Evaluating Various Machine Learning Techniques in Selecting Different Modulations in G3-PLC Protocol

The dynamic nature of power line channels, has made it difficult to effectively estimate power line link quality. Incorrect estimation can lead to data packet retransmissions and even unnecessarily to the creation of new routes consuming additional communication channel bandwidth or flooding the communication network. G3-PLC and Hybrid protocol standard is based on instantaneous Link Quality Indicator (LQI) estimation to determine the suitable modulation schemes (ROBO, BPSK, QPSK and 8PSK), but the instants estimation may not be able to foresee the impending channel state which may result to communication impairment. In this context, a promising approach to estimating channels is the data-based link quality classifier using machine learning (ML) algorithms. This article outlines a measurement campaign using a local network of smart meters, collecting link quality data, and implementing ML prediction models. Hence, we present models for classifying the PLC link quality in different modulation patterns. Accuracy performance is compared against various classification models such as LR (Logistic Regression), SVM (Support Vector Machine), RFC (Random Forest Classifier), and ANN (Artificial Neural Networks) algorithms. The link quality classifier shows a clear improvement in the modulation selection process as the classical approach in the G3-PLC protocol.