Inverse Synthetic Aperture Radar (ISAR) has been widely applied in remote sensing and space target monitoring. Automatic Target Recognition (ATR) based on ISAR imagery plays a critical role in target interpretation and pose estimation. With the growing adoption of intelligent methods in the ATR domain, the quantity and quality of ISAR data have become decisive factors influencing algorithm performance. However, due to the complexity of ISAR imaging algorithms and the high cost of data acquisition, high-quality ISAR image datasets remain extremely scarce. As a result, learning the underlying characteristics of existing ISAR data to generate large-scale usable samples has become a pressing research focus. Although some preliminary studies have explored ISAR image data augmentation, most of them rely on image sequence interpolation or conditional generation, both of which exhibit critical limitations: the former requires densely sampled image sequences with small angular intervals, while the latter can only model the mapping between limited azimuth conditions and ISAR images. Neither approach is capable of generating images of new targets under unseen azimuth conditions, resulting in poor generalization and leaving substantial room for further exploration. To address these limitations, we formally define a novel research problem, termed ISAR azimuth angle extrapolation. This task fundamentally involves high-dimensional, structured, cross-view image synthesis, requiring the restoration of visual details while ensuring physical consistency and structural stability. To address this problem, we propose ISAR-ExtraNet, a foundation-model-based framework for ISAR azimuth angle extrapolation. ISAR-ExtraNet leverages the strong representation, modeling, and generalization capabilities of pretrained foundation models to generate ISAR images of new targets under novel azimuth conditions. Specifically, the model employs a two-stage coarse-to-fine fine-tuning strategy, incorporating optical image contours and scattering center distribution constraints to guide the generation process. This design enhances both semantic alignment and structural fidelity in the generated ISAR images. Comprehensive experiments demonstrate that ISAR-ExtraNet significantly outperforms baseline methods and fine-tuned foundation models, achieving 28.76 dB in PSNR and 0.80 in SSIM. We hope that the training paradigm introduced in ISAR-ExtraNet will inspire further exploration of the ISAR azimuth extrapolation problem and foster progress in this emerging research area.

Synthetic aperture radar (SAR) automatic target recognition (ATR) is crucial for achieving efficient information sensing. With the development of real-time imaging, there is an increasing demand to perform both imaging and ATR on edge platforms that are constrained in computing power, memory, and energy consumption, which makes it challenging to deploy CNN- or transformer-based models with large parameter counts on such platforms. Owing to their structural flexibility and compact parameterization, graph neural networks (GNNs) offer a promising solution for lightweight SAR image target recognition. To this end, this article proposes a GNN with collaborative attention and position embedding (GRAPE). First, we design a collaborative attention layer (CAL) to enhance the collaborative interaction between the features of a central node and those of its neighboring nodes. Second, we introduce a relative positional embedding layer, which explicitly incorporates relative positional information between nodes to further strengthen the spatial modeling capability of the SAR graph. Experimental results demonstrate that GRAPE achieves substantially higher recognition accuracy than existing lightweight CNN and transformer models. On three publicly available datasets, SAR-AIRcraft-1.0, OpenSARShip, and MSTAR, GRAPE yields improvements of 1.89%, 7.86%, and 0.73%, respectively, while attaining the best performance in terms of parameter size and inference-time resource consumption. Compared with mainstream GNN-based methods, GRAPE further improves recognition accuracy by 7.23%, 6.09%, and 6.72%. In summary, the proposed GRAPE framework markedly enhances SAR image target recognition performance under low computational and memory budgets, providing an effective solution for building high-performance and easily deployable SAR ATR systems.

To address the challenges of high hardware costs and structural complexity in traditional antenna array-based Direction of Arrival (DOA) estimation systems, this paper proposes an orthogonal inversion DOA estimation method based on a space-time coding metasurface (STCM). Using an orthogonal coding scheme, the proposed method performs secondary modulation on the received signals. By employing an efficient regeneration algorithm, the incident wave signals at each antenna element are retrieved using only a single receiving antenna, thereby achieving high-precision angular estimation in conjunction with the MUSIC algorithm. Numerical simulation results demonstrate that the proposed method significantly reduces system complexity and hardware costs while maintaining high estimation accuracy, thereby validating its feasibility and robustness.

