TinyResNetDOA-DCS: a tiny residual network for direction-of-arrival estimation with dual channel switched-element system

In conventional direction finding system with dual channel switched-element receiver, the direction-of-arrival (DOA) estimation just can be performed for single source, the accuracy of which deteriorates in the low signal-to-noise ratio case. In this paper, a tiny residual network for DOA estimation with dual channel switched-element system named TinyResNetDOA-DCS is proposed, in which the residual connections, squeeze-and-excitation block (SEBlock), cross stage partial (CSP) connections and DropBlock are employed. The proposed network has the merits of small parameter scale, low computational complexity, real-time of inference and strong generalization ability. As far as we know, it is the first time to estimate the DOA of sources based on neural network with dual channel switched-element system. With the data set formed by the angular spectra features with the minimum variance distortionless response estimator, the advantages of the proposed neural network are illustrated. Simulation results show that the accuracies of TinyResNetDOA-DCS outperform the counterparts in the case of single source with low signal-to-noise ratio.