BLADE: A Universal, Blind Learning Algorithm for ToA Localization in NLOS Channels

Fernando Pérez-Cruz, Chih-Kuang Lin, Howard Huang

Abstract: We study wireless localization systems where multiple synchronized infrastructure devices simultaneously transmit signals whose arrival times are measured at an object of interest. The performance of these time-of-arrival (ToA) localization systems is degraded in non-line-of-sight (NLOS) channels because multipath reflections cause a positive bias in the ToA measurements. To address this major impairment, we propose an algorithm that adaptively learns the probability distribution function of the NLOS bias for each infrastructure device. This algorithm is universal in that it can be applied to any ToA based system, where the only prior assumption about the channel environment is a conservative bound on the maximum bias delay. The algorithm is also blind in that it does not require training with ground truth samples. We characterize the performance of the proposed algorithm and the resulting maximum a posteriori estimator using two sets of over-the-air ToA measurements. In both cases, we show significant performance gains compared to a nonlinear multilateration baseline, and in particular, we show a reduction in the location estimate error by up to a factor of 4 for a live cellular LTE localization network.