Interrupted sampling repeater jamming (ISRJ) has attracted extensive attention in radar electronic countermeasures due to its low time delay and strong coherence. However, the jamming effectiveness of ISRJ is closely related to waveform characteristics, often failing to produce the desired multi-false target deception effect. To address this issue, an extended ISRJ (EISRJ) method focusing on the construction of a frequency-domain periodic sampling pattern is proposed, which can realize deception jamming effects, similar to ISRJ on linear frequency modulation (LFM) waveforms, for various non-linear frequency modulation (NLFM) waveforms by simply using a non-periodic sampling function. Based on the principle of stationary phase, the intrinsic mechanism behind why ISRJ achieves multi-false-target deception effects on LFM waveforms but fails on NLFM waveforms is revealed from the frequency-domain perspective. Building upon this insight, the EISRJ theory, based on waveform time-frequency modulation correspondence and the design of extended interrupted sampling functions, is established. Implementation analysis indicates that electromagnetic metamaterials have advantages over traditional digital radio frequency memory-based jammers in achieving non periodic interrupted sampling and retransmission, without the need to address transceiver antenna isolation issues. Simulation experiments across multiple parameters and NLFM waveforms further validate the correctness and superiority of the EISRJ.

Traditional reinforcement learning (RL) methods typically assume complete observation data, but real-world scenarios are often far more complex. Real-world deployments face missing data where part of the state dimensions become unavailable due to sensor failures, communication blackouts, or temporal sampling discontinuities. While recent research has achieved significant progress in handling missing data scenarios, current methodologies remain fundamentally constrained by their inability to effectively manage prolonged observation gaps and their low tolerance thresholds for high missing rates. In this work, we investigate the performance of reinforcement learning under the condition of missing data. We introduce a novel method, Mutual Information Aligned Generative Reinforcement Learning(MIA-GRL), which employs a spatiotemporal collaborative reconstruction to learn from historical reinforcement learning trajectories. This method synthesizes trajectories that encapsulate environmental characteristics and diversity through contextual information. We design an auxiliary loss function based on mutual information maximization, aiming to maximize the mutual information between the completed data and the original data, thereby ensuring that the completed data retains as much critical information from the original data as possible. Additionally, we utilize missing-aware contrastive learning to learn representations robust to missing patterns, enabling the policy network to capture intrinsic features relevant to the task. Experiments show our method outperforms state-of-the-art methods under the missing data conditions.

This paper proposes a direction-of-arrival (DOA) estimation method for coherent signals using an asynchronous space-time-coding metasurface (ASTCM). Different from traditional array processing, the metasurface modulates incident waves via space-time-coding, generating multiple harmonic components in the received spectrum. Under asynchronous modulation, low-order harmonics are decomposed into multiple same-order harmonics, which suppresses noise. The amplitude and phase of these harmonics carry the angle information of incident signals, and their number matches the number of asynchronous modulation frequencies, facilitating angle recovery. By selecting suitable harmonics from the spectrum of coherent signals, the iterative adaptive approach (IAA) is applied to achieve high-precision DOA estimation. Simulation results verify the effectiveness of the proposed method under different signal-to-noise ratio (SNR). Even at low SNR, the DOA estimation error for coherent signals is still below 0.2°.

This paper proposes a Direction of Arrival (DOA) estimation method based on Sparse Bayesian Learning (SBL) using an asynchronous space-time-coding metasurface. The Space-Time-Coding Metasurface (STMC) technology enables singlechannel DOA estimation, offering significant advantages in cost and energy consumption compared to phased array antennas. However, in traditional metasurfaces, different units are modulated at the same frequency, which fails to fully utilize spectral information and consequently limits performance. To address this issue, this study adopts an asynchronous space-time-coding metasurface and integrates the SBL method for DOA estimation, thereby reducing system complexity and hardware costs while improving convergence speed, direction-finding accuracy, and anti-interference capability. Simulation experiments demonstrate that the proposed method achieves higher estimation accuracy and convergence efficiency compared to other algorithms, while also exhibiting stronger robustness.

