Generalized blind subspace channel estimation

We present a systematic study of the subspace-based blind channel estimation method. We first formulate a general signal model of multiple simultaneous signals transmitted through vector channels. Based on this model, we propose a generalized subspace-based channel estimator by minimizing a novel cost function, which incorporates the set of kernel matrices of the signals sharing the target channel via a weighted sum of projection errors. We investigate the asymptotic performance of the proposed estimator - bias, covariance, mean square error (MSE) and Cramer-Rao bound - for large numbers of independent observations. We show that the performance of the estimator can be optimized by increasing the number of kernel matrices and by using a special set of weights in the cost function. Finally, we consider the application of the proposed estimator to a downlink CDMA system. The results of the computer simulations fully support our analytical developments.