Robust Wideband Adaptive Beamforming by Convolutional Neural Network Structures in an Array System

The structure of arrays capable of receiving wideband signals differs from arrays which can only receive narrowband signals. These arrays should be capable of receiving several GHz instant bandwidth signals over the entire operating frequency, like High Resolution Radars or Terahertz in 6G communication systems. In these arrays, the number of beamformer coefficients increase due to using a time delay line structure, hence leading to a high computational complexity. This is one of the challenges of beamforming in wideband systems. Additionally, there are other factors in classic Wideband beamformers. These factors include poor performance against input DOA (direction of arrival) error, array calibration error, and too many snapshots to reach the steady state of the beamformer. Therefore, this paper focuses on the robustness of beamforming using deep learning technique, which has not been discussed before. The basis of network design is convolutional layers. Two different proposed approaches have been proposed to obtain optimal structure. In the first approach, appropriate values of hyperparameters are estimated, whereas the network model is selected in the second approach. Also, the network training method is done to become more robust against the mentioned factors. In the proposed structures, the constraints of the previous methods have been evaluated. The proposed structure is shown to have a superior performance when compared with other existing algorithms. In addition, the matrix of signal samples is replaced with the covariance matrix at the network's input, so the calculation time is significantly improved compared to the previous methods.