Applied ML to SWIPT systems

Abstract

The demand for dependable and energy-efficient wireless communication systems has increased due to the quick spread of Internet of Things (IoT) devices. Devices can now harvest energy and decode information from the same radio frequency (RF) signals thanks to the promising paradigm known as Simultaneous Wireless Information and Power Transfer (SWIPT). The trade-off between optimizing harvested energy and guaranteeing reliable information decoding is frequently difficult for conventional signal demodulation and resource allocation techniques in SWIPT receivers, particularly when hardware non-idealities, channel noise, and fading are present. This thesis investigates the use of machine learning (ML) techniques, specifically one-dimensional convolutional neural networks (1D-CNNs) and artificial neural networks (ANNs), to optimize energy harvesting and signal demodulation in hardware-based SWIPT systems. Using both single-antenna and multi-antenna SWIPT architectures, the study examines how machine learning (ML)-driven models can learn and infer the best practices for power splitting, modulation recognition, and demodulation under different channel conditions. Particular focus is placed on digital modulation schemes like Amplitude Shift Keying (ASK), Phase Shift Keying (PSK), and Quadrature Amplitude Modulation (QAM) and Analog Modulation (AM, FM),