Recovering clipped OFDM symbols with Bayesian inference

A major problem with multicarrier transmissions is the near Gaussian behavior of the data stream entering the High Power Ampli er (HPA). This causes distortion (some samples are clipped) that must be corrected at the transmit or receive end, in order to improve detection performance. In this paper, we propose a new approach that recovers the distorted samples at the receiver. It is based on an \augmented" Bayesian model which captures the nonlinear behavior of the HPA. Estimates of the input symbols are then obtained with a hybrid deterministic/stochastic algorithm. Simulations over frequency selective channels show that when clipping is severe, our method outperforms existing methods.