Attention based DOA estimation in the presence of unknown nonuniform noise

Abstract

To solve the problem of performance degradation of the direction-of-arrival (DOA) estimate in non-uniform noise environment, we propose a novel attention mechanism using deep learning technology, namely array covariance attention (ACA). Specifically, to design the ACA, according to the structural characteristics of the covariance matrix, the pooling operation is removed, and one-dimensional convolution kernels are used to aggregate correlation characteristics in two spatial directions. With fully connected layers and non-linear activation layers, the characteristics are then coded into the perceptual attention matrix to improve useful information of the covariance matrix. Furthermore, to achieve a better performance, the integration position of the attention mechanism is also discussed in the network. Finally, a new deep-learning network is created for DOA estimation in the presence of non-uniform noise. The experimental results demonstrate the efficiency and superiority of the proposed network.