Visual object tracking based on Unmanned Aerial Vehicles (UAVs) faces critical challenges posed by specular glare, which causes severe feature degradation. To address this, we propose a robust hardware-software collaborative tracking system centered on a custom-designed Adaptive Polarized Light (APL) sensor. Specifically, the APL sensor is driven by real-time image quality evaluation and proactively suppresses specular glare at the optical level by adaptively optimizing its polarization angle. Leveraging the APL sensor, a closed-loop visual tracking and control framework is developed by integrating motion-constrained tracking and reflection-aware control to maintain target localization and a best observation angle based on Brewster's law. Real-world experiments on a resource-constrained UAV edge platform show that the system achieves real-time and robust tracking at 30.5 FPS, with a 68.0% Intersection over Union (IoU) and tracking precision within 35 pixels under extreme glare conditions.
Three-dimensional (3D) reconstruction of building structures using through-the-wall radar (TWR) is essential for situational awareness in urban surveillance and rescue operations. However, existing methods typically suffer from high computational costs and structural fragmentation due to the lack of spatial continuity constraints. To address these issues, this paper proposes an efficient voxel-column sequential structural variational Bayesian (VC-SSVB) framework. Specifically, by leveraging the directionality of wall scattering, we decompose the global inversion into tractable voxel-column sub-problems through an aperture pruning strategy, which significantly reduces the computational complexity. Building on this reduced model, we establish a sequential structural modeling framework incorporating a hierarchical coupled prior to enhance structural continuity. This mechanism propagates spatial information across both inter-column and intra-column neighbors, thereby ensuring structural continuity. The entire framework is calculated via a fast variational inference algorithm that integrates the majorization-minimization (MM) strategy and the preconditioned conjugate gradient (PCG) solver to facilitate efficient closed-form updates of the 3D building map. Extensive simulations and real-world experiments conducted with an unmanned aerial vehicle (UAV)-based radar system demonstrate that the proposed method achieves superior 3D reconstruction quality with enhanced structural continuity and significantly higher computational efficiency compared to existing methods.
Robust and computationally efficient interference suppression in complicated non-Gaussian measurement environments is of paramount importance for aerospace electronic applications. While conventional adaptive filters (AFs) offer computational efficiency, their reliance on static parameters, often customized for specific noise profiles, fundamentally limits their interference suppression capabilities and renders manual tuning impractical in highly dynamic scenarios characterized by multimodal noise. To this end, this paper proposes AdaFilter-Net, a data-driven adaptive filtering framework designed for robust interference suppression in intricate and non-stationary measurement environments. Specifically, unlike traditional AFs that rely on fixed error criteria, AdaFilter-Net employs a fixed-size radial basis function (RBF) network to adaptively learn its error nonlinearities. Benefiting from the universal approximation capability of Gaussian kernel functions, this compact RBF structure allows AdaFilter-Net to adapt its error nonlinearity to different noise conditions while preserving low computational complexity. Furthermore, a spaceefficient sliding window is adopted in AdaFilter-Net to improve its stability in complex tasks, and a deterministic initialization strategy of AdaFilter-Net is introduced to mitigate potential instability caused by the stochastic initialization structure. Moreover, we provide a comprehensive theoretical analysis of AdaFilter-Net, including coupled and weak-coupling steady-state mean square deviation (MSD) analyses, tracking performance analysis under a time-varying system model, convergence analysis with stepsize stability conditions, and computational complexity analysis. Finally, numerical simulations on system identification tasks under complicated noise conditions verify the efficiency and theoretical validity of AdaFilter-Net. Under the same initial convergence-rate setting, AdaFilter-Net achieves a steady-state MSD improvement of approximately 3-8 dB over competitive filters. Additionally, simulations using real-world maneuvering target trajectories demonstrate that AdaFilter-Net maintains robust prediction performance in highly dynamic tracking scenarios, reducing the steady-state RMSE by 12.942 dB and 12.294 dB on two trajectories, respectively. Notably, “complex” in this work refers to complicated and non- Gaussian interference environments rather than complex-valued signal representations, and the proposed formulation and analysis are developed for real-valued scalar-output filtering.
Unimodular waveform design plays a critical role in enhancing the sensing performance of MIMO radar systems. Existing unimodular waveform design approaches can be broadly categorized into methods based on fixed-position arrays, which suffer from limited spatial degrees of freedom, and movable-antenna (MA)–enhanced designs that typically rely on $\ell _{2}$-norm metrics, providing insufficient control over pronounced sidelobe fluctuations. Noting that sidelobe shaping is fundamentally an $\ell _{p}$-norm optimization problem, this paper proposes a joint design of unimodular waveforms and MA positions for MIMO radar beampattern synthesis via weighted $\ell _{p}$-norm minimization. Specifically, we formulate a weighted $\ell _{p}$-norm beampattern matching problem subject to bounded minimum spacing (BMS) constraints on antenna positions and unimodular waveform constraints. The resulting problem is highly nonconvex and strongly coupled, rendering direct solution by conventional methods intractable. To address this challenge, the BMS constraints are incorporated through a softmax-based reparameterization, while the unimodular waveform constraint is modeled on a complex circle manifold (CCM), transforming the original constrained problem into an unconstrained optimization over a Riemannian product manifold. Based on this formulation, we develop an $\ell _{p}$-norm Riemannian-aware mapping ($\ell _{p}$-RAM) framework that jointly optimizes the antenna position vector and the transmit waveform using a Riemannian conjugate gradient algorithm. Simulation results demonstrate that the proposed method achieves quasi-equiripple sidelobe behavior and reduced peak sidelobe levels compared with existing approaches, while maintaining improved mainlobe smoothness.
In next-generation Global Navigation Satellite Systems (GNSS), the transmission of signals with multiple components, multiple carrier frequencies, and diverse spread-spectrum modulations has become an important approach to improve spectral efficiency and system service capabilities. However, this trend also significantly increases the difficulty of constant-envelope multiplexing (CEM) on the satellite payload. In this paper, a general mathematical model of CEM is established and a CEM technique based on convex optimization and time hopping is proposed. The basic idea of this technique is to relax the constant-envelope constraint and the nonrandom periodic component constraint in the original CEM formulation, thereby transforming the otherwise intractable CEM problem into a convex optimization problem that can be efficiently solved. Subsequently, the envelope constancy and the nonrandom periodic component constraint are recovered through mathematical transformations and time-hopping transmission without any loss of multiplexing efficiency. The proposed approach is based on convex optimization and imposes no restrictions on the number, relative power, initial phase and spread-spectrum modulations of components to achieve lower computational complexity, higher flexibility. Moreover, by enforcing the nonrandom periodic component constraint through time hopping rather than by directly constraining the feasible solution space, it further attains improved multiplexing efficiency compared with existing CEM techniques. These advantages provide greater room for next-generation GNSS signal design, enabling support for diverse requirements of future navigation and positioning service.