This paper addresses the performance degradation problem in traditional Direction of Arrival (DOA) estimation caused by off-grid errors, and proposes an off-grid DOA estimation method based on Root Sparse Bayesian Learning(Root-SBL) for asynchronous space-time-coding metasurface(STCM). The method constructs an asynchronous modulated spatiotemporal coding metasurface system, achieving a hardware-simplified architecture with single-channel reception. In terms of signal processing, an off-grid sparse reconstruction model is established, where the grid offset parameters in the continuous angular space are integrated into the Bayesian inference framework as parameters to be estimated. Through hierarchical prior modeling and variational inference, adaptive joint optimization of signal sparsity, noise variance, and grid offsets is achieved. The proposed method is capable of automatically correcting grid mismatch errors and maintains robust performance even in complex scenarios.

In strategic decision-making tasks, determining how to assign limited costly resource towards the defender and the attacker is a central problem. However, it is hard for pre-allocated resource assignment to adapt to dynamic fighting scenarios, and exists situations where the scenario and rule of the Colonel Blotto (CB) game are too restrictive in real world. To address these issues, a support stage is added as supplementary for pre-allocated results, in which a novel two-stage competitive resource assignment problem is formulated based on CB game and stochastic Lanchester equation (SLE). Further, the force attrition in these two stages is formulated as a stochastic progress to consider the complex fighting progress, including the case that the player with fewer resources defeats the player with more resources and wins the battlefield. For solving this two-stage resource assignment problem, nested solving and noregret learning are proposed to search the optimal resource assignment strategies. Numerical experiments are taken to analyze the effectiveness of the proposed model and study the assignment strategies in various cases.

Multichannel coherent RF signal generators (CSGs) are widely used in microwave system testing, such as phased arrays and MIMO systems. In many commercial CSGs (e.g., AnaPico), the relative phases among channels may become random after power-up or frequency switching due to the use of independent phase-locked loops (PLLs). Consequently, amplitude–phase calibration is required to ensure channel coherence. Existing multichannel measurement methods usually rely on multiple receivers, multiple RF channels, or repeated measurements, which increase system complexity and cost. To address this issue, this article proposes a low-cost parallel measurement method for the amplitude and phase relationships of multichannel coherent RF signals using a single RF receiver. In the proposed approach, a field-programmable gate array (FPGA) controls a single-pole $N$ -throw (SPNT) RF switch to periodically connect multiple RF channels to a single receiver. The resulting time modulation generates harmonic components whose amplitudes and phases are determined by the modulation sequence and the amplitude–phase characteristics of the input signals. By analyzing these harmonics, the amplitude ratios and phase differences among channels can be estimated simultaneously. An experimental system using a commercial four-channel CSG and an SP4T RF switch is implemented to validate the proposed method. Compared with reference measurements obtained from a vector network analyzer (VNA), the proposed method achieves a maximum absolute error (MAE) of 0.53 dB and a root-mean-square error (RMSE) of 0.38 dB in amplitude ratio measurement, and an MAE of 1.88° and an RMSE of 1.05° in phase-difference measurement, demonstrating its effectiveness andaccuracy.

High Resolution Range Profile (HRRP) is crucial for Radar Automatic Target Recognition (RATR), but its sensitivity to target aspect angle significantly degrades performance when observational data is incomplete across angles. Traditional recognition methods relying on manually designed features or standard deep learning struggle with this aspect variation, limiting generalization. To address this challenge, this paper proposes SupConResNet, a deep recognition method leveraging Supervised Contrastive Learning (SCL). SupConResNet integrates a 1D ResNet architecture tailored for HRRP signals with a dual-objective loss function utilizing supervised contrastive learning mechanism. By explicitly using label information, the SCL component encourages the learning of feature representations that are not only discriminative between different target classes but also robust to intra-class variations caused by differing aspect angles. We conduct systematic experiments on an electromagnetic simulation dataset, specifically evaluating performance under scenarios simulating incomplete aspect angle coverage, including both cross-angle generalization and coarse angular sampling. The results demonstrate that SupConResNet effectively maintains high recognition accuracy even with significant angular gaps between training and testing data, significantly outperforming implicitly baseline approaches reliant solely on classification loss.

