A deep learning method to estimate independent source number

Blind source separation is one of the main research branches of blind signal processing. However, most of the algorithms of blind source separation are based on the known assumptions of the number of signal sources. Therefore, in the blind source separation field, it is important to determine the number of independent sources. Compared with the unsupervised algorithm of often used in blind source separation, this paper creatively proposes a method of independent sources number estimation based on the neural network which is a supervised learning method. And then, blind signal separation is carried out. The experimental results show the separation of mixed basic signals and the modulation of different signal to noise ratio, and compared with the traditional unsupervised method.