques are analyzed and compared with the performance of receivers that use erasure techniques that require multiple applications of bounded distance decoding. The performance of the Bayesian techniques for dependent erasures are also compared with the performance of receivers that do not permit erasures. It is found that each of the Bayesian techniques offers substantial performance gains over errors-only decoding, and the dependent erasure scheme provides the best performance among the techniques of lower complexity.