2026· IEEE Transactions on Radar Systems· Vol 4, pp. 1377-1394· 0 citations· 32 references
TL;DR
The proposed TD-PDA generalizes to unseen users with an ultralow inference latency, successfully reconstructing legible trajectories even in the presence of strong multipath interference and achieves stability comparable to a well-tuned classical PDA filter via a purely data-driven design.
Abstract
Reconstructing precise character trajectories in multipath environments represents a primary challenge for high-fidelity reconstruction for radar-based in-air writing. Existing tracking methods typically rely on dominant target assumptions, often failing to account for the interference of multipath reflections. Conversely, end-to-end deep learning approaches generally lack the physical interpretability necessary for accurate geometric drawing. In this work, we propose a novel tracking framework that combines high-resolution D-band sensing with a data-driven probabilistic data association (PDA) architecture. We leverage a 56 GHz bandwidth frequency-modulated continuous-wave (FMCW) radar setup to capture fine-grained kinematic signatures, supported by a rigorous sensor placement and calibration strategy. To address multipath interference without relying on rigid parametric assumptions, we propose the Temporal Deep PDA (TD-PDA). This architecture replaces heuristic hard assignments and classical parametric clutter models with a covariance-aware causal temporal convolutional network (TCN) fused directly into the statistical covariance spread equations of an extended Kalman filter (EKF). To train the network without heuristic bias, we introduce a track quality indicator (TQI) ensemble to automatically extract high-fidelity pseudo-labels. The framework is rigorously validated via simulated ground-truth data to quantify the geometric impacts of hardware constraints, and experimentally proven on a 1000-gesture dataset using tenfold leave-one-Subject-Out (LOSO) cross-validation. The proposed TD-PDA generalizes to unseen users with an ultralow inference latency, successfully reconstructing legible trajectories even in the presence of strong multipath interference. The model demonstrates improvements over prior deep-association architectures and achieves stability comparable to a well-tuned classical PDA filter via a purely data-driven design.
Multi-modal synthetic aperture radar (SAR)–optical object detectors raise detection accuracy by fusing complementary physical responses, but require both modalities to be simultaneously available at inference. When the optical stream becomes unavailable—under heavy cloud cover, night-time conditions, or downlink disruption—the detector reverts to SAR-only operation and accuracy degrades sharply. A natural remedy is to distil a multi-modal teacher into a SAR-only student via privileged-information knowledge distillation. However, we observe that the leading channel-wise feature-level method (CWD) reduces the student’s accuracy below the non-distilled baseline, with its smallest-target AP collapsing to near zero, because SAR speckle and target high-frequency edges share the same band and the alignment loss is dominated by broadband speckle energy. We refer to this failure mode as the speckle-fitting trap, formalize it as a gradient-pollution effect, and validate it through spectral and feature-manifold diagnostics. To counter the trap, we propose M-FSAD-KD, a full-link distillation framework whose neck-stage Fourier-gated alignment transfers low-frequency structural content while preserving target-edge high-frequency content; a joint spatial–channel attention mask, a shallow backbone adapter, and a response-level knowledge distillation (KD) term complete the chain. With a MAIENet teacher on OGSOD-1.0, the advantage of M-FSAD-KD over the strongest response-level KD baseline scales with student capacity: it matches KD on a 2.39 M-parameter student (both ≈48% mean average precision at an intersection-over-union (IoU) threshold of 0.5 (mAP50), averaged over multiple seeds) and exceeds it by 2.0 absolute points on a 19.98 M-parameter student (+8 over the non-distilled baseline), where it is the best of all distillation methods; at full convergence the 19.98 M-parameter student reaches 81.9% mAP50, within 8.9 absolute points of the multi-modal teacher. A frozen-feature transfer test to an out-of-domain SAR benchmark (SSDD ship detection) further shows that distilling from the multi-modal teacher yields substantially more transferable SAR features—about ten absolute points above the non-distilled backbone—with M-FSAD-KD transferring best. Cross-architecture validation with a dual-stream DEYOLO teacher yields 48.6% mAP50 at the student—1.1 absolute points below the MAIENet result—indicating that the framework transfers across the two representative teacher architectures tested (single-stream and dual-stream).
Yu-Ming Tong, Kai-Na Xiong, Jun Liu et al.· Remote Sensing· 0 citations
Ultrawideband (UWB) technology is critical for precise distance measurement in complex environments; however, its accuracy is severely degraded under non-line-of-sight (NLOS) propagation. In particular, human-body shadowing introduces multipath distortion and signal attenuation, resulting in significant systematic ranging errors. Existing mitigation methods typically face a tradeoff; they either rely on computationally expensive architectures unsuitable for edge-level measurement devices or use simplified feature sets that fail to capture the temporal channel dynamics required for accurate error characterization. To address this problem, this article proposes a computationally efficient hybrid deep learning framework that integrates a convolutional long short-term memory (CNN–LSTM) network with a multilayer perceptron (MLP). The dual-path architecture jointly exploits raw channel impulse response (CIR) waveforms and static channel features, enabling simultaneous modeling of spatial–temporal dependencies and nonlinear feature relationships. Experimental validation across four diverse measurement environments shows that the proposed floating-point model reduces the ranging root-mean-square error (RMSE) to 0.1186 m, significantly outperforming raw two-way ranging (TWR) and single-modality baselines. Moreover, the model remains compact, with only 17 570 parameters, and after full-integer INT8 quantization, it still achieves an RMSE of 0.1219 m, supporting practical deployment on microcontroller unit (MCU)-class edge platforms. These results demonstrate the potential of the proposed framework as a practical high-precision UWB error-mitigation solution for resource-constrained edge instrumentation and wearable IoT devices.
