Covariance based linear precoding in the case of identical longterm channel state information

We extend the covariance based linear precoding theory to the spatial division multiple access (SDMA) downlink processing in the case that all the streams belonging to the same user have identical channel covariance matrices, which happens when the antennas of the user are located in close proximity. In this case, the precoders designed for those streams based on covariance channel state information (CSI) will be identical, therefore, they can not separate the user successfully. To overcome this, we modify the covariance matrices by keeping the eigenvectors unchanged and choosing one eigenvalue per stream if the number of streams equals to the rank of the covariance matrix. Otherwise, the eigenvalues will be divided into different groups with the rule making the sums of each group as similar as possible. Then, the precoders are designed based on the modified covariance matrices with the constraint of total transmit power. The proposed method is tested and simulation results show that it works well and can improve the bit error ratio (BER) of the system significantly.