Error probabilities for Viterbi decoders with correlated noise

An upper bound can be calculated for the decoder error probability of convolutional coded data with Viterbi decoding by summing the numbers of error paths weighted by the path preference probabilities. This method can be readily applied for uncorrelated errors, but, for correlated noise, particularly with soft decision, the calculation of the path preference probabilities is more difficult. The method of importance sampling has been applied to this calculation using the noise-covariance matrix as input. Some results obtained by applying this technique are presented and compared with simulated decoding.