Abstracts:The large elliptical orbit aerospace vehicle has complex motion characteristics and harsh flight environment. When tracking its trajectory, it is prone to the loss of measurement data caused by the interference of the detector. The traditional tracking algorithm cannot accurately describe its maneuvering characteristics, and the tracking accuracy and robustness are poor. In order to solve this problem, this article proposes a trajectory tracking method for large elliptical orbit aerospace vehicles based on a long short-term memory (LSTM) network. Aiming at the situation in which the acceleration and turning rate of the large elliptical orbit aerospace vehicle are not constant, tangential acceleration and 3-D turning rate are modeled, and a trajectory tracking model with 3-D variable turning rate is proposed. Aiming at the problem of intermittent measurement, a continuous tracking algorithm based on LSTM is proposed, which uses the memory characteristics of LSTM to predict the measurement data reasonably. Finally, simulation verification is carried out in the equatorial plane orbit and the polar plane orbit. The results show that the trajectory tracking algorithm of the large elliptical aerospace vehicle based on LSTM realizes the accurate tracking of the large elliptical orbit of the aerospace vehicle under maneuvering conditions. In the case of intermittent measurement, the LSTM network is used to predict the measurement data online, which can achieve continuous trajectory tracking. Compared to the traditional algorithm, the tracking accuracy and system robustness have been significantly improved.

Abstracts:With the continuous development of deep learning in the field of synthetic aperture radar (SAR) image processing, it is found that image recognition is vulnerable to interference and the accuracy is greatly reduced. Therefore, if a general adversarial attack method is designed to generate adversarial examples capable of deceiving different types of convolutional neural network models, it can protect target images from being correctly recognized by adversaries and safeguard image privacy information. However, most of the current adversarial attack methods have too large attack scope and interference intensity, and do not have the ability to universally attack different neural network models, resulting in poor concealment, deception, and transfer deception of adversarial attacks. In this article, we propose a general keypoints sparse attack (KPSA) method for SAR images, which achieves excellent deception and transfer deception while controlling the attack range and jamming strength. It is a general and efficient adversarial attack method. KPSA uses the generator to generate intensity interference images and imposes amplitude constraint, uses the keypoints extraction method to generate position interference images to realize sparse attack, uses the difference value loss function to achieve fast convergence of the training model, and achieves transfer deception performance without depending on the type of discriminator. The deception rate and transfer deception rate experiments were conducted on man portable surveillance and target acquisition radar (MSTAR) and automatic target recognition network-surveillance and target acquisition radar (ATRNet-STAR). By comparing with the advanced methods in the field of adversarial attacks, it was verified that the KPSA method is superior to the current advanced methods, and it was confirmed that the KPSA method has application value in various complex practical scenarios.

Abstracts:Conventional visual servoing techniques, such as position-based visual servoing (PBVS) and image-based visual servoing (IBVS), rely on inverse Jacobian computations to estimate the desired states of a multirotor, including position and velocity profiles. This reliance not only increases computational complexity but also heightens sensitivity to image noise. Furthermore, these methods typically inject reference trajectories into the outer position control loop, which exacerbates error accumulation as these references propagate to the inner attitude loop. To overcome these limitations, this article proposes the augmented dynamics visual servoing (ADVS) framework, which establishes a direct mapping between image pixel variations and the multirotor’s torque and thrust inputs. By bypassing inverse Jacobian computations, this approach treats image noise as system noise, enabling the application of robust control strategies to mitigate its effects. The proposed framework leverages a time-varying finite-time sliding mode control strategy, where control gains dynamically adapt based on the desired error convergence time. Simulation and experimental results highlight the superiority of ADVS when compared to the existing PBVS, IBVS, and dynamics-based visual servoing approach.

Abstracts:The low probability of intercept (LPI) strategy can significantly reduce the probability of suffering electronic intelligence (ELINT) and jamming for radar. However, some unreasonable LPI designs may seriously weaken the target detection and tracking performance of the radar. Thus, this article explores a deep reinforcement learning-based optimization strategy for radar LPI signals to better counter jamming systems. The entire optimization process is modeled as a Markov decision process and implemented via a double-deep Q-learning network through online interaction with the unknown jamming environment. First, the strategy provides three actions of transmit power, modulation type, and carrier frequency for radar agent to evade the reconnaissance by the ELINT system, which create more challenges for the entire workflow of the ELINT system. Thereafter, a reward function with freely adaptable weights is designed to guide the DRL algorithm for dynamically balancing the LPI performance and the coherent integration (CI) performance of the radar, where the signal strategy duration and the transmit power are focused on. The simulations demonstrate that the proposed method performs well in the unknown environment, and the detection performance of the radar is proved by performing cell averaging constant false alarm rate detection on the range-Doppler images. When less than 1$\%$ of the pulses are jammed, the proposed strategy can get higher CI performance than the compared strategy. This work can provide a dynamic strategy for radar pulses and ensure radar target detection performance in electronic warfare.

Abstracts:In recent decades, a significant advancement in the Global Navigation Satellite System (GNSS) has been adopting the vector tracking (VT) technique, particularly in environments where stable and reliable positioning is essential. It facilitates rapid reacquisition and continuous tracking of GNSS signals by linking the data processing and signal processing modules. In challenging environments affected by multipath and non-line-of-sight errors, such interaction adversely degrades VT’s performance by introducing non-Gaussian error propagation between the navigation processor and baseband channels. This study presents a robust filtering approach based on maximum correntropy criterion optimization to address the above challenges. A multikernel assignment strategy has also been formulated to enhance VT stability in demanding scenarios. Extensive tests have been conducted to assess the performance and efficacy of the proposed methodology. The results indicate that the proposed filter model significantly reduces positioning errors in static and dynamic scenarios. Moreover, the method demonstrates resilience and reliability across diverse urban settings.

