Abstract:Compared with the conventional broadside imaging mode used in traditional low-frequency radar, the high squint mode of airborne terahertz Synthetic Aperture Radar(SAR) can fully exploit its advantages of high-resolution and high-frame-rate, enables high-definition dynamic imaging of the forward squint area and significantly expands the platform's dynamic sensing range. In the broadside mode, terahertz SAR can achieve a larger Doppler bandwidth with a smaller aperture, resulting in relatively small range migration and comparatively simpler imaging processing. However, in the high squint mode, the increase in aperture length and the highly space-variant characteristics of range migration greatly raise the difficulty of motion compensation. To address this issue, this paper investigates high-squint terahertz SAR imaging techniques, employing a joint estimation method using system parameters and echo data to achieve high-precision motion parameter estimation and motion compensation. Based on airborne terahertz radar measured data, high-resolution imaging is achieved at a squint angle of 70°, with a resolution better than 0.2 m. Compared with conventional imaging methods, the proposed imaging algorithm achieves a reduction in Peak Side Lobe Ratio(PSLR) of 16.53 dB and a reduction in image entropy of 39.9%.

Abstract:Object detection technology enhances the decision-making efficiency of intelligent systems through real-time localization and accurate classification, playing a vital role in traffic monitoring, industrial inspection, and medical image analysis. Heterogeneous image fusion-based detection integrates multimodal data such as Synthetic Aperture Radar(SAR), infrared, and visible images, effectively overcoming single-modality limitations and providing more comprehensive feature representations, thereby achieving superior detection performance. This paper systematically investigates heterogeneous image fusion strategies for object detection and their adaptation to various detection frameworks. Specifically, it introduces fusion strategies at the point, region, feature, and multi-level, analyzes their technical characteristics, and examines their application within single-stage and two-stage detection frameworks along with their respective advantages and limitations. Furthermore, it synthesizes state-of-the-art approaches for joint optimization of fusion and detection. Finally, commonly used datasets and evaluation metrics are presented, along with promising application prospects in intelligent transportation, industrial inspection, and smart healthcare.

Abstract:To rapidly acquire target position information and achieve high-precision imaging detection of targets in the forward-downward-looking scenario of a flying platform, this paper proposes a real-time imaging algorithm for terahertz-band Synthetic Aperture Radar(SAR) applicable to dive scenarios, based on a forward-downward-looking three-dimensional imaging model under the uniform linear array mode. The algorithm simplifies signal processing through reasonable approximations to reduce computational complexity, thereby enabling rapid three-dimensional scene reconstruction. The feasibility of the algorithm is verified through a simulation platform, and the algorithm is mapped to a high-speed parallel Field-Programmable Gate Array(FPGA) hardware architecture. By comparing the FPGA processing results with the simulation results, the feasibility and real-time performance of the algorithm in an actual hardware system are further validated.

Abstract:Based on YOLOV8, a specialized algorithm named Synthetic Aperture Radar(SAR)-You Only Look Once(YOLO) is developed for small target detection in SAR images. The proposed architecture incorporates a Pyramid Multi-Scale Feature Aggregation(PMSFA) module, an Efficient Multi-Scale Feature Fusion(FPSConv) module, and a Lightweight Shared Detection Convolutional Head(LSDECD) into the native YOLOV8 framework. The proposed algorithm is validated by using both a self-constructed SAR image small target detection dataset and publicly available datasets. Ablation studies demonstrate the effectiveness of the three introduced modules. Comparative experiments further indicate that the proposed SAR-YOLO algorithm significantly improves small target detection accuracy compared to other object detection methods: the key metric mAP50 is increased by 11.93%, while the detection time per frame is reduced to 18 ms. The proposed algorithm provides a robust, accurate, and real-time detection framework for small target detection in SAR images.

