Machine Learning Enhanced Indoor Positioning with RIS-aided Channel Configuration and Analysis

With the rapid development of wireless technologies, future communication networks should not only enhance data transmission but also provide accurate and reliable location services. In complex indoor environment, traditional positioning methods encounter challenges related to accuracy and cost due to unpredictable attenuation, multipath interference, and other factors. This paper designs a Machine Learning Enhanced Indoor Positioning method that utilizes RIS-aided Channel Configuration and Analysis, called RCCA-MLEIP, for large-scale warehouse applications. This method consists of two stages: the RIS-aided Channel Configuration and Analysis (RCCA) stage, and the Machine Learning Enhanced Indoor Positioning (MLEIP) stage. In the RCCA stage, multiple RISs are deployed to create reflection paths associated with product tags. The phase coefficients of the RISs are adjusted multiple times, and the mixed signals observed by the reader after each adjustment are recorded. Based on these mixed signals, a system of equations is established to resolve both the phase fading and amplitude fading of each reflection path. In the MLEIP stage, a feature dataset is created using simulation methods, consisting of the phase fading and amplitude fading of multiple reference tags analyzed in the RCCA stage. We use the coordinates of the reference tags as the label dataset to train a Back Propagation (BP) neural network. The trained model can then output the coordinates of a target product tag based on the channel fading features associated with that tag, which are solved/obtained in the RCCA stage. The proposed RCCA-MLEIP utilizes RIS to create a multipath environment for positioning, effectively reducing the hardware costs of the positioning system. Moreover, employing machine learning techniques to estimate the target position can enhance both accuracy and response speed.