Gaussian Mixture Model Learning for Multipath Assisted Positioning

The wireless signal distortion decreases the precision of the estimated position. However, the distortion caused by the multipath propagation was recently shown not to decrease but even improve the precision when utilized correctly. This approach is called mutipath assisted positioning. In this paper, we propose a particle filter resampling algorithm for multipath assisted positioning exploring high likelihood areas allowing a better approximation of the posterior probability density function. Thanks to the posterior probability density function modeled as a Gaussian mixture model, we can perform the exploration with the same computational load as a regularized particle filter. The proposed algorithm allows decreasing the number of particles orderly while preserving the state of the art approach's precision. We show a comparison of the state of the art Channel-SLAM algorithm with the proposed Gaussian mixture model-based method demonstrating the significance of the achieved improvement.