Importance sampling simulation for evaluating the lower-bound BER of the Bayesian DFE

An importance sampling (IS) simulation technique, originally derived by Iltis (1995) for Bayesian equalizers, is extended to evaluate the lower-bound bit error rate of the Bayesian decision feedback equalizer (under the assumption of correct decisions being fed back). Using a geometric translation approach, it is shown that the two subsets of opposite-class channel states are always linearly separable. A design procedure is presented, which chooses appropriate bias vectors for the simulation density to ensure asymptotic efficiency of the IS simulation