Hao Zhang

Frequency-modulated continuous-wave (FMCW) radar systems suffer from severe mutual interference in dense low-altitude uncrewed aerial vehicle (UAV) environments, critically degrading target detection fidelity and spectral efficiency in shared microwave bands. To address the challenges above, this article proposes an adaptive perceptual learning target/interference network, named APLF-Net. Initially, a multivariate coupled perception model is proposed to co-optimize the objective function with auxiliary sparse priors. Subsequently, a stepwise alternating optimization framework is designed and integrated into a hierarchical deep feedforward network for enhancing flexibility. Within this framework, the hierarchical network structure is refined to mimic optimization steps, accelerating the model data processing. The threshold convolution module based on echo energy inference is proposed to prevent excessive sparsification. Furthermore, a novel unsupervised training strategy is introduced, which leverages the intrinsic connections within the signal model to invert the sparse representations output by the network back into the original input echo, thereby completing the loss term mapping. A series of simulations and real-world experiments demonstrates that, compared to existing state-of-the-art methods, APLF-Net can rapidly suppress interference and accurately extract targets, exhibiting good performance.

The rapid development of the low-altitude economy has made reliable UAV surveillance in urban and field areas a critical task. For FMCW radar, detecting these low, slow, and small targets is severely compromised by strong clutter, particularly the pervasive sidelobe clutter from static infrastructures that masks faint target echoes. To address this challenge, this paper proposes the Physics-Informed Guided Decomposition Network (PIGD-Net) for robust and adaptive radar clutter suppression. Firstly, we reformulate the clutter suppression task as a guided component decomposition problem. Subsequently, a multi-objective optimization framework is introduced. The framework employs a convolutional sparse-enhancement module to accurately capture the target component, while concurrently suppressing the low rank clutter via a guided nuclear norm minimization, which is dynamically steered by a data-driven mask. This entire optimization process is then unfolded into the PIGD-Net architecture, enabling efficient training by accommodating the distinct characteristics of its internal module. Crucially, PIGD-Net operates under a composite unsupervised loss, enforces the inherent physical properties of each component by fusing a reconstruction term, a Hoyer-based sparsity prior, and an orthogonality penalty, thereby obviating the need for ground-truth labels. Extensive experiments on simulated and real world data validate the performance of the method. Compared to existing methods, PIGD-Net achieves efficient suppression with a fast processing time, while significantly improving the Signal-to Clutter Ratio and target fidelity.

The widespread adoption of frequency-modulated continuous wave (FMCW) radar operating in microwave bands has made mutual interference inevitable. Currently, conventional hardware-based solutions increase microwave system complexity, while microwave signal-processing methods struggle with effective interference suppression under limited samples. To address these challenges, this article proposes a range-Doppler coupled learning and reconstruction network (RDLR-Net) for adaptive suppression of interference between FMCW radars. First, the interference suppression is reformulated as a microwave signal reconstruction problem with a coupled range-Doppler optimization. Subsequently, a novel regularization constraint model is introduced, which combines the $l_{1}$ norm and target capture operator to effectively separate strong and weak scattering targets. A framework based on distributed optimization is designed and unfolded into a deep feedforward network to accelerate model training and inference. RDLR-Net adopts an unsupervised learning strategy to invert the target signal onto the original echo with interference, thus completing the loss term mapping. The simulation and multiple measured experiments demonstrate the proposed method achieves efficient interference suppression performance. Compared to state-of-the-art methods, the time required to process a single frame of measured data is only 0.0872 s, with the best interference suppression performance.

High-precision 2-D/3-D Synthetic Aperture Radar (SAR) image reconstruction from the indirect scattered echoes of hidden targets represents a core technical challenge in millimeter-wave (mmW) Non-Line-of-Sight (NLOS) environmental perception. Deep learning approaches have demonstrated exceptional performance in SAR imaging. However, existing methods are predominantly designed for Line-of-Sight (LOS) scenarios, where clean LOS simulation signals can be acquired for training purposes, a condition often difficult or impossible to meet in NLOS imaging due to complex multipath environments and noise. To tackle this issue within specific NLOS configurations, particularly those involving strong specular reflections from discrete, isolated hidden objects, we propose an Equivariant Imaging (EI) framework tailored for mmW SAR. The EI framework is a fully self-supervised learning approach that leverages the group invariance present in signal distributions, enabling robust image reconstruction from partial NLOS measurements contaminated with noise and multipath artifacts. In our method, the reconstruction function is based on a deep unfolding network with Total Variation (TV) constraints, mapping the NLOS scattered echoes to the target image. Moreover, we introduce an Adaptive Peak Convolution Network (APConv) into the reconstruction process to dynamically adjust thresholds, replacing traditional fixed-threshold methods. This enhances imaging flexibility and quality under these defined NLOS conditions. Finally, we validate the proposed method using various NLOS echo data collected through an experimental mmW system. Numerical and visual results both demonstrate the effectiveness of our approach for NLOS mmW SAR imaging tasks. The proposed EI framework thus offers a promising approach for advancing NLOS mmW SAR perception capabilities, particularly for environments and target configurations aligning with those investigated and supported by our current experiments.

