Reliable Near-Field Multi-User Positioning Informed by Two-Stage MUSIC

The paper introduces MUSIC-Net, an end-to-end deep learning framework for near-field multi-user positioning that incorporates a two-stage MUSIC algorithm to isolate line-of-sight signal components and estimate surrogate distances. By embedding these MUSIC-derived objects into training, the method bypasses separate parameter estimation and path/source association, directly recovering user positions even in mixed LoS/NLoS multipath scenarios. Additionally, the authors employ split conformal prediction to provide statistically guaranteed confidence sets for each user’s position, achieving lower mean positioning error and tighter prediction regions compared to existing benchmarks.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv AI.

• benchmarks