.png){kind=icon}
.png){kind=icon}
Return
Learning-Based Flexible Dual-Path Iterative Framework for Interference Suppression in Automotive FMCW Radars
.png){kind=icon}
PRE
AI
H
S
R
Y
X
J
M
J
G
DOI：10.1109/taes.2026.3713283.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The rapiddevelopment of frequency-modulated continuous wave (FMCW) radar has introduced critical mutual interference challenges. Currently, compressed sensing (CS) and deep learning offer promising interference suppression capabilities, while conventional CS implementations face computational bottlenecks and hyperparameter dependence. Meanwhile, the limited interpretability and generalization ability of generic deep networks are also concerns. To address these issues, a learning-based flexible dual-path iterative network (LFDPI-Net) is proposed for suppressing interference between FMCW radars. First, the interference suppression is transformed into model-driven optimization. Second, it combines the interpretability of CS-based methods with feature extraction of deep learning, using a designed mirrored convolutional neural network to perform nonlinear mapping to the target, thereby expanding the receptive field. To enhance generalization, the model flexibly learns hyperparameters in a layered manner. In addition, LFDPI-Net is devised as a dual-path feedforward model to better synchronize the processing of multiple complex-valued pulses. Finally, a multidomain joint constraint term is proposed to stabilize the optimization process by simultaneously considering both target and interference signals. A series of experiments demonstrate that LFDPI-Net can efficiently suppress interference and accurately extract target information, offering a practical solution to mutual interference in dense FMCW radar scenarios.
Keywords:
Compressed sensing (CS)
deep learning
frequency-modulated continuous wave (FMCW) radar
interference suppression
Journal
IF:
5.7
Papers:
651
Citations:
2.4W
Organization
Cited Papers