Iet Radar Sonar and Navigation最新文献

Space-time adaptive processor (STAP) has been widely used in the GNSS antenna array anti-jamming receiver. The cost of STAP algorithms, including the minimum variance distortion-less response (MVDR) algorithm and power inversion (PI) algorithm, is introducing measurement errors. However, there is no systematic answer to the principle of error introduction, the magnitude of error and its influence on single point position (SPP) and real-time kinematic (RTK). We have conducted a systematic study on the above-mentioned issues. Firstly, the principle of error introduction was theoretically studied. Then, a large number of simulations were conducted to evaluate the magnitude of the error. Finally, simulated errors are introduced into the B1I and B3I real measurements to implement SPP and RTK to evaluate the influence of the STAP algorithms on SPP and RTK. Results show that for SPP, the influence of STAP algorithms on the B1I + B3I dual-frequency ionosphere-free combination SPP is larger than that on the B1I single-frequency SPP; for RTK, the influence of STAP algorithms on the B1I + B3I dual-frequency uncombined RTK is smaller than that on the B1I single-frequency RTK. In addition, the influences of the MVDR algorithm on SPP and RTK are smaller than those of the PI algorithm.

Multi-agent deep reinforcement learning (MARL) for self-driving vehicles aims to address the complex challenge of coordinating multiple autonomous agents in shared road environments. MARL creates a more stable system and improves vehicle performance in typical traffic scenarios compared to single-agent DRL systems. However, despite its sophisticated cooperative training, MARL remains vulnerable to unforeseen adversarial attacks. Perturbed observation states can lead one or more vehicles to make critical errors in decision-making, triggering chain reactions that often result in severe collisions and accidents. To ensure the safety and reliability of multi-agent autonomous driving systems, this paper proposes a robust constrained cooperative multi-agent reinforcement learning (R-CCMARL) algorithm for self-driving vehicles, enabling robust driving policy to handle strong and unpredictable adversarial attacks. Unlike most existing works, our R-CCMARL framework employs a universal policy for each agent, achieving a more practical, nontask-oriented driving agent for real-world applications. In this way, it enables us to integrate shared observations with Mean-Field theory to model interactions within the MARL system. A risk formulation and a risk estimation network are developed to minimise the defined long-term risks. To further enhance robustness, this risk estimator is then used to construct a constrained optimisation objective function with a regulariser to maximise long-term rewards in worst-case scenarios. Experiments conducted in the CARLA simulator in intersection scenarios demonstrate that our method remains robust against adversarial state perturbations while maintaining high performance, both with and without attacks.

The coordinated deployment of multi-sensor systems significantly enhances group target detection capabilities, yet persistent tracking remains challenging due to inherent limitations in single-sensor field of view (FoV) coverage. This paper proposes a novel labelled multi-Bernoulli (LMB) filter for resolvable group target (RGT) tracking under the centralised fusion (CF) framework, abbreviated as the CF-LMB-RGT filter. The proposed method introduces the virtual leader kinematic model to capture intra-group motion constraints and incorporates group structure undirected graph into the LMB recursion for interaction prediction. A key innovation lies in the Kullback–Leibler divergence minimised fusion rule that optimally integrates local posteriors within joint FoV regions while explicitly modelling common FoV overlaps, enabling complementary information fusion across nonoverlapping sensor FoVs. Simulation results demonstrate that our method achieves impressive tracking accuracy for RGTs by integrating information from all sensors.

The fast simultaneous localisation and mapping (FastSLAM), utilising the Rao-Blackwellised particle filter, provides a robust navigation solution in urban environments. Ensuring the integrity of FastSLAM is critical for the safety of autonomous driving applications. Our previous work proposed an empirical integrity risk evaluation method for nominal conditions and a probabilistic bound using PAC (probably approximately correct)–Bayesian theory. However, it was limited by overly conservative risk estimates and a lack of consideration for fault conditions. This study introduces a refined integrity evaluation framework with three main contributions. First, a modified weighting and resampling technique is proposed to reduce conservatism in empirical risk without compromising estimation accuracy. Second, a fault monitoring method is introduced to detect and isolate control input faults during the dynamic update step. Third, a conservative integrity risk evaluation approach is developed for FastSLAM to account for data association faults using probabilistic modelling. Simulation results show that the proposed methods significantly improve integrity performance under both nominal and faulted scenarios.

