On the performance of the Viterbi decoder with trained and semi-blind channel estimators

Maximum-likelihood sequence estimation is often used to recover digital signals transmitted over finite memory convolutive channels when an estimate of the channel is available. We study the impact of channel estimation errors on the quality of sequence detection. The general case of single input multiple output (SIMO) channels is considered. An asymptotic upper bound for the symbol error rate is presented which allows to treat channel estimation errors as equivalent losses in signal-to-noise ratio (SNR). This relationship is studied and numerically validated for the standard least squares channel estimate and for the semi-blind estimator which makes use of the empirical subspace of the observed data