Deep Learning for MPC Extraction and False Detection

Published

May 28, 2025

Author(s)

Howard Dai, Jack Chuang, Jian Wang, Samuel Berweger, David Griffith

Abstract

Multipath component (MPC) extraction is critical for channel modeling and joint communications and sensing (JCAS). The super-resolution algorithm known as CLEAN-Space-Alternating Generalized Expectation (CLEANSAGE) is widely used for MPC extraction. Because CLEAN-SAGE is based on maximum likelihood estimation (MLE), it can experience performance degradation and false detection events when a mismatch occurs between the theoretical model and the received signal. Moreover, the complexity of CLEAN-SAGE makes it challenging to support real-time or near-real-time applications because the algorithm iteratively extracts parameters of MPCs with gradient search. This paper proposes two machine learning (ML)-based solutions to address these issues: a classification model for false signal detection and regression models for direct MPC parameter extraction. The classification model uses a convolutional neural network (CNN) to predict true or false detection from the beamformed signals, which can be used alongside CLEAN-SAGE to reduce false detection rates. Our model achieves greater than 90 % classification accuracy on simulated data and a smaller set of real-world examples collected by NIST's 140 GHz phased array channel sounder. Two regression models have been developed using YOLO to focus on reducing estimation errors or a CNN-based autocoder to focus on processing speed. Both models directly predict the number of MPCs, the range, and the azimuth angle of each MPC from a heatmap. We leverage the NIST Wi-Fi JCAS simulation bench to generate heatmap training data in the signal-to-noise ratio (SNR) range of –10 dB to 25 dB before pulse compression. Our models achieve accuracy comparable to CLEAN-SAGE and a false detection rate less than 0.5 % in multi-targets scenarios. The results also demonstrate all our regression models are running faster than CLEAN-SAGE.

Proceedings Title

SPIE Defense+Commerical Sensing

Conference Dates

April 13-17, 2025

Conference Location

Orlando, FL, US

Conference Title

"Signal Processing, Sensor/Information Fusion, and Target Recognition"

Pub Type

Conferences

Download Paper

Local Download

Keywords

Deep learning, NextG, Joint communications and sensing, Channel sounding

Citation

Dai, H. , Chuang, J. , Wang, J. , Berweger, S. and Griffith, D. (2025), Deep Learning for MPC Extraction and False Detection, SPIE Defense+Commerical Sensing, Orlando, FL, US, [online], https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=959380 (Accessed March 5, 2026)