S. Huang, Yu-Hsiang Lin, Yi-Cheng Lai et al.· IEEE Transactions on Instrum...· 0 citations
4D millimeter-wave (mmWave) radar enables all-weather 3D object detection with reliable Doppler sensing. However, its practical application is hindered by inherent sparsity and noise. While multi-frame accumulation densifies point clouds, it inevitably introduces motion-induced spatiotemporal misalignment and geometric distortion, degrading detection accuracy in dynamic environments. To address these challenges, we propose the Kinematic Compensation Temporal Fusion Network (KCTF-Net), a novel 3D detection framework. Specifically, the Dynamic Kinematic Compensation (DKC) module explicitly aligns dynamic points in physical space, rectifying the motion-induced “smearing” effect. Furthermore, the Temporal Pillar Flow Enhancement (TPFE) module captures latent inter-frame kinematics, adaptively suppressing noise and mitigating sparsity. Finally, the Motion-Guided Attention Fusion (MGAF) module synergistically integrates density-filtered geometric priors with temporal features to reconstruct precise object geometries. Extensive experiments on the View-of-Delft (VoD) and TJ4DRadSet benchmarks demonstrate that KCTF-Net achieves state-of-the-art performance. Notably, it yields a 3D mean average precision (mAP) of 57.73%.
Xing-Kai Jin, Guang-Xian Xu, Fei Ma et al.· IEEE Robotics and Automation...· 0 citations
Millimeter-wave (mmWave) radar enables privacy-preserving and illumination-robust human motion reconstruction, but training generalizable models typically requires costly paired radar-motion recordings. Simulation can scale such supervision, yet even physics-based simulators cannot fully reproduce real-world multipath, clutter, hardware-specific response statistics, or distance-dependent resolution degradation, leaving a sim-to-real gap. We present mmSimPrior, a simulation-pretrained framework that factorizes transferable knowledge into signal, motion, and radar-to-motion mapping priors. To learn transferable signal and motion priors, we pretrain a multimodal radar encoder with a physics-informed domain-randomization curriculum designed to mitigate the sim-to-real gap by approximating real-world propagation- and acquisition-level variations, while a joint-temporal tokenizer learns a discrete prior over plausible human motion. A dual-mode mapping module predicts either motion-code distributions for structurally constrained zero-shot reconstruction or continuous motion parameters for flexible adaptation from limited real data. We further construct a 4.2M-frame, 31K-sequence dataset suite and introduce a No-Overlap Setting that prevents any exact subject-environment-location-motion tuple from appearing in both the adaptation and test sets. Experiments on mmSimPrior-Real and RT-Pose demonstrate consistent gains: with only 24 paired real sequences, mmSimPrior-Reg reduces MPJPE by 24.7-39.0% over the strongest baseline across the three environments, while mmSimPrior-Cls reduces zero-shot MPJPE by 8.5% without fine-tuning.
Ground-penetrating radar (GPR) denoising is typically optimized in the radargram domain, whereas target interpretation is often performed after migration. This separation causes an imaging-inconsistency problem: a denoiser may improve visual quality or data-domain metrics while attenuating weak diffractions and phase-coherent events required by reverse time migration (RTM). To address this problem, this letter proposes MigSup-Net, a migration-supervised learning framework that couples radargram restoration with image-domain consistency. Clean synthetic observations are generated by wave-equation forward modeling and corrupted by mixed structured–Gaussian noise to emulate both random and field-like disturbances. Smooth-background RTM of the clean scattered data is then used to construct migration-domain supervision. MigSup-Net adopts a shared encoder–decoder backbone with two prediction heads for denoised radargrams and migration-domain images, and is trained using radargram fidelity, migration fidelity, and total variation (TV) regularization. At inference, the trained network directly outputs both a restored radargram and a migration-domain image in a single forward pass. Experiments on held-out synthetic models show that MigSup-Net achieves the highest output signal-to-noise ratio (SNR), peak SNR (PSNR), radargram structural similarity index measure (SSIM), and migration SSIM among the tested conventional and learning-based baselines. A measured pipe-profile example further indicates that the proposed migration supervision suppresses incoherent field noise while preserving migration-relevant responses. These results demonstrate that migration supervision provides an effective imaging-aware constraint for GPR denoising.
Interrupted sampling repeater jamming (ISRJ) severely degrades high-resolution range profile (HRRP)-based radar target recognition by generating deceptive scatterers. However, existing learning-based ISRJ suppression methods tend to ignore the structural priors inherent in radar echoes, limiting their robustness and generalization. To address this issue, this letter proposes a structure-informed ISRJ filtering network, termed ISRJ-FNet, for robust HRRP recovery. The proposed method embeds two explicit priors into a deep filtering framework: the symmetric range distribution of ISRJ and the sparse range-localized structure of target echoes. A symmetric dual-branch architecture is further introduced to improve mask consistency under target perturbations. Experiments on measured and simulated data show that ISRJ-FNet achieves better performance than representative deep suppression baselines. The results demonstrate the effectiveness and robustness of structure-informed deep filtering for radar anti-jamming applications.
Mei Liu, Xunzhang Gao, Xiangfeng Qiu et al.· IEEE Signal Processing Lette...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
