High SNR Analysis for MIMO Broadcast Channels: Dirty Paper Coding vs. Linear Precoding

Abstract: We study the MIMO broadcast channel and compare the achievable throughput for the optimal strategy of dirty paper coding to that achieved with sub-optimal and lower complexity linear precoding (e.g., zero-forcing and block diagonalization) transmission. Both strategies utilize all available spatial dimensions and therefore have the same multiplexing gain, but an absolute difference in terms of throughput does exist. The sum rate difference between the two strategies is analytically computed at asymptotically high SNR, and it is seen that this asymptotic statistic provides an accurate characterization at even moderate SNR levels. Furthermore, the difference is not affected by asymmetric channel behavior when each user a has different average SNR. Weighted sum rate maximization is also considered, and a similar quantification of the throughput difference between the two strategies is performed. In the process, it is shown that allocating user powers in direct proportion to user weights asymptotically maximizes weighted sum rate. For multiple antenna users, uniform power allocation across the receive antennas is applied after distributing power proportional to the user weight.