Simulated Viterbi decoding using importance sampling

The simulation of Viterbi decoding often requires excessive computer time to estimate low bit error probabilities since large sample sizes are needed to produce the rare error events. In this paper, importance sampling for Viterbi decoding is featured as a means of reducing the number of samples required. This reduction is achieved by modifying or `biasing¿¿ the noise statistics to produce more errors, and subsequently scaling the number of errors to produce the desired bit error probability. If the bias is chosen properly and the effective memory of the decoder is limited, a significant reduction in the sample size is shown to occur.