In this letter, a deep learning framework for direction of arrival (DOA) estimation is developed. We first show that the columns of the array covariance matrix can be formulated as under-sampled noisy linear measurements of the spatial spectrum. Then, a deep convolution network (DCN) that learns the inverse transformation from large training dataset is introduced. In contrast to traditional sparsity-inducing methods with computationally complex iterations, the proposed DCNbased framework could efficiently obtain DOA estimates in near real time. Moreover, the utilization of the sparsity prior improves DOA estimation performance compared to existing deep learning based methods. Simulation results have demonstrated the superiority of the proposed method in both DOA estimation precision and computation efficiency especially when SNR is low.

This letter presents a support vector regression-based direction finding scheme in a wide range of frequencies. The proposed solution consists of two innovations: preprocessing and post-processing. Preprocessing takes advantage of the conjugate symmetry and Toeplitz property of the array covariance matrix to reduce the dimension of the input feature. Post-processing provides the directions of signals with different frequencies by exploiting the inherent relationship between direction and frequency. The experimental results show that this strategy is able to complete wide-frequency-range direction finding with high estimation accuracy and computation efficiency.

This study considers the problem of wideband direction of arrival (DOA) estimation for spatial optical beam‐forming network (SOBFN). Since its system configuration is different from that of the traditional microwave beam‐forming networks, the authors first establish the observation model of SOBFN when multiple broadband signals are received. Two kinds of DOA estimation strategies are proposed for wideband direction finding based on the established signal model. The proposed methods decompose the wideband array output into narrowband outputs in several discrete frequency bins. Then, they take advantages of the spatial sparsity of the discrete narrowband observation models to estimate the spatial power spectrum, either coherently or non‐coherently. A coarse DOA estimate is obtained via peak‐searching in the reconstructed spectrum. The preliminary result is then refined by a linear interpolation procedure. Profound experiments are carried out to demonstrate and compare the performance of the two strategies. Simulation results show that the proposed approaches can achieve wideband DOA estimation of high precision and super‐resolution for SOBFN. Moreover, coherent process is proved to have superior performance than the non‐coherent strategy.