Iet Radar Sonar and Navigation最新文献

Automatic modulation recognition (AMR) of radar signals plays a pivotal role in intelligent spectrum management for modern electronic warfare systems. Although deep learning has improved AMR accuracy, existing approaches encounter challenges when identifying signal types beyond those seen during training. To overcome this limitation, this work proposes a boundary-aware learning (BAL) framework with three innovations. First, an adaptive pyramid network automatically highlights important time-frequency information at multiple dimensions, producing richer and more discriminative signal representations. Second, a boundary-aware learning strategy shapes the embedding space so that samples of the same modulation are drawn tightly together, whereas different modulations are pushed farther apart. Third, a distance-based rejection mechanism measures how far a new signal lies from known clusters, enabling reliable detection of previously unseen modulations. Together, these components create a unified feature space that both sharpens class boundaries and isolates unknown signal types. Extensive experiments on both simulated and measured radar datasets show that the proposed framework outperforms conventional architectures in open-set signal recognition, confirming its superior robustness in realistic electromagnetic environments.

The track association problem aims to determine which tracks belong to the same real target, supporting subsequent processes such as target state estimation, target tracking and situation assessment. Existing algorithms are primarily based on statistical mathematics, fuzzy mathematics, grey system theory and neural networks. However, they suffer from several limitations, including overly idealised modelling, manually set thresholds, two-level traversal and poor association performance under few-shot conditions. In the light of the above problems, we propose a K-Point Input LSTM-KNN-DPC (KLKD) algorithm. Firstly, considering the rationality of feature selection, we introduce a similarity measurement method with an adaptive cutoff distance. Secondly, a method for cluster centre selection is presented. Finally, an association assignment strategy is provided. The proposed algorithm eliminates the need for timestamp alignment and exhaustive pairwise matching across different track sequences, thereby improving the efficiency of track association. Experimental results demonstrate that, in both typical manoeuvring and irregular manoeuvring scenarios, the KLKD algorithm achieves higher association accuracy and lower end-to-end latency compared to existing methods.

Given rising UAV security concerns, this work addresses a critical counter-UAV limitation: the inability to rapidly discern UAV versus non-UAV targets (birds and balloons), causing response delays and false alarms. We establish this binary classification as a pivotal security decision node, directly activating response protocols to enhance system timeliness, reliability and efficacy. We propose an interpretable LightGBM framework integrating fused radar features (motion, RCS and track). Hyperparameters are optimised via grid search with performance validated through 5-fold cross-validation. SHAP values quantify feature contributions. Validation experiments show that the proposed framework achieves 92.55% overall accuracy in UAV and non-UAV classification based on low-altitude radar data, outperforming all comparative models including SVM, RF, BPNN, FT-Transformer, TabNet and LSTM, in both accuracy and inference speed. The SHAP-based interpretable framework simultaneously ensures high classification accuracy and reliable decision validation, thereby providing dual technical assurance in accuracy and interpretability for low-altitude security system deployment.

To address the limitations of existing UAV-mounted LiDAR systems, this paper proposes an innovative real-time transmission and rendering technology for massive point cloud data. The technology utilises an onboard LiDAR placed on the front of the UAV to collect 3D spatial point cloud data, which is transmitted to an onboard Intel NUC for critical processing steps such as data analysis and lossy compression. The compressed data are then transmitted via the UAV's video transmission link to a mobile control app, enabling real-time transmission and rendering of massive point cloud data on Android terminals. This approach effectively overcomes the portability drawbacks of traditional methods that rely on bulky computers, requiring only an Android mobile terminal. A well-designed lossy compression strategy significantly improves data transmission efficiency and reduces computational pressure for point cloud rendering on low-memory mobile devices. Integrated with the SLAM (simultaneous localisation and mapping) algorithm on a flight test platform composed of a DJI M350 RTK UAV, Velodyne VLP16 LiDAR, Intel NUC onboard computer and DJI RC Plus Android controller, the system achieves high-precision 3D point cloud real-time transmission and rendering, enabling real-time Beyond Visual Line of Sight (BVLOS) UAV control. Experimental results demonstrate that this technology can process millions of point cloud data per second on the UAV mobile controller, exhibiting excellent real-time data transmission and rendering performance under various environmental conditions, including low latency and high frame rates, meeting stringent requirements for high precision and real-time responsiveness.

Geosynchronous synthetic aperture radar (GEO SAR) provides extensive beam coverage and strong continuous observation capabilities, making it a research focus in the remote sensing. Based on the analysis of the GEO SAR illuminated scene characteristics, it is found that the nonplanar and undulating elevation of the ground surface leads to significant azimuth and range spatial variance in the echo signals. To attain precise, geometrically undistorted fast back-projection (BP) SAR images, we analyse the impact of elevation errors on the signal's quadratic phase. Then, a fast back-projection imaging method for GEO SAR geometric distortion calibration based on dynamic selection of local sub-aperture images using a coarse digital elevation model (DEM) is proposed. Firstly, the elevation information of the imaging grid is obtained through coarse DEM interpolation. Secondly, the coordinates of the dynamically selected local sub-aperture images can be approximated by a fully expanded quadratic coordinate polynomial. And the compression function of the two-stage spectrum compression method is modified by utilising this polynomial. Ultimately, the full-aperture SAR image is produced by conducting multi-level fusion operations and mosaic techniques. Additionally, a detailed flowchart is provided. The efficacy of the suggested approach is verified through simulated echo data.

