Framework for the Machine Learning Based Wireless Sensing of the Electromagnetic Properties of Indoor Materials

Abstract

Available digital maps of indoor environments are limited to a description of the geometrical environment, despite there being an urgent need for more accurate information, particularly data about the electromagnetic (EM) properties of the materials used for walls. Such data would enable new possibilities in the design and optimization of wireless networks and the development of new radio services. In this paper, we introduce, formalize, and evaluate a framework for machine learning (ML) based wireless sensing of indoor surface materials in the form of EM properties. We apply the radio-environment (RE) signatures of the wireless link, which inherently contains environmental information due to the interaction of the radio waves with the environment. We specify the content of the RE signature suitable for surface-material classification as a set of multipath components given by the received power, delay, phase shift, and angle of arrival. The proposed framework applies an ML approach to construct a classification model using RE signatures labeled with the environmental information. The ML method exploits the data obtained from measurements or simulations. The performance of the framework in different scenarios is evaluated based on standard ML performance metrics, such as classification accuracy and F-score. The results of the elementary case prove that the proposed approach can be applied for the classification of the surface material for a plain environment, and can be further extended for the classification of wall materials in more complex indoor environments.