Variational Bias Classification and Mitigation Filter for Robust Localization Based on Cellular Signals

Cellular signals have become a promising candidate for localization due to advantages such as broad coverage and high signal quality, especially in dense urban environments. However, pedestrians, vehicles, and buildings substantially increase non-line-of-sight (NLOS) signal propagation, leading to gradually varying biases in the ranging measurements. Additionally, ranging measurements may contain outliers due to factors such as electromagnetic interference and hardware impairments. The lack of consideration for the statistical differences between these biases has diminished the performance of traditional robust methods. In this paper, we propose a novel bias classification and mitigation filter (BCMF) model for robust localization using cellular signals and apply variational Bayesian inference (VBI) to derive the prediction and update equations for this model. In the proposed filtering model, irrelevant outliers will be mitigated, while the gradually varying bias information will be propagated across time steps, thus improving localization accuracy. We also conduct numerical stability and computational complexity analyses to demonstrate the feasibility of the proposed filtering method in engineering applications. Finally, simulation and real-world localization results demonstrate that the proposed filtering method significantly outperforms several state-of-the-art filtering techniques in scenarios where ranging measurements are contaminated by both outliers and NLOS biases.