This paper proposes a game-theoretic fuzzy multi-agent reinforcement learning framework for cooperative multiple unmanned aerial vehicles (UAV) pursuit-evasion in complex dynamic environments. Three-dimensional Apollonius dominance regions are integrated with an ordinal potential game (OPG), for which the existence of a Nash equilibrium is established. Then, a Game-Theoretic Fuzzy Multi-Agent Soft Actor-Critic (GT-FMASAC) algorithm is developed by incorporating interval type-2 fuzzy logic into centralized training with decentralized execution to improve robustness under uncertain observations. A hierarchical hybrid reward further promotes cooperative pursuit. Simulation and ablation studies demonstrate improved capture performance, scalability, and computational efficiency over representative MARL baselines.
In airborne stepped-frequency waveform stripmap synthetic aperture radar (SAR), the stepped-frequency (SF) relationship between subband chirp signals is utilized to synthesize broadband range spectrum, thereby enabling high-resolution range profile. In practice, residual constant inter-subband phase errors can reduce spectral coherence and distort the synthesized range impulse response, resulting in increased sidelobes and even multi peak responses. This paper derives an explicit approximate relationship between the distorted synthesized range impulse response and the inter-subband phase errors, showing that the single-subband sinc response is modulated by a finite cosine series term. Based on this analysis, phase calibration is formulated as a maximum mixed $\ell _{\infty,1}$-norm (MLN) optimization that searches for the phase vector maximizing the $\ell _{\infty,1}$-norm of the subband synthesized data. To mitigate the sensitivity to initialization and reduce computation, a subband spectrum overlap model (SSOM) exploits the practical spectral overlap between adjacent subbands to provide a fast rough estimate, which initializes the subsequent MLN refinement. Simulated scenes and real three-subband airborne measurements demonstrate that, compared with traditional methods, the proposed analytical characterization framework quantifies the impact of inter-subband phase errors on the synthesized impulse response, and the proposed phase-error calibration scheme achieves a 2-3.5 dB improvement integrated sidelobe ratio (ISLR) and peak sidelobe ratio (PSLR).
This paper investigates the suppression of interrupted sampling repeater jamming (ISRJ) in multiple-input multiple-output (MIMO) radar systems from the perspectives of the spatial, fast-time, and slow-time domains via joint transceiver design. While traditional MIMO radar effectively mitigates sidelobe interference, suppressing mainlobe ISRJ solely from the time domain presents a severe physical challenge: since the target and jammer share the same spatial bin, the system must rely heavily on waveform uncorrelation to separate the jamming, which is highly difficult and inherently risks severe target cancellation. To tackle this issue, we propose a joint space-time design framework that maximizes the signal-to-interference-plus-noise ratio (SINR) to eliminate sidelobe jamming and optimally filter mainlobe jamming. At the core of this framework is a jamming-aware integrated sidelobe level (JA-ISL) constraint. We equip a subset of the space-time filters with mismatched filtering capabilities, enabling them to act as fast-time mismatched filters to suppress mainlobe jamming in the fast-time domain when the maximum SINR criterion fails to do so effectively. However, to minimize the impact on the overall optimization, this constraint serves as a safety-net mechanism without incorporating waveform diversity, which correspondingly degrades the MIMO radar performance to a phased-array-like level. Furthermore, we develop an efficient solver based on the MM-ADPM framework. Numerical results under the specified jammer-parameter settings demonstrate that the proposed method effectively suppresses simultaneous mainlobe and sidelobe ISRJ and achieves a high output SINR even when conventional spatial filtering is severely limited.
This paper proposes a resilient consensus control framework for heterogeneous multi-agent systems under false data injection attacks, where each agent is either an Un manned Aerial Vehicle or an Autonomous Surface Vehicle. In spired by graph-aware digital twin, a distributed leader-follower controller is derived in a two-layer cyber-physical architecture. Unlike conventional residual-based methods, this framework employs dual detection mechanisms: adaptive residuals for non IMP attacks and Kullback-Leibler divergence for IMP-based stealthy attacks. This work is unique in the sense that the GNN-LSTM digital twin enables the network for real-time attack detection as well as state compensation comparatively. Lyapunov analysis proves asymptotic consensus under nominal conditions and uniform ultimate boundedness under attack. Simulation results demonstrate 97.9% true positive rate with 2.1% false positive rate, 87% peak error reduction, and 1.8-3.6 s recovery times across multiple scenarios.
This paper proposes a fractional-order-free predefined-time guidance law via asymmetric feedforward for the interception of a high-speed target. First, a state-driven performance function is designed to constrain the line-of-sight (LOS) angle rate. It can autonomously tighten the performance boundary based on the convergence state of the LOS angle rate and the available interceptor maneuverability. This mechanism actively accelerates the convergence rate and alleviates the parameter-tuning burden. Second, a stability-guaranteed asymmetric feedforward strategy is developed to actively exploit the beneficial disturbance and the inherent nonlinear dynamic of the interceptor, thereby enhancing system stability and accelerating the response rate. This strategy is augmented by a secant-style practical prescribed-time disturbance observer, which provides practical disturbance estimation at a prescribed time. Based on the performance function and the feedforward strategy, the fractional-order-free predefined-time guidance law is designed, which can achieve the exact predefined-time convergence of the LOS angle rate. The predefined-time convergence characteristic does not rely on fractional-order term or singular time-scaling gain, which significantly mitigates the computational burden while avoiding the time-scaling-induced singularity problem. Theoretical analysis and numerical simulations confirm that the proposed approach guarantees rigorous system stability and achieves precise target interception.
