A Data-Driven Signal Subspace Approach for Indoor Bluetooth Ranging

Abstract

Bluetooth ranging relies on two-way multicarrier phase difference (MCPD) channel frequency response measurements to mitigate time and phase offsets. However, the challenge of doubled multipath components under the two-way MCPD approach, especially with a low number of snapshots, further degrades the performance of the commonly utilized multiple signal classification (MUSIC) algorithm. In this article, we investigate a reduced complexity signal-subspace-based approach for wireless ranging using bluetooth low energy (BLE) in high multipath environments. We propose a novel signal subspace decomposition (SSD) algorithm where we utilize the span of individual signal subspace eigenvectors for range estimation. We formulate the integration of the Fourier transform and randomized low rank approximation into the SSD algorithm to reduce the computational complexity for better utilization in embedded devices. We then make use of the output features from the estimated pseudospectrum of the individual eigenvectors, obtained from the enhanced SSD algorithm, as an input to the long–short-term-memory (LSTM) recurrent neural network to obtain a data-driven SSD-LSTM wireless range estimator for the BLE. We evaluate our proposed approach using our real-world BLE data for single- and multiple-antenna scenarios. Our results show an improved performance of our proposed approach by more than 37%, while still enjoying the lowest computational complexity than existing MUSIC and support vector regression approaches for BLE ranging.