Deep learning based channel estimation in PLC systems

Abstract

Power line communication systems (PLC) are used for data transmission. Accurate channel state information (CSI) is essential for the receiver design in such systems, however, impulsive noise poses a challenge for the channel estimation task. In this paper, we propose a deep learning based method for PLC channel estimation which is resistant against impulsive noise as well as the additive white Gaussian noise (AWGN). The proposed deep neural network consists of three sub-networks: The first one is a denoising network which aims to remove the noise from the received signal. The second sub-network offers a low-accuracy estimation of the channel using the denoised signal. The third sub-network is designed for high-accuracy channel estimation. The training of the proposed network is done in two stages: Firstly, the denoising sub-network is trained. Secondly, by freezing the trained parameters of the denoising network, the two-channel estimation sub-networks are trained. Moreover, we have derived the Cramer Rao lower bound of the PLC channel estimation problem. The proposed method has been evaluated through various simulation scenarios which confirm the superiority of the proposed method over its counterpart. The suggested algorithm indicates acceptable resistance against impulsive and Gaussian noises.