Coping with Non-Line-of-Sight

NLOS mitigation with machine learning

We designed and trained a neural network that can recognize and correct non-line-of-sight (NLOS) conditions between two UWB devices by analyzing the CIR. The neural network was then quantized and optimized to run directly on the MCU (at-the-edge).
The results showed how the ranging error can be reduced by up to 50%, with a 40% improvement in the position error.

Dealing with NLOS due to the human body

In parallel with the development "generic NLOS" mitigation, the activities that involved people wearing UWB tags (social distancing and MUSE visitors mobility) showed how the impact of the body significantly reduces the ranging accuracy. This indicates the need to tackle human body NLOS (HNLOS) specifically.
A second study therefore focused on the PHY-level information provided by the radio to directly determine this special NLOS condition, and on the identification of the appropriate radio configuration. The outcome was a custom PHY configuration that proved to be significantly better than the generic one recommended by the manufacturer of the UWB board.

Related publications

Robust Ultra-wideband Range Error Mitigation with Deep Learning at the Edge
S. Angarano, V. Mazzia, F. Salvetti, G. Fantin, M. Chiaberge. Engineering Applications of Artificial Intelligence, Volume 102, 2021. https://doi.org/10.1016/j.engappai.2021.104278
POLITO

Human Occlusion in Ultra-wideband Ranging: What Can the Radio Do for You?
V. Le, M. Trobinger, D. Vecchia, G.P. Picco. International Conference on Mobility, Sensing and Networking (MSN), Guangzhou (China), December 14-16, 2022. https://doi.org/10.1109/MSN57253.2022.00016
UNITN