Abstract: Aiming at the problem of millimeter-wave multipath direction-of-angle estimation, a deep learning-based direction-of-angle estimation method is proposed. This method is realized by constructing the mapping between the covariance matrix and multipath direction-of-angle. The proposed method first constructs a sampling covariance matrix based on the received signal. Then, a multi-label classification model using the deep residual shrinkage network is introduced into the direction-of-angle estimation of the line-of-sight path. Finally, a regression model using the proposed convolutional neural network is used to achieve direction-of-angle estimation of multiple non-line-of-sight paths. A series of simulation results show that the proposed method significantly reduces root mean square error compared with the traditional methods. The lower angle estimation error is achieved in low signal-to-noise ratio. Moreover, under different scenarios, the proposed approach has the good applicability.

Key words: mmWave, multipath direction estimation, deep residual shrinkage network, convolutional neural network