Channel Estimation in Block Fading Channels Using the Factor Graph EM Algorithm

Factor graphs, and message-passing over these graphs using the Sum-Product Algorithm (SPA), are an important method for unifying the tasks of channel estimation and decoding. In this paper, we use a variant of the Expectation-Maximization (EM) Algorithm that is formulated as a modification to the sum-product algorithm (SPA) over factor graphs. We firstly advance the theory of this factor graph EM Algorithm, showing that it has attractive complexity and graph-theoretic properties in the case of a factor graph with cycles. Subsequently, we apply this algorithm to the problem of state estimation in block interference channels. Simulation results are given, confirming that EM-based estimation-decoding improves significantly on conventional methods, in spite of low complexity.