Radio Map Estimation (RME) with Deep Progressive Network

Jashia Mitayeegiri, Shaohua Dong, Chenxi Qiu, Qing Yang, Xinrong Li, Heng Fan, Yan Huang

Abstract: The goal of radio map estimation (or RME) is to predict the radio strength at each position in a region. Due to deep learning RME has been subjected to rapid advancement in recent years. We propose a deep progressive network for RME. Our method is to gradually refine the predicted radio map using multiple networks in a recursive manner. Compared to previous approaches that predict the radio map within a single feed-forward pass, our method leverages the prediction result from the last stage as “a priori”; which contains useful contextual information of the radio map, for further refining the radio map in the next stage, resulting in improved performance. On the RME benchmark we demonstrate that the proposed method, dubbed DPN-RME, can enhance the RME results under different sampling rates and outperforms the baseline model; evidencing the efficacy of progressive architecture for RME. Code is provided at https://github.com/Jashia515/DPN-RME.