Deep Networks for Direction-of-Arrival Estimation in Low SNR

Abstract

In this work, we consider direction-of-arrival (DoA) estimation in the presence of extreme noise using Deep Learning (DL). In particular, we introduce a Convolutional Neural Network (CNN) that predicts angular directions using the sample covariance matrix estimate. The network is trained from multichannel data of the true array manifold matrix in the low signalto-noise-ratio (SNR) regime. By adopting an on-grid approach, we model the problem as a multi-label classification task and train the CNN to predict DoAs across all SNRs. The proposed architecture demonstrates enhanced robustness in the presence of noise, and resilience to a relatively small number of snapshots. Moreover, it is able to resolve angles within the grid resolution. Experimental results demonstrate significant performance gains in the low-SNR regime compared to state-of-the-art methods and without the requirement of any parameter tuning in both cases of correlated and uncorrelated sources. Finally, we relax the assumption that the number of sources is known a priori and present a training method, where the CNN learns to infer their number and predict the DoAs with high confidence. The increased robustness of the proposed solution is highly desirable in challenging scenarios that arise in several fields, ranging from wireless array sensors to acoustic microphones or sonars.

A Generalized Noise Reconstruction Approach for Robust DOA Estimation

“…Recently, data-driven deep learning-based algorithms have been developed rapidly [31], and their strong nonlinear map-ping capability and robustness have attracted a lot of attention. In terms of DOA estimation, deep neural networks (DNN) [32], convolutional neural networks (CNN) [33], and convolutional recurrent neural networks (CRNN) [34], [35] yield very accurate estimation results through multi-label classification (MLC) or regression space spectrum. However, since the data-driven algorithms are very sensitive to data, while dynamic range of data is not taken into consideration, resulting in the network learning from the experience of prejudice to any fixed signal power during the training process [33].…”

Environment Sensing Considering the Occlusion Effect: A Multi-View Approach

“…In the regime of internet of vehicles, [17] designed a fast beam alignment and tracking algorithm based on a hardware test platform to achieve vehicle tracking while performing millimeter wave communication with low latency and high data rate. As the research interest on ISAC keeps on increasing, more and more potential technologies are also introduced into this field, such as channel modeling [18], joint beam optimization [19] and machine learning [20], etc.…”

“…We plug ( 18), ( 19) and ( 75) into (74), and absorb the constant term into the constant c, we obtain (20) Plugging ( 76) and ( 77) into (78), we obtain (75).…”

Fast Cross-Correlation for TDoA Estimation on Small Aperture Microphone Arrays

“…Chakrabarty and Habets [47], [48] demonstrate that a convolutional neural network (CNN) could be trained on white noise signals to estimate the DoA for a Uniform Linear Array (ULA). Other approaches also demonstrate the robustness of CNNs and fully connecter networks in low signal-to-noise ratio (SNR) scenarios [49], [50], [51], [52]. C ¸akır et al [53] showed that convolutive and recursive neural networks (CRNNs) could be used for sound event detection, followed by Adavanne et al [54], [55] who demonstrated CRNNs could estimate the DoAs for a specific class of sound.…”