This paper investigates the linear precoder design for $K$-user interference channels of multiple-input multiple-output (MIMO) transceivers under finite alphabet inputs. We first obtain general explicit expressions of the achievable rate for users in
the MIMO interference channel systems. We study optimal transmission strategies in both low and high signal-to-noise ratio (SNR) regions. Given finite alphabet inputs, we show that a simple power allocation design achieves optimal performance at high SNR whereas the well-known interference alignment technique for Gaussian inputs only utilizes a partial interference-free signal space for transmission and leads to a constant rate loss when applied naively to finite-alphabet inputs. Moreover, we establish necessary conditions for the linear precoder design to achieve weighted sum-rate maximization. We also present an efficient iterative algorithm for determining precoding matrices of all the users. Our numerical results demonstrate that the proposed iterative algorithm achieves considerably higher sum-rate under practical QAM inputs than other known methods.
