On Gaussian Interference Channels with Mixed Gaussian and Discrete Inputs

This paper studies the sum-rate of a class of memoryless, real-valued additive white Gaussian noise interference channels (IC) achievable by treating interference as noise (TIN). We develop and analytically characterize the rates achievable by a new strategy that uses superpositions of Gaussian and discrete random variables as channel inputs. Surprisingly, we demonstrate that TIN is sum-generalized degrees of freedom optimal and can achieve to within an additive gap of O (1) or O (log log (SNR)) to the symmetric sum-capacity of the classical IC. We also demonstrate connections to other channels such as the IC with partial codebook knowledge and the block asynchronous IC.