Blind compensation of memoryless nonlinear distortions in sparse signals

This paper focuses on blind compensation of memoryless nonlinear distortions in sparse signals. A nonlinear distortion tends to decrease the sparsity. We propose to compensate for the unknown distortion using a general monotonic function, such that the compensated signal is as sparse as possible. Novel estimator for the compensation function is presented, which is able to compensate for both symmetric and asymmetric distortions. The selection of a sparsity measure is discussed with respect to compensation performance. The functionality of the method is evaluated in experiments with artificially distorted real-world speech signals.