Channel capacity and mutual information have been studied extensively for various types of wire-line and wireless multi-user communication channels. Among the vast information theoretic literature, most of the results are based on the assumption that the channel inputs are Gaussian distributed. However, Gaussian inputs can never be realized in practical systems. The inputs are usually taken from finite alphabets, which can significantly depart from Gaussian distribution. A large nonlinear gap exists between the theoretical capacity and practical achievable rate. This nonlinear gap indicates that Gaussian-input assumption may not provide a realistic design guideline to practical systems. Maximizing mutual information over channels with finite alphabet inputs will benefit not only bandwidth efficiency but also bit error rate performance. However, much less work has been done for this important topic. This is mainly due to lack of closed-form solution and high computational complexity.
The project investigates the direct maximization of mutual information and throughput over multi-user channels with finite alphabet inputs. The computational complexity problem is tackled by developing mathematically tractable and practically accurate algorithms via employing graph-based message-passing techniques. Shaping matrices are introduced to the maximization of mutual information. Parameterized approaches are developed to solve optimal shaping matrices which lead to the global maxima of the mutual information. Research efforts focus on multiple access channels, broadcast channels and interference channels, which are the fundamental channel scenarios of multi-user communications. The channel state information and/or channel covariance information are made available to the transmitter and receiver for the maximization of mutual information. Both frequency-flat fading and frequency-selective fading channels are explored.
