Crowdsourced Radio Environment Mapping by Exploiting Machine Learning

Accurate and cost-efficient radio environment mapping (REM) is of great importance to realize dynamic spectrum sharing. Two kinds of existing approaches, i.e., propagation model based approach and sensor monitoring based approach, are suffering from either inaccurate spectrum availability or high deployment cost. To solve these problems, crowdsourced REM is proposed which recruits users to fulfill the sensing tasks. In this work, we propose a novel crowdsourced REM method which exploits machine learning techniques to choose crowdsourced data for radio field intensity interpolation. The evaluation results demonstrate that the proposed method is capable of reducing the estimation error substantially compared to the existing method.