In wireless channels and many other channels, interference is a central phenomenon. Mitigating interference is a key to improving system performance. To find the limit of the achievable rates for these channels in the presence of interference, the two-user Gaussian interference channel has been the subject of intensive study in network information theory. However, most current results have been obtained by assuming Gaussian input distributions. While optimal in single-user Gaussian channels, the issue with this assumption is that the Gaussian noise becomes the worst noise when the input distribution is also Gaussian. In this thesis, we propose a class of discrete constellations. We show that this class of constellations can automatically achieve the same sum rates as schemes that treat interference as noise or perform time sharing.

In wireless channels and many other channels, interference is a central phenomenon. Mitigating interference is a key to improving system performance. To find the limit of the achievable rates for these channels in the presence of interference, the two-user Gaussian interference channel has been the subject of intensive study in network information theory. However, most current results have been obtained by assuming Gaussian input distributions. While optimal in single-user Gaussian channels, the issue with this assumption is that the Gaussian noise becomes the worst noise when the input distribution is also Gaussian. In this thesis, we propose a class of discrete constellations. We show that this class of constellations can automatically achieve the same sum rates as schemes that treat interference as noise or perform time sharing.