Capacity Analysis for Gaussian and Discrete Memoryless Interference Networks

Interference is an important issue for wireless communication systems where multiple uncoordinated users try to access to a common medium. The problem is even more crucial for next-generation cellular networks where frequency reuse becomes ever more intense, leading to more closely placed co-channel cells. This thesis describes our attempt to understand the impact of interference on communication performance as well as optimal ways to handle interference. From the theoretical point of view, we examine how interference affects the fundamental performance limits, and provide insights on how interference should be treated for various channel models under different operating conditions. From the practical design point of view, we provide solutions to improve the system performance under unknown interference using multiple independent receptions of the same information. For the simple two-user Gaussian interference channel, we establish that the simple Frequency Division Multiplexing (FDM) technique suffices to provide the optimal sum- rate within the largest computable subregion of the general achievable rate region for a certain interference range. For the two-user discrete memoryless interference channels, we characterize different interference regimes as well as the corresponding capacity results. They include one- sided weak interference and mixed interference conditions. The sum-rate capacities are derived in both cases. The conditions, capacity expressions, as well as the capacity achieving schemes are analogous to those of the Gaussian channel model. The study also leads to new outer bounds that can be used to resolve the capacities of several new discrete memoryless interference channels. A three-user interference up-link transmission model is introduced. By examining how interference affects the behavior of the performance limits, we capture the differences and similarities between the traditional two-user channel model and the channel model with more than two users. If the interference is very strong, the capacity region is just a simple extension of the two-user case. For the strong interference case, a line segment on the boundary of the capacity region is attained. When there are links with weak interference, the performance limits behave very differently from that of the two-user case: there is no single case that is found of which treating interference as noise is optimal. In particular, for a subclass of Gaussian channels with mixed interference, a boundary point of the capacity region is determined. For the Gaussian channel with weak interference, sum capacities are obtained under various channel coefficients and power constraint conditions. The optimalities in all the cases are obtained by decoding part of the interference. Finally, we investigate a topic that has practical ramifications in real communication systems. We consider in particular a diversity reception system where independently copies of low density parity check (LDPC) coded signals are received. Relying only on non-coherent reception in a highly dynamic environment with unknown interference, soft-decision combining is achieved whose performance is shown to improve significantly over existing approaches that rely on hard decision combining