Synthetic Power Line Communications Channel Generation with Autoencoders and GANs

Abstract

Power Line Communication (PLC) technologies have a relevant role in smart energy grids. Channel modeling is important to assess their performance and enable the development of advanced PLC solutions. In this paper, we propose an approach to channel modeling that exploits AutoEncoders (AEs) and Generative Adversarial Networks (GANs) to synthetically generate PLC Channel Transfer Functions (CTFs). A dataset obtained from measurements of CTFs is handled with an AE to extract its complete description through features. Then, a GAN is trained to generate new features that possess the same statistical distribution of the extracted ones. This allows the generation of new CTFs with the previously trained decoding part of the AE. The presented method is evaluated through simulations using a measured dataset and the results are verified with traditional metrics used to statistically characterize the channel.