Impact of Room Size on Machine Larning-based Material Prediction using Channel Impulse Response

A comprehensive and accurate description of the indoor radio environment is a prerequisite for environment-aware wireless communications. Approaches for the identification of the materials of the surfaces are scarcely investigated in the literature. The use of machine learning approaches is promising for automatic and seamless identification. The ability of the models to correctly identify the materials based on channel impulse response (CIR) can be affected by the properties of the rooms used in the training process. In this paper, we study the impact of the sizes of the rooms used to train the models on their ability to identify the materials in new room sizes. Based on the results, training the models using various room sizes–both smaller and larger than the rooms where the model will be applied–results in accurate models.