Memory-Efficient 1D Neural Network for PMCW Interference Mitigation in FMCW-Based Sensing Systems

View at publisher Summarise with VIRAT
Authors: Heekwon Yoon, Seonmin Cho, Seongwook Lee
Cited by: 0
References: 8

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

In this paper, we propose a memory-efficient onedimensional (1D) neural network to mitigate interference generated by phase-modulated continuous-wave signals in frequencymodulated continuous-wave radar systems. The coexistence of heterogeneous automotive radar systems operating in overlapping frequency bands leads to significant mutual interference. Recently, deep learning-based techniques have been proposed to suppress interference in range-Doppler (RD) maps, owing to their effectiveness in handling complex and diverse interference patterns. However, existing deep learning approaches using twodimensional (2D) convolutional neural network (CNN) generally process the entire RD map as input, leading to quadratic growth in memory usage as the size of RD map increases. To address this issue, we propose a line-wise processing approach in which each range-bin slice of the RD map is independently restored by a 1D neural network. The proposed architecture maintains nearly constant peak GPU memory (PGM) usage regardless of the size of RD map. Simulation results demonstrate that the proposed model achieves interference suppression performance comparable to a 2D CNN-based method, with an average difference in masked peak signal-to-noise ratio within 0.7 dB. At the same time, the proposed model significantly reduces memory requirements. Specifically, for an RD map of size$256 \times 256$, the proposed model reduces PGM usage by 85.89 % compared to the 2D CNN-based approach.

Fields of Study & Subjects

Radar Systems and Signal Processing Advanced SAR Imaging Techniques Advanced Neural Network Applications

Publication details

Journal / source
International Conference on Artificial Intelligence in Information and Communication (ICAIIC)
Pages
346–350
Publication date
24 February 2026
Language
EN