The radar jamming technology based on metasurfaces has attracted widespread attention in recent years due to its advantages of flexible modulation and low power consumption. However, the deception effect of the radar target characteristics imitation technique is greatly reduced by the limited modulation states of the digital coding metasurface. To address this issue, this article proposes a novel time-domain equivalent amplitude and phase (EAP) generation method that can significantly increase the number of time modulation states of the metasurface without increasing hardware complexity. The time-domain EAP generation model based on the spatial vector synthesis is established first, and theoretical analysis demonstrates that the number of modulation states of 1- and 2-bit phase-coding metasurfaces under vertical incidence can be expanded from 2 to (the number of units +1), and from 4 to the square of (the number of units +1), respectively. Based on the time-domain EAP, a more efficient and accurate metasurface-based target high-resolution range profile (HRRP) imitation scheme compared with the state of the art is proposed, and its superiority is verified through simulations. Experiment results based on a 1-bit phase-coding metasurface further validate the practicality of both the EAP generation method and the EAP-based target HRRP imitation scheme.

Cross-eye jamming is one of the most effective countermeasures against monopulse radar, demanding precise cooperative control of jamming signal direction, amplitude and phase. In this paper, we propose a novel two-element retrodirective cross-eye jamming scheme based on digital coding metasurface and derive the corresponding metasurface retro-reflector model. By employing metasurface-based retrodirective reflectors, directional, amplitude, and phase modulation of the jamming signals can be achieved. Compared to conventional techniques, the proposed method offers significant advantages in both cost-effectiveness and control flexibility. Simulation results confirm the effectiveness of the proposed method.

To address the issues of large volume and high hardware costs in traditional antenna arrays when estimating the direction of arrival (DOA), we innovatively utilize programmable metasurfaces to replace conventional array architecture and achieve dynamic control of spatial electromagnetic waves through their reconfigurable electromagnetic properties. To tackle the significant decline in estimation accuracy of the traditional phase interferometer method under low signal-to-noise ratio (SNR) conditions, we adopt the method of space-time coding modulation (STCM) and directly solve the DOA estimation value by using the analytical relationship between the steering vector and the modulation matrix. Furthermore, to significantly improve the estimation accuracy, we strategically nonuniformly position the metasurface units to augment the array aperture. The simulation experiments have validated the robustness and superiority of the NSTCM method in low SNR environments.

Offline reinforcement learning suffers from critical vulnerability to data corruption from sensor noise or adversarial attacks. Recent research has achieved a lot by downweighting corrupted samples and fixing the corrupted data, while data corruption induces feature entanglement that undermines policy robustness. Existing methods fail to identify causal features behind performance degradation caused by corruption. To analyze causal relationships in corrupted data, we propose a method, Robust Causal Feature Disentanglement(RCFD). Our method introduces a learnable causal feature disentanglement mechanism specifically designed for reinforcement learning scenarios, integrating the CausalVAE framework to disentangle causal features governing environmental dynamics from corruption-sensitive non-causal features. Theoretically, this disentanglement confers a robustness advantage under data corruption conditions. Concurrently, causality-preserving perturbation training injects Gaussian noise solely into non-causal features to generate counterfactual samples and is enhanced by dual-path feature alignment and contrastive learning for representation invariance. A dynamic graph diagnostic module further employs graph convolutional attention networks to model spatiotemporal relationships and identify corrupted edges through structural consistency analysis, enabling precise data repair. The results exhibit highly robust performance across D4rl benchmarks under diverse data corruption conditions. This confirms that causal feature invariance helps bridge distributional gaps, promoting reliable deployment in complex real-world settings.

