Enabling SWIPT with Machine Learning-Based Multisine Signal Classification

This paper presents a novel methodology for employing multisine waveforms in simultaneous wireless information and power transfer (SWIPT) systems, utilizing software-defined radio tools. The proposed approach encodes information by varying the number of carriers in the multisine signals, while simultaneously enabling the receiver to harvest energy. A comprehensive dataset is generated by transmitting various waveforms and measuring the harvested power across different distances. The primary objectives are to accurately classify the received waveforms to extract information and validate the dataset using established machine learning techniques. Experimental evaluations demonstrate that basic supervised machine learning models, specifically multinomial logistic regression and support vector machine, achieve a high accuracy of 99.2% and 100%, respectively. These results underscore the capability of the proposed system to effectively distinguish not only between binary signal classes but also among multiple signal classes.