Label Noise Resistant f-Divergence Receiver for Power Line Communications

Nicola Novello, Andrea M. Tonello

Abstract: Security in power line communications (PLC) is a crucial topic, given their deployment in smart grids. In this paper, we consider the presence of an attacker who aims to damage the communication between two PLC nodes by inserting label noise during the channel estimation phase. We present a new approach that uses an f-divergence-based neural receiver to mitigate the effect of label flipping. Although the presented receiver works for any scenario, we study the performance of different f-divergences on a real dataset collected from a PLC system employing joint voltage and impedance modulation. The numerical results demonstrate that the shifted log (SL) divergence achieves a better performance with respect to the famous cross-entropy minimization approach, both in the absence and presence of label noise.