Enabling Accurate Positioning in NLOS Scenarios by Hybrid Machine Learning with Denoising and Inpainting

Positioning for non-line-of-sight (NLOS) scenarios is a long-live challenging task, since the conventional approaches for positioning are mostly LOS-dependent, and perform poorly in NLOS scenarios. Thanks to the powerful computing and learning abilities of artificial intelligence (AI), the present study on machine learning (ML) based approaches shows promising potential to conquer the challenge in theory. Nonetheless, one critical aspect for using AI model is the generalizing ability, which tells the inference level in practice. This problem is more vital for ML based positioning, since the channel information (CI) for actual usage may not be aligned with that for training. One typical case is the noisy or incomplete CI, which may degrade the inference performance and even leads the trained model unusable. In this paper, a novel hybrid machine learning (HML) approach is introduced by exploiting both supervised and unsupervised learning models developed with denoising and inpainting abilities to enable accurate positioning in NLOS scenarios. The simulation shows the proposed approach can have 10 times higher accuracy than conventional approaches.