Information geometry of maximum partial likelihood estimation for channel equalization

Information geometry of partial likelihood is constructed and is used to derive the em-algorithm for learning parameters of a conditional distribution model through information -theoretic projections. To construct the coordinates of the information geometry, an Expectation-Maximization (EM) framework is described for the distribution learning problem using the Gaussian mixture probability model. It is shown that the information-geometric em-algorithm is equivalent to EM to establish its convergence. The algorithm is applied to channel equalization by distribution learning and its rapid convergence characteristics is demonstrated through simulation studies.