Fast convergence with q-expectation in EM-based blind iterative detection

We develop the q parameterized Expectation Maximization (q-EM) algorithm for parameter estimation based on incomplete observations. The q-EM algorithm is a one-parameter generalization of the standard EM algorithm that has been successfully used in many applications. With q-EM algorithm, we investigate iterative schemes for joint channel estimation and signal detection over frequency selective channels. We show that the convergence speed is improved by replacing the standard expectation with q-expectation, which was first introduced in the Tsallis entropy literature. Simulation results with different q values are given. A variable-q strategy is proposed to further improve the system performance.