Efficient CNN-Based Time-Domain Denoising of Impulsive Noise in NB-PLC Systems

In this study, a convolutional neural network (CNN)-based time-domain denoising approach is proposed to suppress impulsive noise which is considered as the most sever impairments in narrowband powerline communications (NB-PLC). Unlike conventional techniques, such as clipping and blanking, the proposed method does not require prior knowledge of noise statistics. The introduced CNN network is trained using synthetically generated OFDM signals corrupted by Middleton Class-A impulsive noise, calibrated from real NB-PLC measurement data. Extensive G3-PLC-compliant simulations demonstrate that the proposed method significantly outperforms classical blanking and clipping schemes. At an SNR of 10 dB, the proposed CNN achieves a mean squared error (MSE) of 1.2×10−4, compared to 2.3×10−4 and 2.5 × 10−4 for blanking and clipping, respectively, under time-varying impulsive noise conditions. Moreover, the receiver incorporating the denoising method closely approaches the ideal AWGN reference under low impulsive noise density and for SNR values above 12 dB.