This paper proposes a novel method for Direction of Arrival (DOA) estimation using a deep unfolded LISTA network in a non-uniform metasurface. Traditional DOA estimation methods often face challenges such as limited accuracy, high computational complexity, and poor adaptability to complex signal environments. To address these issues, we optimize a non-uniform metasurface array to reduce hardware costs and mutual coupling effects while enhancing resolution. Additionally, a deep unfolded Learned Iterative Shrinkage Thresholding Algorithm (LISTA) network is constructed by transforming Iterative Shrinkage Thresholding Algorithm (ISTA) iterative steps into trainable neural network layers, combining model-driven logic with data-driven parameter optimization. Simulation results prove that this method enhances higher precision and reduces computational complexity in comparison with traditional algorithms, especially under low SNR conditions. Furthermore, the method exhibits greater generalization ability, making it a reliable approach for high-precision DOA estimation in practical applications.

A novel method for the direction of arrival (DOA) estimation of coherent signals under a space–time–coding metasurface (STCM) is proposed in this paper. Noticeably, the STCM can replace multi-channel arrays with a single channel, which can be utilized to modulate incident electromagnetic waves and generate harmonics. However, coherent signals are overlapping in the frequency spectrum and cannot achieve DOA estimation through subspace methods. Therefore, the proposed method transforms the angle information in the time domain into amplitude and phase information at harmonics in the frequency domain by modulating incident coherent signals using the STCM and performing a fast Fourier transform (FFT) on these signals. Based on the harmonics in the frequency spectrum of the coherent signals, appropriate harmonics are selected. Finally, the ℓ1 norm singular value decomposition (ℓ1-SVD) algorithm is utilized for achieving high-precision DOA estimation. Simulation experiments are conducted to show the performance of the proposed method under the condition of different incident angles, harmonic numbers, signal-to-noise ratios (SNRs), etc. Compared to the traditional algorithms, the performance of the proposed algorithm can achieve more accurate DOA estimation under a low SNR.

This paper proposes a wideband signal direction finding method based on Sparse Bayesian Learning (SBL) with a non-uniform array arrangement of metasurfaces. The Space-Time-Coding Metasurface (STMC) enables single-channel Direction of Arrival (DOA) estimation, offering significant advantages over phased array antennas in terms of cost and energy consumption. However, the uniform arrangement of traditional metasurface units may result in the inability to install two units within a half-wavelength range at the operating frequency. To address this issue, this study adopts a non-uniform design approach for the STMC unit array and integrates the SBL method for DOA estimation of Linear Frequency Modulation (LFM) signals, thereby reducing system complexity and hardware costs while improving convergence speed, direction finding accuracy, and anti-interference capability. Simulation experiments demonstrate that the proposed method achieves higher estimation accuracy and convergence efficiency compared to other algorithms, while also exhibiting stronger robustness.

Monopulse radar angle measurement technology is crucial for modern missile precision guidance systems due to its high accuracy and real-time capabilities. Cross-eye jamming (CEJ) is recognized as one of the most effective countermeasures against monopulse radar. However, traditional CEJ implementation requires complex amplitude and phase modulation through specialized hardware. The digital coding metasurface (DCM) is an advanced electromagnetic (EM) wave modulation technology. Synchronized modulation of phase and amplitude can be achieved by dynamically adjusting the digital coding state of each individual unit of the DCM. In this paper, a cross-eye jamming method based on DCM is proposed. The EM modulation model of DCM is derived, and it is theoretically demonstrated that cross-eye jamming signals can be generated by two DCMs. The simulation results demonstrate that the method attains more than twice the cross-eye gain when the error of the DCM phase modulation is maintained within 40°.

We propose a novel deep unfolded network approach designed to handle DOA estimation difficulties, particularly under conditions involving mutual coupling. At the outset, we convert the array output, where an unspecified mutual coupling coefficient (MCC) is involved, into the real domain of the neural network. The deep spread network is then trained to locate the MCC and acquire the DOA spatial spectrum. We illustrate that a deeply unfolded network can achieve superior generalization without the need for training tags or extensive training datasets. Simulation results validate the efficacy of our deep expansion network compared to existing methods.