Generalized Viterbi algorithms for error detection with convolutional codes

Abstract

Presented are two generalized Viterbi algorithms (GVAs) for the decoding of convolutional codes. They are respectively, a parallel algorithm that simultaneously identifies the L best estimates of the transmitted sequence, and a serial algorithm that identifies the lth best estimate using the knowledge about the previously found l-1 estimates. These algorithms are applied to combined speech and channel coding systems, concatenated codes, trellis-coded modulation, partial response (continuous-phase modulation), and hybrid ARQ (automatic repeat request) schemes. As an example, for a concatenated code more than 2 dB is gained by the use of the GVA with L=3 over the Viterbi algorithm for block error rates less than 10/sup -2/. The channel is a Rayleigh fading channel.<>