To overcome the limitations of slow convergence in traditional iterative methods and poor generalization in deep learning approaches for high-contrast electromagnetic inverse scattering problems (ISPs), this paper proposes a physics-enhanced variational Born iterative network (PE-VBIM-Net). Inspired by the variational Born iterative method (VBIM), the proposed network architecture mimics the alternating update process between total field and contrast through a cascade of sub-networks. A physics-enhanced module (PEM) based on singular value decomposition (SVD) is introduced to extract dominant induced current components from scattered fields, enabling more accurate initialization of the total field. This module is integrated throughout the iterative process to strengthen the physics-guided nature of the network. Initial contrast is obtained by fusing reconstructions from backpropagation (BP) and diffraction tomography (DT), which are then refined by the contrast update network. Furthermore, a hierarchical physics-consistency loss function is designed to jointly constrain the contrast, total field, and induced current, thereby enforcing physical consistency during reconstruction. The effectiveness and generalization of PE-VBIM-Net are validated on both synthetic and experimental data, demonstrating its superior performance in reconstructing high-contrast scatterers.
In the proposed work, a flexible, multilayered, Frequency Selective Surface (FSS)-based microwave absorber is proposed that is also suitable for covering cylindrical surfaces. The issue of misalignment in multilayered FSS while applying it to cylindrical surfaces is addressed by incorporating changes in unit cell simulation methodology. A novel unit cell simulation technique for cylindrical surfaces is proposed which significantly reduces the time taken for simulation of FSS on a cylindrical geometry. For fabrication of the prototype, unit cell patterns are printed on a matte finish PET (Polyethylene terephthalate) sheet with resistive ink, using screen printing technology. For uniform dielectric spacing between the resistive layers, an airlike mattress foam is characterized and used. For flexible ground, aluminum foil is used instead of conventional copper sheet. Use of aluminum foil, mattress foam and PET sheet, makes the proposed absorber cost-effective as well. The proposed absorber gives more than 90% absorption for 3.9 GHz - 17.3 GHz, with fractional bandwidth of 126.4%. The absorber is polarization-insensitive and maintains significant absorption up to an incident angle of 500. Measurement results match well with simulation for flat, as well as cylindrical geometries.
Mainlobe deceptive jamming is difficult to suppress when multiple digital radio-frequency memory (DRFM) jammers are range-spread and angularly co-located with the target, since the received echoes are coherent with the radar waveform and the local range training data can be severely contaminated. Conventional polarimetric countermeasures usually operate in the two-channel H/V receive domain and therefore provide limited degrees of freedom under dense target-jammer mixing. This paper proposes a receive-side polarization-time linearly constrained minimum variance (PT-LCMV) processing framework for fully polarimetric radar with intra-pulse polarization agility. Conventional polarimetric pulse compression and adaptive range-gate selection are first used to isolate the target-jammer mixing region. Guided by a prescribed transmit polarization trajectory, the matched filters are then segmented in fast time, and the segment-wise H/V outputs are stacked into a 2M-dimensional polarization-time snapshot for each range cell. A nominal target polarimetric scattering matrix is used to construct a model-based target steering vector, while a target-suppressed local covariance matrix is estimated from neighboring range cells to characterize the dominant residual interference structure. The resulting PT-LCMV filter preserves the nominal target response and suppresses the dominant mainlobe deceptive components in the expanded polarization-time space. Simulation results under several single- and multi-jammer DRFM scenarios show that the proposed method restores a sharp target peak, achieves substantial output signal-to-interference-plus-noise ratio (SINR) gains, and outperforms representative benchmark methods.
The design of a radar phased array two-way radiation pattern, which is the product of transmit and receive array factors, typically suffers from undesired minor lobes near the main beam, along with the hardware cost and complexities associated with transmit/receive (T/R) modules and beamforming network design. This paper presents a systematic approach for reducing the peak sidelobe level (PSLL) in two-way radiation patterns of phased arrays with both linear and rectangular planar geometries. Specifically, for a given linear or planar transmit array with all elements active, we propose a receive array configuration that employs only a limited number of active elements. This approach reduces cost, complexity, and power consumption while maintaining a low PSLL in the two-way radiation pattern. The proposed receiving array is based on a Quasi-Uniform Sparse Receive Array (Q-USRA) configuration, whose aperture can be either smaller or larger than that of the transmit array. This configuration aligns the receive nulls (SLL peaks) with the SLL peaks (nulls) of the transmit array pattern, thereby achieving a low PSLL in both linear and planar two-way radiation patterns. The design parameters of the Q-USRA configuration (i.e., the number of active elements and the receive aperture length relative to the transmit array) are obtained via an exhaustive search and are independent of transmit array size for both linear and rectangular planar array geometries. This approach enables flexible array design while overcoming the computational challenges associated with methods based on genetic algorithms (GAs), combinatorial traversal algorithms (CTAs), and mixed-integer programming (MIPs). The theoretical findings are further validated through full-wave simulations with microstrip patches and dipoles as radiating elements, thereby confirming the suitability of the proposed array design framework for realistic implementations.
The parabolic wave equation (PWE) method is widely used for modeling radio wave propagation in tunnel environments. However, performing simulations over long distances or electrically large domains remains computationally intensive. This paper proposes an efficient on-the-fly multi-stage generative adversarial network (GAN) model that integrates deep convolutional GAN (DCGAN), cycle-consistent GAN (CycleGAN), and super-resolution GAN (SRGAN) for predicting received signal strength (RSS) in tunnels without the need for extensive pre-training. Unlike traditional supervised models that rely on large paired datasets, the proposed method first employs unsupervised data augmentation to generate structurally consistent coarse–fine mesh RSS pairs using DCGAN and CycleGAN. These synthetic pairs are then used to train SRGAN, enabling high-fidelity fine-mesh RSS predictions from coarse-mesh inputs. Experimental results demonstrate that the proposed model achieves accurate RSS predictions with significantly reduced computational cost compared to conventional PWE simulations. Furthermore, the model’s effectiveness has been validated through measurement campaigns in a real-world tunnel environment.
