Linear Precoding for Finite-Alphabet Inputs Over MIMO Fading Channels With Statistical CSI

This paper investigates the linear precoder design that maximizes the average mutual information of multiple-input multiple-output fading channels with statistical channel state information known at the transmitter. It formulates the design from the standpoint of finite-alphabet inputs, which leads to a problem that is very important in practice but extremely difficult in theory: First, the average mutual information lacks closed-form expression and involves prohibitive computational burden. Second, the optimization over the precoder is nonconcave and thus easily gets stuck in local maxima. To address these issues, this study first derives lower and upper bounds for the average mutual information, in which the computational complexity is reduced by several orders of magnitude compared to calculating the average mutual information directly. It proves that maximizing the bounds is asymptotically optimal and shows that, with a constant shift, the lower bound actually offers a very accurate approximation to the average mutual information for various fading channels. This paper further proposes utilizing the lower bound as a low-complexity and accurate alternative for developing a two-step algorithm to find a near global optimal precoder. Numerical examples demonstrate the convergence and efficacy of the proposed algorithm. Compared to its conventional counterparts, the proposed linear precoding method provides significant performance gain over existing precoding algorithms. The gain becomes more substantial when the spatial correlation of MIMO channels increases.