GAN-based Cyclostationary Noise Generator for Narrowband Powerline Communication Systems

Abstract

Conventionally, mathematical models are used to generate noise traces. However, those approaches may lose some details during the modeling processes. A generative adversarial network (GAN) has recently attracted researchers’ attention due to its excellent capability to generate indistinguishable data. This paper modified the SpecGAN by learning the phase information into the SpecGAN, called Phase-learned SpecGAN. The effectiveness of the proposed method is evaluated by generating cyclostationary noise, which is the dominant impulse source that existed in the narrowband powerline communication (NB-PLC) systems.