EM-based channel estimation for coded multi-carrier transmissions

Expectation-maximization (EM) based iterative algorithms are investigated in order to estimate the impulse response of a frequency-selective multipath channel in a coded OFDM system. Two ways of choosing the EM complete data are compared: a complete data built from observations and transmitted symbols (CL-EM) and a complete data chosen by decomposing noise and observation components (NCD-EM). Both CL-EM and NCD-EM algorithms are derived for a coded OFDM system. The rate of convergence of both EM algorithms is theoretically determined. It is found that the rate of convergence of CL-EM is independent from the number of channel taps at high signal-to-noise ratio (SNR), while that of NCD-EM varies with the number of taps. It is shown that CL-EM converges in a few iterations. Furthermore, considering the complexity per iteration, CL-EM has a lower complexity than its counterpart. We also establish a Cramer-Rao bound (CRB) for coded OFDM transmission. Simulation results show that CL-EM has a good performance-complexity trade-off and it achieves the CRB.