An EM-based procedure for iterative maximum-likelihood decoding and simultaneous channel state estimation on slow-fading channels

The use of channel state information (CSI) is known to result in significant performance advantages in coded systems operating on fading channels. The relative performance advantages in using CSI are generally established through assessment of the two extreme cases of perfect and no CSI. Typically, the more severe the fading the greater the predicted relative performance advantage of perfect CSI. Little work has been done, however, in the development and characterization of explicit estimation techniques for recovering CSI on representative fading channels. We present one such scheme based upon use of the expectation-maximization (EM) algorithm. More specifically, we pose the maximum-likelihood (ML) decoding problem as an incomplete data problem which is easily solved using the EM algorithm. The resulting EM-based procedure provides an iterative scheme for simultaneous ML decoding and channel state estimation. We demonstrate through simulation that this scheme is capable of providing performance close to that predicted on the slow-fading Rician channel when perfect CSI is available