Original title: Reliable Near-Field Multi-User Positioning Informed by Two-Stage MUSIC
Jiaying Li, Haifeng Wen, Changsheng You, Yuanwei Liu +1 · September 2026
Pinpointing multiple wireless devices close to an antenna is tricky. Reflected signals bounce around and interfere with direct signals, which ruins accuracy. Older tools try to measure every bounce and match each path to the right device. This math takes heavy computing power, builds up mistakes, and cannot promise reliable results. Researchers made a deep learning tool called MUSIC-Net to fix this. It uses a signal technique called two-stage MUSIC during training. This helps the network separate the direct signal path and find a stand-in distance. Because of this, it predicts user locations directly without mapping out extra reflections. It also adds a statistical check to draw reliable confidence zones. Tests show it cuts positioning errors and shrinks these uncertainty zones compared to older benchmarks.
Imagine trying to find a friend in a crowded, echo-filled stadium by listening for their voice. The echoes and other voices mix with the direct sound, making it nearly impossible to tell exactly where the friend is standing. In this paper, the positioning system faces a similar challenge with wireless signals that bounce off obstacles. It uses a two-stage technique to isolate the direct signal from the reflections, and then provides a statistically guaranteed zone where the user must be located, rather than just a single guess.
Reliable positioning is crucial for future wireless systems like 5G and IoT, but signal reflections from obstacles often cause errors. This work introduces a method that not only improves accuracy but also provides a statistical guarantee on the positioning confidence, which is vital for safety-critical applications such as autonomous navigation. It achieves this without the heavy computation of traditional approaches.
The abstract reports only numerical simulations, not real-world experiments. The method is designed for near-field mixed LoS/NLoS scenarios; its effectiveness in far-field or purely NLoS conditions is not addressed. The computational cost of training the deep network and the requirement of a calibration set for conformal prediction are not examined.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Reliable Near-Field Multi-User Positioning Informed by Two-Stage MUSIC — Jiaying Li, Haifeng Wen, Changsheng You et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.