Abstract:To address the cross-modal registration challenges between Synthetic Aperture Radar(SAR) and optical images, which arise from differences in imaging mechanisms, noise characteristics, and textures—particularly the difficulties in feature extraction from low-texture SAR images, significant domain disparities, and high computational costs of existing methods—this study proposes an improved algorithm named EnhancedXFeat based on the Accelerated Features(XFeat) network. The algorithm enhances the ability to preserve shallow-level details and distinguish deep-level semantics by doubling the number of channels in key modules of the backbone network; introduces a cross-modal dual-branch structure to independently process modality-specific features; and applies a dual attention mechanism twice. These designs effectively resolve the domain difference issue and improve the performance of feature extraction and fusion for low-texture images. Comparative experimental results on a self-constructed dataset demonstrate that EnhancedXFeat outperforms mainstream algorithms significantly: the average number of detected feature points reaches 37.0, which is much higher than 4.3 of Scale-Invariant Feature Transform(SIFT), 6.2 of Oriented FAST and Rotated BRIEF(ORB), and 17.9 of Optical-SAR Match Net(OSMNet); the matching accuracy achieves 88.1%, remarkably superior to 9.2% of SIFT,12.7% of ORB, 68.0% of OSMNet and 68.7% of VGG-16; meanwhile, the single-operation time is controlled to 0.801 2 s, achieving a favorable balance between accuracy and efficiency. The conclusion indicates that EnhancedXFeat effectively improves the accuracy and robustness of SAR-optical image registration through its targeted design, and its lightweight architecture provides an efficient and reliable solution for multi-source remote sensing image registration applications in resource-constrained environments.

Abstract:The terahertz(THz) band(0.1~10 THz) has become a key technology direction for future wireless communications due to its large bandwidth and high-speed transmission characteristics. In this paper, a 300 GHz terahertz wireless communication system is designed and implemented based on an all-electronics scheme. Compared with photonics-aided approaches, this system offers advantages of simple structure and low complexity. Through the construction of an experimental platform, wireless transmission tests over 50 m are completed. Residual blocks are incorporated into the One-Dimension Convolutional Neural Network(1D-CNN) in the experiments, which can effectively avoid gradient vanishing and enhance signal equalization performance. The effects of transmission rate and different neural network signal recovery schemes on system performance are analyzed in detail. Meanwhile, the relationship among the number of 1D-CNN neurons, the depth of hidden layers, and transmission performance is discussed. The proposed 1D-CNN signal recovery method effectively compensates for nonlinear effects in the system, reducing the system Bit Error Rate(BER) from 4.6×10^-3 to 1.4×10^-3.

Abstract:As the cornerstone of information technology, the semiconductor industry relies on doping the surface of intrinsic semiconductor wafers as one of its core processes to fabricate high-performance optoelectronic devices and integrated circuits. To improve production efficiency and ensure device quality, there is an urgent need to develop a rapid, non-contact, and non-destructive technique for extracting doping layer thickness and doping concentration parameters. This paper proposes a method that utilizes Fabry-Perot resonance spectroscopy to calculate the doping layer thickness and doping concentration of semiconductor wafers. Based on electromagnetic wave scattering matrix theory, this method enables efficient numerical calculations of the terahertz transmission, reflection, and absorption spectra of semiconductor wafers. By constructing a correlation model between doping layer thickness, doping concentration, and Fabry-Perot resonance spectra, it provides a theoretical foundation for fitting experimental spectra and extracting doping parameters with high precision. This method is expected to offer an efficient, non-destructive detection means for quality control and process optimization in semiconductor manufacturing.

Abstract:To meet the requirements of miniaturization, low profile, polarization switching, and wide-angle scanning for satellite communication receiving terminals, a novel K-band antenna array is designed. Switchable circular polarization is synthesized in the magneto-electric dipole antenna element through dual linear polarizations. By constructing a double-layer patch, uniformly distributed metallic vias, and "П" shaped orthogonal grounded feeding structure, it attains 20% relative bandwidth with less than -22 dB isolation. The element maintains stable wide-beam radiation patterns across the entire frequency band, exhibiting an Axial Ratio(AR) below 3 dB within ±65° beam coverage. An 8×8 antenna array is constructed by integrating the antenna unit with a multi-stage sequential rotation feeding technique. Simulation results demonstrate that within a scanning range of ±60°, the K-band array achieves a gain exceeding 17.8 dBic and an AR below 3.6 dB, indicating excellent circular polarization beam scanning capability. The antenna array features a compact structure and lightweight design, making it suitable for mobile millimeter-wave satellite communication receiving terminals.