Non-line-of-sight (NLOS) mmW radar imaging technology can bypass obstacles to accurately detect hidden targets and achieve high-precision 3-D imaging, demonstrating significant potential in complex environments. Conventional sparse reconstruction methods based on regularization are primarily designed for line-of-sight (LOS) scenarios and typically rely on prior scene information and empirical parameters. However, challenges such as image blurring and multipath artifacts in NLOS imaging limit the adaptability of these methods. To address this challenge, we propose a novel Adaptive Artifact Cancellation and Target Enhancement Network (ACTE-Net) for precise target reconstruction from NLOS array mmW echoes. Specifically, a Gaussian convolution operator is designed to filter out artifacts generated by multipath echo signals. Following this, an energy enhancement operator is proposed for NLOS target reconstruction, enabling precise focus on hidden targets and recovery of fine details. Based on this model, we have developed ACTE-Net by integrating data-driven deep-learning approaches. This network utilizes a 2-D holographic imaging operator to avoid large-scale matrix computations and systematically determines optimal network parameters through unsupervised training, ensuring its adaptability and generalization capability in NLOS imaging. Finally, we constructed an mmW experimental system and collected NLOS echo data for the letter "N" and trellis under two different reflector conditions for imaging validation. Extensive simulation and real-data experiments show that ACTE-Net outperforms several advanced algorithms, indicating significant advantages.

Frequency Modulated Continuous Wave (FMCW) Radar is an excellent sensor, however, the widespread popularity and application of millimeter-wave automotive radar, greatly increases the risk of mutual interference between vehicles, weakening the abilities of automotive radar to sense the environment. As a result, efficiently suppressing interference has been a challenge. To address this issue, this paper presents a modified block coordinate descent optimization with dual-domain sparsities named BCD-DS, to synchronously suppress mutual interference between automotive FMCW radars by jointly leveraging the priors in time and frequency domains. Firstly, the sparsities of the target and the interference on multiple transformation domains are analyzed. Then, the interference suppression problem is modified into an optimization framework with regularizers being sparse priors in dual domains subject to the L ₁ norm constraint. Based on the joint optimization model, the target and interference are cooperatively estimated by the Alternating Direction Method of Multipliers (ADMM) scheme. One iteration consists of two phases, with the solution obtained at each phase updating the echoes, promoting a good convergence performance. BCD-DS balances interference suppression and target signal power. Both simulated and measured experiments validate the effectiveness of the proposed algorithm. Compared to state-of-the-art algorithms, the results show that BCD-DS achieves better interference suppression performance and increases the signal-to-interference and noise ratio (SINR) by at least 20 dB, simultaneously retrieving the speeds and ranges of multiple targets with less than 1% error.

The widespread application of frequency-modulated continuous-wave (FMCW) radars leads to a significant increase in the risk of mutual interference. Conventional compressed sensing (CS)-based interference mitigation methods are limited by large computational capacity, troublesome parameter settings, and inefficient data processing. The supervised deep learning approach can overcome the above challenges; however, the measured data are hard to invert the labels. To address these problems, an effective framework, called the feature-oriented unsupervised adaptive suppression network (FUAS-Net), is proposed to suppress mutual interference between FMCW radars. FUAS-Net is mainly composed of an encoder and an interference restoration decoder. First, the signal model is constructed, and the task of interference suppression (IS) is modified into an optimization framework with regularizers. Then, the design of the encoder is inspired by approximate message passings (AMPs) by taking advantage of a priori information in the frequency domain to obtain targets. Second, the decoder is designed to complete the mapping from the target to the predicted interfered echo, alleviating the data dependency. FUAS-Net provides robust target extraction capabilities and optimizes parameters without ground truth. Finally, a novel training loss function is proposed that combines the mean square error (mse) and IS ratio (ISR) constraints to guarantee the power and sparsity of the targets. Both simulation and measured experiments validate the effectiveness of the proposed method, in comparison with existing state-of-the-art methods, and the results show that FUAS-Net can accomplish effective target extraction, expressing the potential for efficient IS.

Spaceborne synthetic aperture radar (SAR) is a promising remote sensing technique, as it can produce high-resolution imagery over a wide area of surveillance with all-weather and all-day capabilities. However, the spaceborne SAR sensor may suffer from severe radio frequency interference (RFI) from some similar frequency band signals, resulting in image quality degradation, blind spot, and target loss. To remove these RFI features presented on spaceborne SAR images, we propose a multi-dimensional calibration and suppression network (MCSNet) to exploit the features learning of spaceborne SAR images and RFI. In the scheme, a joint model consisting of the spaceborne SAR image and RFI is established based on the relationship between SAR echo and the scattering matrix. Then, to suppress the RFI presented in images, the main structure of MCSNet is constructed by a multi-dimensional and multi-channel strategy, wherein the feature calibration module (FCM) is designed for global depth feature extraction. In addition, MCSNet performs planned mapping on the feature maps repeatedly under the supervision of the SAR interference image, compensating for the discrepancies caused during the RFI suppression. Finally, a detailed restoration module based on the residual network is conceived to maintain the scattering characteristics of the underlying scene in interfered SAR images. The simulation data and Sentinel-1 data experiments, including different landscapes and different forms of RFI, validate the effectiveness of the proposed method. Both the results demonstrate that MCSNet outperforms the state-of-the-art methods and can greatly suppress the RFI in spaceborne SAR.