Towards Automated and Interpretable Pathloss Approximation Methods

Ahmad Anaqreh, Shih-Kai Chou, Irina Barašin, Carolina Fortuna

Keywords: machine learning, automated discovery, pathloss, wireless, interpretable, kolmogorov arnold network, symbolic regression, genetic programming

TL;DR: Studying pathloss appoximation using interpretable machione learning methods such as symbolic regression and KANs.

Abstract: Modeling propagation path loss is crucial for optimizing next-generation wireless communication systems, including 5G and beyond. This work explores the use of Deep Symbolic Regression and Kolmogorov-Arnold Networks as innovative methods for approximating path loss models such as Alpha-Beta-Gamma and Close-In which are commonly used in urban micro- and macro-cellular scenarios. By integrating the predictive power of machine learning with the symbolic approaches, these methods achieve high accuracy approximation across a wide range of frequencies and propagation conditions. Through neural-guided symbolic regression and interpretable architectures, this work demonstrates how these approaches can simplify path loss modeling while maintaining robust performance. Validating these methods highlights their potential to effectively approximate path loss models in wireless communication systems.

Submission Number: 17