Fundamental Trial on DOA Estimation with Deep Learning

Abstract

Direction of arrival (DOA) estimation of wireless signals has a long history but is still being investigated to improve the estimation accuracy. Non-linear algorithms such as compressed sensing are now applied to DOA estimation and achieve very high performance. If the large computational loads of compressed sensing algorithms are acceptable, it may be possible to apply a deep neural network (DNN) to DOA estimation. In this paper, we verify on-grid DOA estimation capability of the DNN under a simple estimation situation and discuss the effect of training data on DNN design. Simulations show that SNR of the training data strongly affects the performance and that the random SNR data is suitable for configuring the general-purpose DNN. The obtained DNN provides reasonably high performance, and it is shown that the DNN trained using the training data restricted to close DOA situations provides very high performance for the close DOA cases.

Accuracy Improvement in DOA Estimation with Deep Learning

“…Our previous study verified the estimation capabilities of DNNs under a simple estimation situation and discussed the parallel use of a general-purpose DNN and the DNN designed for a specific scenario where two close DOA signals were incident [15]. However, [15] treats an integer DOA estimation problem. Issues of on-grid problems in the DNN have not been discussed yet in comparison to other off-grid estimation techniques.…”

“…These papers discuss applicationoriented performance, and the DNN configurations are not general in terms of the array size [11], [13], [14], the array geometry [13], the use of subregion decomposition [12], or the use of complex-valued network [14]. Our previous study verified the estimation capabilities of DNNs under a simple estimation situation and discussed the parallel use of a general-purpose DNN and the DNN designed for a specific scenario where two close DOA signals were incident [15]. However, [15] treats an integer DOA estimation problem.…”

“…[10] uses a convolutional neural network (CNN) to estimate near-field sources with spherical wavefront model. In the radio resource localization field, several papers can be seen recently [11]- [15]. In [11], the DOA estimation results are used for massive multiple-input multiple-output (MIMO) channel estimation, and estimation performance related to the number of hidden layers was precisely demonstrated.…”

“…As a solution for this problem, we propose a method of combining two DNNs whose grids are staggered. Furthermore, we train the DNN suitable to the close DOA case, evaluate the estimation accuracy, and propose a parallel use of multiple DNNs designed for different scenarios as in [15]. In the rest of paper, the configuration and training process of DNN are shown after formulation of the array structure and signal arrival model.…”

Unsupervised Learning Strategy for Direction-of-Arrival Estimation Network

“…With the rapid development of artificial intelligence (AI), researchers have designed many deep-learning-based (DLbased) networks to achieve data-driven estimation approaches [16]- [18]. There are generally two kinds of models to realize DL-based estimation: the classification model and the regression model [19], [20].…”