Classification of Indoor Environments in Time-Varying Channels

The growing demand of indoor applications necessitates accurate environment classification to enhance the performance and efficiency of sensor-based tracking and positioning systems. This paper introduces a deep learning approach utilizing Convolutional Neural Networks (CNNs) for real-time classification of various indoor environments, based on real-time radio frequency (RF) measurements in slow-fading settings. The proposed method captures both static and slow-fading environments, providing a more comprehensive solution. The CNN model, trained on realistic data, achieves a classification accuracy of 99.63%, significantly outperforming traditional classifiers such as decision trees, support vector machines, and k-nearest neighbors. Moreover, we analyze the impact of bandwidth on classification performance, revealing a crucial tradeoff between error rate and inference time. With a minimal prediction time and high accuracy, the proposed approach demonstrates its suitability for broad range of indoor applications.