Comparative analysis of different physics informed neural network architectures for spectrum occupancy prediction in Ara

Abstract

In this study, we present a comprehensive analysis of different neural network architectures integrated into Physics-Informed Neural Networks (PINNs) for spectrum occupancy prediction in wireless communication systems. The rapid expansion of wireless technologies has intensified the demand for efficient spectrum management strategies, making accurate spectrum occupancy prediction essential. Traditional methods often fall short in dynamic and complex environments, necessitating advanced approaches like PINNs that embed physical laws into neural network frameworks to enhance prediction accuracy and robustness. The work focuses on the application and comparative performance of Convolutional Neural Networks, Fully Connected Neural Networks, and Encoder-Decoder architectures within PINNs. Our findings highlight the strengths and limitations of each architecture in capturing spatial patterns and handling in pathloss prediction.