Comparison of methods for iterative joint data detection and channel estimation

This paper compares iterative deterministic and Markov chain Monte Carlo algorithms approximating the maximum likelihood of joint data detection and channel estimation with respect to the quality of an initial channel estimate. The quality of the initial channel estimate is measured by the normalized mean squared error between estimated and true channel. The deterministic method does not take the instantaneous quality of the channel estimation or of the current data estimate into account and might get trapped in a local maximum of the likelihood function, whereas the Monte Carlo methods theoretically almost converge to the joint maximum likelihood. Based on simulation results, it is shown that a performance gain can be achieved by applying the second class of algorithms at the expense of slower convergence speed.