Abstract:Based on a 0.25 μm GaN High-Electron-Mobility Transistor(HEMT) process platform on SiC substrate, a compact, high-power, high-efficiency Microwave Monolithic Integrated Circuit(MMIC) Power Amplifier(PA) is designed using electro-thermal co-design and microwave device/circuit simulation techniques. Based on output power requirements and load-pull characteristics,the dimensions and optimum impedances of the transistors at each stage are determined. According to the circuit gain requirements, a three-stage cascaded configuration is adopted, with total gate width ratios of 1:3.6:16 for the pre-driver, driver, and final stages, respectively. Through electro-thermal co-design, the die layout is optimized to reduce thermal coupling effects and enhance the heat dissipation capability of the circuit. By employing electromagnetic parasitic parameter extraction technology, a compact circuit layout is achieved with a chip size of 2.7 mm×5.6 mm. Test results demonstrate that the power amplifier achieves a saturated output power greater than 46.5 dBm, a Power-Added Efficiency (PAE) greater than 42%, and a saturated power gain greater than 26.5 dB over the 14~18 GHz frequency band. Owing to its excellent electrical performance, it is suitable for wide applications in communications and electronic countermeasures.

Abstract:To address the insufficient near-field coupling efficiency in traditional Passive Intermodulation(PIM) test systems, a compact near-field coupling test board operating at 2.6 GHz has been designed and fabricated. The board employs a Substrate-Integrated Waveguide(SIW) architecture and incorporates, for the first time, an adjustable rectangular-slot structure whose dimensions can be optimized to match different materials under test, thereby markedly enhancing the near-field test intensity. While maintaining stable RF electrical performance, the board achieves a high test sensitivity of -170 dBc with a measurement deviation kept within 3 dB. Experimental results demonstrate that the compact geometry is especially suited for evaluating the PIM characteristics of common microscale conductive materials, offering excellent repeatability and stability. The proposed design provides an effective test method for material-level PIM investigations and holds significant value for PIM analysis of components and materials used in mobile-communication systems.

Abstract:To address the spectrum resource contention issue in heterogeneous wireless networks, a priority-based spectrum interference analysis and adaptive optimization method is proposed. A dual-layer heterogeneous communication system model considering spatiotemporal characteristics is constructed. An interference quantification mechanism integrating time-frequency overlap and Signal-to-Interference Ratio(SIR) evaluation is established, and a weighted interference evaluation model accounting for device priority is proposed. On this basis, a parameter-adaptive Particle Swarm Optimization(PSO) algorithm is designed to achieve dynamic optimal allocation of spectrum resources. Simulation results demonstrate that the proposed method reduces the overall system interference level by an average of 63.3% while ensuring the communication quality of high-priority devices, with the suppression effect on severe interference reaching 81.4%. Compared with existing methods, the proposed approach exhibits significant advantages in both interference suppression and computational efficiency, providing a novel technical solution for spectrum management in complex electromagnetic environments.

Abstract:Existing track association models are mostly designed for target association tasks in specific scenarios. When target maneuvering patterns become highly variable or the number of targets fluctuates significantly, these models tend to suffer from degraded association accuracy or extensive missed detections. To address the need for real-time perception of track target attributes in multi-target scenarios, this paper proposes a track association model based on multi-scenario threshold switching. The applicable scenarios and limitations of various track association algorithms are analyzed. Simulation verification is conducted to identify the advantageous performance intervals of different track association algorithms under varying scene densities and noise environments, thereby determining environmental thresholds. During practical operation, the track association model with better performance is selected and switched according to current environmental parameters, enabling more accurate and efficient target attribute classification. Experimental results demonstrate that the proposed threshold-switching track association model achieves a comprehensive Correlation Accuracy Rate(CAR) of 98.16%, maintains an association processing time within 1 s for 50 targets, and keeps the comprehensive Missed Detection Rate(MDR) below 17%.

Abstract:To mitigate the communication interruption problem caused by plasma encountered by hypersonic vehicles, a channel equalization method based on channel prediction is proposed. By investigating the correlation between reflected and transmitted signals, the relationship between transmitted and reflected signals is established; the reflected signal is utilized to predict plasma channel parameters in real time, and corresponding adaptive adjustment strategies are formulated based on the prediction results to counteract the impact of plasma parasitic modulation effects. The phase shift induced by plasma causes special rotation of the constellation diagram for phase-modulated signals, but exerts no significant influence on the main frequency range of frequency-modulated signals; moreover, frequency-modulated signals employ non-coherent demodulation, rendering the phase effect negligible. Consequently, Frequency-Modulated(FM) signals are less affected by plasma. Under the frequency modulation scheme, this paper leverages reflected signals to predict plasma channel characteristics in real time, and compensates for the attenuation caused by plasma parasitic modulation effects on transmitted signals through an adaptive power amplifier at the transmitting modulation stage. Simulation results demonstrate that, compared with non-adaptive modulation schemes, the algorithm adopting adaptive modulation to compensate for channel effects can effectively improve bit error performance.