New blind/semi-blind equalization algorithms for GSM systems

We propose a GSM channel identification algorithm based on the decision feedback of the derotation-based constant modulus (DB-CMA) equalizer. By using the entire DB-CMA equalizer output as the reference "training", the channel impulse response can be more accurately identified for sequence estimation. We also propose a new semi-blind channel identification method. Based on the blind subspace method, a shorter training sequence is combined with "reliable" subspace information for channel identification with lower sensitivity to channel order mismatch and robustness to common zeros among diversity subchannels. Both methods are based on an approximated linearization of GMSK modulation and its single-input two-output model resulted from a derotation scheme. Overall improvement in the system efficiency is achieved.