Attenuation Modeling Using Physics Guided Deep Reinforcement Learning: A Channel Estimation Use Case

Abstract

Along the path of propagation, the radio waves are subjected to a number of losses such as attenuation, refraction, obstruction etc., which can affect the signal strength and quality. Attenuation can be caused even due to changes in environmental conditions along the path of propagation. The impact of rainfall attenuation is mathematically modelled using the recommendations from International Telecommunication Union. These real time physical losses are modelled using the approach of providing the physical losses to the neural architecture. In this work, the physical loss information is provided to the neural architecture. From the results of the simulation, it can be noted that the model has learnt the variations in the dynamic environment when exposed to environmental changes and shows scientifically consistent performance. Proximal Policy optimization algorithm has exhibited better network utility and higher training rewards in comparison to Advantage Actor Critic algorithm.