An Asymptotic Analysis of the MIMO BC under Linear Filtering

We investigate the MIMO broadcast channel in the high SNR regime when linear filtering is applied instead of dirty paper coding. Using a user-wise rate duality where the streams of every single user are not treated as self-interference as in the hitherto existing stream-wise rate dualities for linear filtering, we solve the weighted sum rate maximization problem of the broadcast channel in the dual multiple access channel. Thus, we can exactly quantify the asymptotic rate loss of linear filtering compared to dirty paper coding for any channel realization. Having converted the optimum covariance matrices to the broadcast channel by means of the duality, we observe that the optimal covariance matrices in the broadcast channel feature quite complicated but still closed form expressions although the respective transmit covariance matrices in the dual multiple access channel share a very simple structure. We immediately come to the conclusion that block-diagonalization is the asymptotically optimum transmit strategy in the broadcast channel. Out of the set of block-diagonalizing precoders, we present the one which achieves the largest sum rate and thus corresponds to the optimum solution found in the dual multiple access channel. Additionally, we quantify the ergodic rate loss of linear coding compared to dirty paper coding for Gaussian channels with correlations at the mobiles.