In this work, an ultra-thin frequency selective surface is developed with a square unit cell loaded with three ring slot pairs. The outer ring slot pair is loaded with metal shorts It achieves one polarization-insensitive passband with sharp transition at Ka-band. The design is verified with full-wave simulated results, as well as the measured results. The fabricated FSS has a flat passband and its measured 3-dB bandwidth is from 33 GHz to 37.5 GHz. A complete angular stability study of the FSS has been conducted. Meanwhile, it can directly attach to the aperture of the horn antenna to lower the whole profile of FSS-antenna configuration. Working as a radome, this FSS can be a good candidate for radar communications and radar cross section reduction in radar stealth research area at Ka-band.

In this letter, we proposed a novel unsupervised learning strategy for direction-of-arrival (DOA) estimation network. Inspired by the sparse power spectrum and $\ell_1$ -norm optimization, we develop a novel loss function to cooperate with the estimation network. Unlike the prior DL-based methods, the proposed method does not need any manual annotations for training and validation datasets. Compared with state-of-art methods, the proposed method can automatically increase the degree of freedom of the array without further pre-processing on the covariance matrix of array observation data. Moreover, the proposed method can obtain clear spectrum and precise DOAs under harsh estimation environments without any time-consuming problem.

In this paper, we present a novel cascaded deep neural network architecture for high-precision direction-of-arrival (DOA) estimation in the presence of multiple array errors. The proposed framework addresses complex nonlinear distortions through an end-to-end feedforward computing strategy, eliminating the need for explicit error modeling. It consists of two key components trained in parallel: (1) four-layer adaptive spatial filters optimized using spatial-domain supervision on covariance matrices, and (2) an eight-layer MUSIC-NET that maps synthetic covariance matrices to MUSIC-like spectrum via self-supervised learning. The output of spatial filters serves as the input for MUSIC-NET, forming a two-stage architecture that enables adaptive array error mitigation during DOA estimation. Simulation results demonstrate that the proposed architecture exhibits superior robustness and generalization capabilities across unseen scenarios, outperforming state-of-the-art methods in low signal-to-noise ratio and limited snapshot conditions.