Blind FIR channel estimation in multichannel cyclic prefix systems

In this paper, we revisit a number of classical blind estimation techniques for FIR multichannels when applied to communication systems that are based on the introduction of a cyclic prefix. These techniques include techniques based on deterministic modeling of the unknown symbols such as (signal and noise) subspace fitting methods, subchannel response matching (SRM), deterministic maximum likelihood (DML), and techniques based on a Gaussian white noise model for the unknown symbols such as Gaussian ML (GML) methods and covariance matching. The presence of a cyclic prefix transforms spatiotemporal channels into a set of parallel spatial channels, coupled by the discrete Fourier transform (DFT) of the FIR channel impulse response. The associated blind channel estimation methods become computationally much more attractive and also become more straightforward to analyze and to compare in terms of performance. Working in the DFT domain reveals immediately that temporal whiteness of the additive noise is unessential, only spatial whiteness matters. Furthermore, the blind channel identifiability conditions become extremely weak when zero padded (ZP) systems are considered.