Joint estimation of DOA and polarization parameters based on CNN-Transformer hybrid architecture

Compared with traditional scalar arrays, polarization-sensitive arrays can simultaneously obtain spatial domain information and polarization information.So it has stronger anti-interference ability and stronger resolution rate.Traditional algorithms such as MUSIC and ESPRIT have problems of low parameter accuracy and large operation time cost under low SNR.To solve the above problems, this paper proposes a joint estimation method of DOA and polarization parameters based on CNN-Transformer hybrid architecture.First, a four-layer CNN (E-CNN) is used to estimate the real part of the polarization electromagnetic vector of the signal. The estimated values are then combined with the real and imaginary parts of the covariance matrix and input into the parameter estimation network to enhance the polarization features of the data.The UCA-CT network combines CNN and Transformer for multi-level feature extraction. It achieves joint estimation of DOA and polarization parameters.The experimental results show that the UCA-CT network achieves an RMSE of 0.98° for DOA and 2.4° for polarization parameters at -10 dB. Compared to traditional subspace algorithms, the RMSE of DOA is reduced by 0.1°, and the RMSE of polarization parameters is reduced by 0.5°. The results demonstrate that the proposed method improves polarization parameter estimation while maintaining DOA estimation accuracy.