Transionospheric SAR autofocus is a variational algorithm designed to circumvent the deficiencies of conventional autofocus techniques in correcting the distortions of spaceborne SAR images due to ionospheric turbulence. It has demonstrated superior performance in a variety of computer-simulated imaging scenarios. In the current work, we conduct a systematic statistical analysis of transionospheric SAR autofocus aimed at corroborating its robustness and identifying limitations and sensitivities across a broad range of factors that affect the autofocus performance. We employ the range-compressed domain representation where the target reflectivity, antenna signal, and the phase screen depend only on the azimuthal coordinate. The three main factors included in the study are the levels of turbulent perturbations, clutter, and noise. We use the normalised cross correlation (NCC), integrated sidelobe ratio (ISLR), and peak desynchronisation (PD) as a-posteriori performance metrics. A key objective of the current analysis, beyond assessing the autofocus performance, is to identify the directions of how to further improve the algorithm, in terms of both the quality of focusing and associated computational cost.

In response to the complex scenario where multiple jammers navigate through a netted radar system (NRS), this study presents an optimised allocation algorithm for cooperative jamming resources, namely the Multi-Agent Jamming Resource Allocation (MJCJRA) algorithm, which is based on multi-agent deep reinforcement learning. Initially, the research develops a target fusion detection probability function and a global performance index optimisation function, which are tailored to the specific jamming and radar detection models of the scenario. Subsequently, the multiple jammers are mapped into a multi-agent system with a greedy strategy employed to generate targeted rewards for the jamming agents, enhancing their learning efficiency and performance. The study culminates in the design of evaluation and mixed-strategy networks for the jamming agents. It utilises an exponential mean shift method for soft updates of the target network, adopts priority experience replay and importance sampling methods, and incorporates reward centring into the loss function for network updates. Experimental findings demonstrate that MJCJRA algorithm significantly surpasses the baseline method, the particle swarm optimisation (PSO), the snow ablation optimiser (SAO), the multi-agent deep deterministic policy gradient (MADDPG) and multi-agent proximal policy optimisation (MAPPO), effectively diminishing the detection capability of NRS.

In the experimental test of the radar system, it is extremely important to build a realistic experimental environment for electromagnetic target testing, which is often realised by the radar target simulation technology. Corner reflectors often simulate radar RCS features by the spatial arrangement; however, their electromagnetic characteristics are solidified and the RCS features differ from those of real targets. This paper proposes a target RCS simulation method based on AFSS echo power modulation. The core idea is to use AFSS reflection modulation to dynamically regulate the target power information to achieve flexible and fast control of the target radar RCS characteristics. Based on the AFSS echo power modulation model, the theoretical relationship between the modulation parameters and the RCS value is deduced, and the duty cycle of the scattering state control signal is used as an adjustable variable to realise the simulation of the dynamic RCS sequence of the mid-range target. The RCS simulation experiment is carried out based on the target measured data, and the simulation effect is analysed in terms of statistical characteristics and similarity coefficients. The simulation results show that the statistical characteristics of the simulated RCS sequence and the target RCS sequence are very close to each other with the mean value and standard deviation within 1 dBsm and the extreme value and extreme deviation within 3 dBsm. The method is of great significance in the field of radar system tests and electronic protection.

Low-correlation sidelobes are critical for spectrally compatible waveforms in multiple-input multiple-output (MIMO) radar systems. This study presents a novel algorithm for designing spectrally compatible waveforms for MIMO radar with low auto- and cross-correlation sidelobes to enhance weak target detection capability. We adopt the minimum auto- and cross-correlation-weighted integrated sidelobe level (ACWISL) as the objective function. Under spectral and constant modulus constraints, we formulate a nondeterministic polynomial time (NP)-hard problem. To solve this problem, we combine the block successive upper-bound minimisation (BSUM) and majorisation-minimisation (MM) algorithms to develop the BSUM-MM algorithm. The original problem is decomposed into several independent subproblems, which are iteratively solved using the MM algorithm. We also employ the fast Fourier transform (FFT) to significantly accelerate the calculation. Simulation results demonstrate that the proposed algorithm is superior in terms of computational efficiency and sidelobe performance.

The existing radar emitter modulation recognition methods typically assume that the data distribution across different types is balanced. But in reality, the number of signals of various kinds often follows a long-tail distribution, leading to model overfitting for the head classes and underfitting for the tail classes. As a result, the overall recognition performance of models under such data imbalances is suboptimal. A long-tail distribution automatic modulation recognition method based on decoupled training is proposed to address this issue. Based on the ResNeXt network, the proposed method decouples the model training process into two stages: a feature extraction phase under the imbalanced dataset and the classifier learning stage under a balanced dataset. The classifier boundary is fine-tuned by $��\tau �$-normalization method. Compared to existing radar emitter modulation recognition frameworks, the proposed method achieves an overall recognition accuracy of 86.8% when the data imbalance factor is 0.01, surpassing the baseline model by 5%, and improves the performance of radar emitters modulation recognition in the real environment.