Linear Array DOA Estimation Based on CNN-SEBlock Under Low Signal-to-Noise Ratio

Numerous scholars have delved into the DOA (Direction of Arrival) estimation quandary concerning arrays, predominantly focusing on environments exhibiting substantial SNR (signal-to-noise ratios). However, the Music algorithm encounters complete failure in scenarios characterized by low SNR. This paper introduces CNN-SEB, an approach tailored for estimating the orientation of uniform line arrays under conditions of low SNR. CNN-SEB leverages the convolution module within CNN architecture to facilitate the estimation of the line array. This method operates on the acceptance covariance data matrix and incorporates the SE attention mechanism module, enabling the application of varying weights to different feature maps. Comparative analysis against conventional methods such as the traditional CNN algorithm, Music algorithm, encoder algorithm, and CNN-trans algorithm reveals an enhancement in test accuracy. Remarkably, the prediction accuracy achieves 0.7822 in scenarios with a low SNR.