Reduced-complexity soft-ln-soft-out decoding of variable-length codes
Claudio Weidmann · 2003
We present approaches to complexity reduction for a recently introduced soft-in-soft-out (SISO) decoder for variable-length codes (VLC). One method is based on a new trellis construction that reduces complexity by an order of magnitude, but in- curs only a small performance loss in many settings. I. INTRODUCTION Several methods to use soft information to decode variable length codes have been proposed over the last few years. Our starting point is a symbol-by-symbol MAP method, which may be described in a BCJR-type framework (l) or as an instance of Bayesian networks (2). It can exploit the resid- ual redundancy left in the code sequence when symbols from a (first-order) Markov source are coded independently with a fixed VLC. Additionally, it provides bit-level soft outputs, which may be used to iterate between a SISO channel decoder and a SISO source decoder.