Sphere decoding inspired approximation method to compute the entropy of large Gaussian mixture distributions

The computation of mutual informations of large scale systems with finite input alphabet and Gaussian noise has often prohibitive complexities. In this paper, we propose a novel approach exploiting the sphere decoding concept to bound and approximate such mutual information term with reduced complexity and good accuracy. Using Monte-Carlo simulations, the method is numerically demonstrated for the computation of the mutual information of a frequency- and time-selective channel with QAM modulation.