Reconfigurable trellis decoding of linear block codes

A class of methods for soft-decision decoding of linear block codes, referred to as reconfigurable trellis (RT) decoding, is presented. In RT decoding a reduced trellis (or tree) search is facilitated by carrying out the search on a reconfigured trellis (or tree) that corresponds to an equivalent code. The equivalent code is formed by reordering the received symbols according to their reliabilities. Consequently, the trellis reconfiguration is determined 'on-the-fly', but only a small portion of the trellis needs to be constructed, as guided by the reduced search. The search efficiency improves for channels where the soft-decisions provide a good indication of which symbols are in error. For example, using the M algorithm on an erasure channel, only a single survivor (i.e. M=1) is sufficient to attain maximum-likelihood decoding of maximum-distance codes. For more typical channels, we present simulation results and a detailed assessment of the number of metric and binary-vector operations for the M algorithm.