Abstracts:This article presents a prescribed performance optimal backstepping control method based on reinforcement learning (RL) to address the challenge of achieving optimal control for hypersonic vehicles (HSVs) while accounting for dynamic performance under disturbances. A nonlinear model of the HSV is developed, and the controller design is divided into altitude and velocity subsystems. Optimal control commands for each subsystem are derived using RL within the actor–critic framework. To enhance the antidisturbance capabilities of RL, the total system uncertainties are estimated through an extended state observer (ESO), effectively balancing optimal control with disturbance rejection. The proposed lightweight RL method incorporates an adaptive update law for weight adjustment, eliminating the need for the trial-and-error process typical of conventional RL. Furthermore, a time-varying barrier Lyapunov functions based on prescribed performance theory ensures the stability of the closed-loop system and ensures global state convergence within prescribed performance bounds. Simulation results confirm the effectiveness and superiority of the proposed method.

Abstracts:Performance lower bounds are utilized as a standard to evaluate the accuracy with which parameters are estimated. However, lower bounds for joint estimation have not been fully explored when it comes to simultaneous estimation of multiple parameters. The Cramér–Rao bound (CRB) can only guarantee asymptotic tightness but not provide tight performance bounds for parameter estimators under low-to-medium signal-to-noise ratio (SNR) conditions. Consequently, we provide a tight performance bound for the joint estimation of the direction of arrival (DOA) and frequency with a uniform linear array (ULA). First, based on the placement of the receiving antennas, the received signals are categorized into two distinct datasets. Then, the output probability density function of the ULA is established based on the additive Gaussian white noise model. Finally, the DOA and frequency entropy error (DFEE) bound is derived as performance lower bound via the joint a posteriori entropy. The DFEE provides a global tight lower bound for the joint estimation of the DOA and frequency. In addition, an approximate expression for the DFEE is derived, capturing the relationship between array performance and parameters, such as the SNR and array elements. Compared to the CRB, the DFEE provides a more effective performance low bound. The performance bound of the DFEE approaches that of the traditional maximum likelihood method, with simulation results confirming its effectiveness and superiority.

Abstracts:This article proposes a scale-independent adaptive control strategy that allows a quadrotor to rely solely on optic flow output to achieve a soft vertical landing on stationary surfaces. Most prior schemes that rely on a fixed-gain approach to regulate optic flow suffer from instabilities induced by nonlinear dynamics of the vertical landing process. To overcome this drawback, this scheme proposes an adaptive sliding mode control strategy with theoretical stability guarantees that incorporates performance and input constraints to achieve desired transient and steady-state performance in the presence of uncertainty arising from wind gusts, ground effects, and possible actuator saturation. In particular, detailed Lyapunov stability analysis is used to demonstrate that the optic flow error converges exponentially to a small prescribed bound in steady-state, which is shown to be the key to achieving a guaranteed soft touchdown. Extensive experimental results substantiate the efficacy of the proposed strategy using a micro-quadrotor platform in a controlled environment. Moreover, a laterally stabilized soft landing with tailored descent behavior is also shown using a mini-quadrotor platform in real-world (unstructured) settings in the presence of wind gusts. Performance comparison studies with an optic flow-based fixed-gain proportional-integral control strategy and an altitude-based vertical velocity control strategy are also included, demonstrating the superior performance of the proposed strategy in comparison with these alternative designs.

Abstracts:In response to the susceptibility of networked systems to factors such as inaccurate modeling and sensor failures, a robust state estimation method for uncertain linear discrete systems with random measurement losses is presented in this article. The impact of random parameter uncertainty on the system model is not limited to specific forms. The resulting robust state estimator shares a structural resemblance with the Kalman filter, facilitating recursive implementation with comparable computational complexity. Moreover, a pseudo-covariance matrix recursive expression suitable for convergence analysis is derived. It is proven that under certain controllable and observable conditions, the pseudo-covariance matrix of the robust state estimator converges with probability one to the corresponding stationary distribution. Numerical simulation results based on a vehicle–trailer system show that, compared with the traditional Kalman filter, the proposed method achieves an improvement of at least 1 dB in estimation accuracy, demonstrating greater robustness to parameter modeling errors.

Abstracts:Forward-looking superresolution imaging is essential for enhancing radar detection capabilities, particularly within real-aperture systems. Traditional subspace methods face challenges in constructing effective subspaces with limited snapshots, and deconvolution methods require substantial computational resources. This article presents a concise description of a nonparametric approach known as the polarization modulation spectrum (PMS), which capitalizes on target polarization differences to bypass the need for covariance matrix construction in subspace superresolution methods. By establishing a polarization filter bank in the spatial domain, the PMS enhances radar angular resolution from a single snapshot. We develop a fully polarimetric multichannel radar echo signal model to elucidate the enhancement of spatial resolution through PMS. A fast PMS superresolution algorithm is proposed, and its performance is analyzed from the aspects of resolution, polarization difference, the influence of noise, and complexity. The method’s effectiveness is validated through numerical simulation and scene inversion data, demonstrating a notable advancement in radar’s forward-looking imaging capabilities.