Global navigation satellite system (GNSS) is widely recognised to be vulnerable to spoofing attacks. A sophisticated form of GNSS spoofing, termed distributed spoofing, transmits each spoofing signal through dedicated antennas. This technique poses significant implementation difficulty in practical scenarios, owing to challenges including diverse propagation paths, inter-node clock synchronisation and transmit-receive isolation. This article proposes a feedback node aided distributed spoofing (FNA-DS) system for executing GNSS time synchronisation attacks, enabling flexible implementation of distributed spoofing. Leveraging observations from a feedback node, the time biases and drifts of each spoofing node are estimated in real time and compensated during spoofing signal generation, ensuring false position, velocity and timing (PVT) solution accuracy and high pseudorange consistency. By dedicating the feedback node exclusively to signal reception and the spoofing nodes solely to signal transmission, the requirement for transmit-receive isolation is relaxed. To comprehensively characterise the distributed spoofing threat, a detailed performance analysis of the FNA-DS system is conducted, quantifying the impact of node position errors and time parameter estimation errors. Field experiments using a self-developed distributed spoofing prototype validate the FNA-DS system's effectiveness and expose limitations in existing direction of arrival (DoA) based anti-spoofing techniques. Collectively, this work expands the capability frontier of GNSS spoofing, advances understanding of distributed spoofing and underscores its significance as a practical GNSS security threat.

This paper presents a comprehensive survey of Direction of Arrival (DoA) estimation techniques that utilise one-bit and low-resolution data. We delve into various approaches, including direct application of quantised data to existing DoA estimators and reconstruction-based methods, such as covariance matrix reconstruction via the arcsine law and recovery of noiseless unquantised measurements. Low-resolution quantisation is increasingly essential in modern communication systems, especially massive MIMO systems, due to its benefits in terms of power consumption, cost and system complexity. One-bit quantisation, in particular, has gained significant attention in wireless communication and cellular and sensor networks. We conduct a thorough evaluation of different methods and algorithms under various scenarios to identify optimal techniques for different conditions. Our analysis includes comparisons of different performance metrics and computational complexity. We also investigate the effect of increasing the number of quantiser output levels on DoA estimation performance. Our findings demonstrate that the Lloyd-Max quantiser consistently outperforms the maximum entropy quantiser for a higher number of quantisation levels. Additionally, we compare the performance of direct use of quantised data with quantised measurement recovery approach at higher quantisation levels. Our results suggest that direct use of quantised data is generally a more efficient and effective approach in such scenarios.

Passive reconnaissance solutions receive increased interest as unjammable fibre-optic drones represent a large number of UAVs in recent military conflicts. In order to equip critical infrastructure with an early warning system against drone attacks, it seems obvious to use the local communication infrastructure as illuminator for a passive radar. In Germany and other European countries, a blackout resistant LTE network in the 450 MHz band for critical infrastructure sites is currently rolled outed or already planned. In this contribution, we provide a proof of concept and present experimental results with a LTE450-based single- and multichannel passive radar for drone detection. To ease the signal processing while achieving a clearer ambiguity-function, only the reference elements contained in the OFDM symbols are used. For removing the dominant direct path contribution from the illuminator, an ad hoc approach is used which exploits the space-time structure of the received OFDM reference elements.

In electronic warfare (EW) systems, accurate time-of-arrival (TOA) prediction for radar signals is critical for effective jamming. TOA depends on both pulse repetition interval (PRI) and radar scan patterns, which are increasingly complex due to technological advancements. Unlike prior research focusing solely on one factor, this paper proposes a machine-learning model that leverages both PRI and scan patterns to predict subsequent radar pulse TOA. The system demonstrates superior prediction accuracy and robust performance in noisy environments and under varying probabilities of detection (POD). This is achieved by separating the PRI sequence and the radar scan interval, an approach that can be applied to different system designs. The proposed method applies a filtering algorithm that separates PRI and scan sequences, feeding them into distinct LSTM models, with a splitting technique addressing missing pulses. Importantly, the model integrates the radar antenna main lobe and side lobe information to enhance jamming effectiveness. Simulation results also demonstrate that the main design concept—considering both PRI and scan type—can be used for different techniques, such as a decision tree. This approach significantly improves TOA estimation, handles diverse radar patterns and represents a valuable contribution to radar technology for improved situational awareness and operational efficiency.

This study addresses positioning errors in high-frequency (HF) over-the-horizon (OTH) localisation that arise from inaccuracies in ionospheric virtual height measurements. We propose a direct localisation algorithm based on the direct position determination (DPD) model in which the initial search range of the target is estimated using HF single-station direction-finding (SSDF). To enhance accuracy, the International Reference Ionosphere (IRI) model is combined with ionosonde data to provide priors on ionospheric virtual heights. These priors are incorporated into a single-layer mirror reflection model of the ionosphere to establish a more accurate signal propagation path, thereby mitigating errors caused by variations in virtual heights across different transmission paths. The algorithm leverages the global search capability of particle swarm optimisation (PSO) to generate high-quality initial solutions, followed by localised refinement through the Gauss–Newton method to further improve positioning accuracy. Experimental results show that, compared with traditional direct localisation methods that assume fixed virtual heights, the proposed approach reduces positioning errors by 5–25 km in typical scenarios and increases computational efficiency by more than 40% compared to the conventional exhaustive grid search method (measured in terms of computational complexity). Overall, the method provides a balanced solution for HF OTH localisation systems, effectively improving both accuracy and efficiency.