NLOS Wireless Localization Algorithm Based on Ray-tracing and Machine Learning

Most of the existing localization algorithms are for line-of-sight (LOS) condition. However, the localization errors of the existing algorithms are not reliable for non-line-of-sight (NLOS) condition. This paper performs wireless channel measurements in campus scenario at 3.3 GHz. The campus scenario contains both LOS and NLOS conditions. Measured channel impulse responses (CIRs) are utilized for electromagnetic (EM) parameter calibration based on ray-tracing (RT) simulation. After that, RT simulation provides accurate path loss information of each point in the campus scenario. Path loss fingerprint dataset is built for neural network training. The paper proposes a residual neural network with 12 layers for NLOS wireless localization. The residual neural network achieves a 4 meters localization minimum error, 6 meters localization mean error, and 5 meters localization root-mean-square error in the campus scenario. Consequently, the proposed wireless localization algorithm is practical and effective for the scenario with NLOS condition.