DOA estimation based on STFT and complex-valued CNN

To solve the problem of performance degradation of the direction-of-arrival (DOA) estimate using covariance matrix in low SNR environment, STFT is used to extract features. The complex signals captured by each array element are processed through the fast Fourier transform (FFT) at varying time intervals, and the maximum Magnitude is calculated to obtain a time-frequency matrix. This novel method is named Maximum Magnitude STFT (M-STFT). According to the complex-valued time-frequency matrix, a complex-valued convolutional neural network (CVCNN) is used to fully extract useful information. Furthermore, to achieve a better performance, 2D-FFT is applied as a kernel function in convolutional layers. Simulation results show that, the method proposed in this article can effectively identify the direction of unknown signals, achieving a success probability of 83.63% at SNR of -15dB.