Message-Passing Receiver Architecture with Reduced-Complexity Channel Estimation

We propose an iterative receiver architecture which allows for adjusting the complexity of estimating the channel frequency response in OFDM systems. This is achieved by approximating the exact Gaussian channel model assumed in the system with a Markov model whose state-space dimension is a design parameter. We apply an inference framework combining belief propagation and the mean field approximation to a probabilistic model of the system which includes the approximate channel model. By doing so, we obtain a receiver algorithm with adjustable complexity which jointly performs channel and noise precision estimation, equalization and decoding. Simulation results show that low-complexity versions of the algorithm - obtained by selecting low state-space dimensions - can closely attain the performance of a receiver devised based on the exact channel model.