A semi-blind EMVA for maximum likelihood equalization of GMSK signal in ISI fading channels

We examine the maximum likelihood (ML) equalization of Gaussian minimum shift keyed (GMSK) signals in GSM systems. The method we employ is based on the expectation maximization Viterbi algorithm. (EMVA). The EMVA is applicable to transmission schemes that can be modeled as a finite state machine (FSM), whose noisy output sequence is thus a hidden Markov chain. The GMSK signal transmitted via an inter-symbol interference (ISI) channel is just one particular instance of a hidden Markov model. Our channel identification procedure makes full use of the known training bits available in each GSM frame and thereby results in a semi-blind EMVA (SbEMVA). For a static ISI channel, simulation results indicate that the SbEMVA is near-optimal in error performance. For a Ricean fading ISI channel with a spread factor of 0.01, a K-factor of 5, and at a BER of 10/sup -3/, we find that the SbEMVA is about 4.dB better titan the ML receiver that uses the channel estimate obtained from just the training data.