Estimation of Room Shape Using Radio Propagation Channel Analysis

Abstract

This study proposes a new method to estimate room shape using radio waves. Unlike existing heuristic methods, the proposed method uses accumulated radio channel data and the neural network to predict room shapes. The radio channel data is generated by three dimensional (3D) ray tracing simulation and actual measurements. The neural network used the channel data for the training process. After training the neural network, we evaluated the performance variation resulting from changes in the channel environment parameters used in the neural network training, the size of the room, hyperparameters of the neural network, and the location of the transceivers. The results showed that better performance was observed when more sets of channel parameters were used as inputs to the neural network. In addition, the results obtained from measurements show that the estimation of the room shape through the neural network works well not only in the simulation but also in the real world.