Blind channel estimation, equalisation and CRB for OFDM with unmodelled interference

A blind channel estimation and equalisation scheme is given for orthogonal frequency-division multiplexing systems with unmodelled interference. The approach uses a generalised multichannel minimum variance principle to design an equalising filterbank that preserves the desired signal components and suppresses the overall interference. A channel estimate is then obtained by deriving an asymptotically tight lower bound of the filterbank output power, which reduces the problem to a quadratic minimisation. By imposing a special structure on the received signal, the performance is shown to be significantly increased. To assess the performance of the proposed scheme, an unconditional Crameacuter-Rao bound (CRB) is derived which, similar to the proposed blind channel estimator, does not assume knowledge of the transmitted information symbols. The CRB serves as a lower bound for all unbiased blind channel estimation schemes. Numerical examples show that the proposed scheme approaches the CRB as the SNR increases. It also exhibits low sensitivity to unknown narrowband interference and compares favourably with a subspace blind channel estimator