DOA estimation using GRNN for acoustic sensor arrays

• Published in: Multidimensional systems and signal processing Vol. 34; no. 2; pp. 575 - 594
• Main Authors: Yao, Qihai, Wang, Yong, Yang, Yixin, Yang, Long
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.06.2023; Springer Nature B.V
• Subjects: Algorithms; Artificial Intelligence; Artificial neural networks; Circuits and Systems; Direction of arrival; Electrical Engineering; Engineering; Neural networks; Sensor arrays; Sensors; Signal classification; Signal,Image and Speech Processing; Sound sources; Support vector machines; Training; DOA estimation; Array signal processing; Generalized regression neural network; Machine learning
• ISSN: 0923-6082; 1573-0824
• DOI: 10.1007/s11045-023-00877-9

This paper proposes a direction of arrival (DOA) estimation method for an acoustic source using linear sensor arrays on the basis of generalized regression neural network (GRNN). The real and imaginary parts of the received data of linear sensor arrays in the frequency domain are vectorized and spliced into a one-dimensional sequence as the input feature. The application of this method is studied in three scenarios on noiseless, noisy, and hybrid training sets. Simulations show that the GRNN algorithm has higher accuracy at high SNRs than the support vector machine (SVM), convolutional neural network (CNN) and multiple signal classification (MUSIC) methods, and only the GRNN method can estimate the DOA effectively at low SNRs. According to the different accuracy requirements in practical applications, this paper also provides the selection rules for an appropriate training set for the GRNN method. Therefore, the GRNN method can achieve effective the DOA estimation in different SNR environments of many scenarios.