Multiuser blind identification using a linear parameterization of the channel matrix and second-order statistics

We observe that the channel matrix in the standard multiuser, multichannel (MIMO) digital communications model is linear in the channel coefficients. Also, previous work incorporating “basis functions” suggests that the multipath channel itself is in a subspace formed by delayed versions of the transmission pulse. Hence the channel matrix is linear in the coefficients of this subspace. We propose two algorithms based on the sample covariance matrix of the received signal (i.e., second-order statistics) that take advantage of this linear parameterization: a new identification algorithm that estimates the outer product of the model coefficients via multiplication by a predetermined matrix, and a multiuser version of the previously presented “subspace method” that employs knowledge of the transmission pulse. While both methods are superior to the original non-parameterized subspace method in terms of computation and performance, the new method requires less computation and in some cases outperforms the other