When the elevation of target is smaller than a beamwidth, the complex multipath signals will distort the feature of direct signal reflected from target. The elevation of target can hardly be estimated accurately. Hence, in this paper, we proposed three kinds of neural networks models including deep neural network (DNN), 1-D convolutional neural network (1-D CNN) and 2-D convolutional neural network (2-D CNN) and their optimization method to mitigate phase distortion caused by multipath signals and enhance the phase feature of direct signal. The direction of arrival (DOA) estimation accuracy of physics-driven methods including digital beamforming (DBF) and multiple signal classification (MUSIC) is effectively improved. Concretely, we analyze the origins of error of DOA estimation in multipath environment and discuss the importance of phase feature to DOA estimation. A complete framework of feature-to-feature phase enhancement is built for DOA estimation in radar systems. The results of experiments with real data collected from a very high frequency (VHF) radar demonstrate the superior DOA estimation performance of proposed feature-to-feature learning methods with respect to other state-of-the-art methods including physics-driven methods and existing data-driven methods.

In this paper, an effective solution for adaptive sensor scheduling integrated with power and bandwidth allocation is developed for centralized multiple target tracking (MTT) in the netted collocated multiple-input multiple-output (C-MIMO) radar system. By incorporating a modified particle filter, the predicted posterior Cramér-Rao lower bound (PCRLB) is calculated. Aiming at improving the sum of target tracking accuracy of location, the predicted PCRLBs for the location of multiple targets is derived as the optimization metric. Owing to the dynamic sensor scheduling is NP-hard, a convex relaxation technique is adopted for the sensor scheduling. Taking account of the problem of power and bandwidth allocation is non-convex, a joint convex relaxation technique and cycle minimizer algorithm is adopted to convert the non-convex optimization problem into a series of convex problems. After the results of resource allocation are fed back to each node, an adaptive closed-loop system is established. Numerical results demonstrate that the location tracking accuracy can be improved efficiently by the proposed algorithm.

In low-altitude target situation, the multi-path signals cause the amplitude-phase distortion of direct signal from targets and degrade the performance of existing methods. Hence, in this paper, we propose a phase enhancement method for low-angle estimation using supervised deep neural network (DNN) to mitigate the phase distortion, thus to improve direction of arrival (DOA) estimation accuracy. The mapping relationship between original phase difference distribution of received signal and desired phase difference distribution is learned by DNN during training. The phase of test data is enhanced by trained DNN, and the enhanced phase is used for DOA estimation. We explain the significance of enhancing phase instead of amplitude by discussing the sensitivity of amplitude and phase on DOA estimation. Moreover, we prove the effectiveness and superiority of the proposed method by simulation experiments. The results demonstrate that the proposed technique has a better performance in terms of estimation error and goodness of fit (GoF) than the physics-driven DOA estimation methods and state-of-the-art methods including feature reversal and support vector regression (SVR).

In this letter, we propose two angle separation learning schemes (ASLs) to address the coherent DOA estimation problem. We first show that the columns of the array covariance matrix can be formulated as under-sampled linear measurements of the spatial spectrum. Secondly, the computational load of coherent DOA estimation contains two-dimensional searching process, which is reduced through introducing angle separation. Correspondingly, two classification learning models integrating the angle separation are proposed to solve the problem of coherent DOA estimation. Compared to classic sparsity-inducing methods with complex computational interactions, the proposed ASLs can efficiently obtain DOA estimates in near real time. Moreover, the proposed ASLs improve the DOA estimation performance compared to existing deep-learning based methods. Numerical experiments prove the effectiveness and superiority of the presented ASLs. Simulation results show that the new techniques have a better performance in term of estimation errors and generalization ability than the classic physics-driven and state-of-the-art deep-learning based DOA estimation methods, especially in demanding scenarios with low SNR and limited snapshots.