A subspace based approach for channel estimation and equalization in GSM receivers

For rapidly changing environment in mobile communications, we need to look for channel identification/equalization techniques that can work with small data records. Non-statistical approaches are preferable for this reason. The 26-bit training sequence in a GSM data frame may not be sufficient for accurate channel identification at lower values of SNR. With finite number of data samples available, we exploit subspace structure (using a deterministic framework) to assist channel estimation at such SNRs. Subspace techniques also make identifiability of larger memory channels (which are not identifiable with only 26 training bits) possible. An algorithm is also suggested for estimation of equalizer coefficients.