Multi-Resolution Mixer Network for Localization of Multiple Sensors from Cumulative Power Measurements

Wireless sensor networks (WSNs) can consist of many inexpensive sensors that communicate using the same wireless channel. In some applications, localization of these sensors can be as crucial as collecting their monitoring data. As the number of sensors increases, the complexity of processing data in such large networks grows significantly. In general, localization methods in WSNs typically rely on data containing sensor-specific information. However, the problem becomes more challenging when data contains no sensor-specific information. To address this issue, we propose a mixer-based deep neural network to estimate sensor positions using the received cumulative signal strength that is devoid of explicit sensor-specific information. Our approach employs wavelet decomposition to extract information from the input time series, combined with patching, embedding, and mixer techniques for position estimation. We compare the performance of our model with a nonlinear Kalman filter-based state estimation method. Extensive evaluations using data generated from our simulator demonstrate that our method consistently outperforms the unscented Kalman filter (UKF) method across all scenarios.