Additive non-Gaussian noise channel estimation by using minimum error entropy criterion

Channel estimation is an important component of wireless communications. This paper deals with the comparison between Mean Square Error (MSE) based neural networks and Minimum Error Entropy (MEE) based neural networks in additive non-Gaussian noise channel estimation. This essay analyzes MEE and MSE algorithms in several channel models utilizing neural networks. The aim of this study is first to compare the performance of an MSE-based conjugate gradient backpropagation (BP) algorithm with MEE-based BP method. The trained neural networks can be applied as an equalizer in the receiver. Moreover, to make a complete comparison between methods, we compare them in both low and high SNR regimes. The numerical results illustrate that MEE-based back propagation algorithm is more capable than the MSE-based algorithm for channel estimation. In fact, with additive non-Gaussian noise the performance of the MSE can be approximately as same as the MEE results in high SNR regime, but the MEE outperforms MSE-based method obviously in low SNR regime with non-Gaussian noise.