A Deep-Learning-Based Method for Diagnosing Time-Varying Plasma Adopting Microwaves

Abstract

Performing as an important technology, microwave interferometry helps understand and analyze plasma characteristics. However, the relatively appropriate diagnosing methods being discovered for investigating fast time-varying plasma have met with limited success so far. The crucial problem of this task lies in eliminating the interference of additive noise while compensating for phase ambiguity of the initial point and of the phase jump points. In particular, randomly varying electron density and noise make it impossible to directly detect these points. Theoretical analysis has found that the time-varying electron density of plasma causes the received signal to rotate on the complex plane. Inspired by this fact, a deep learning (DL) method is proposed to track the time-varying plasma and solve the above two problems, during which the deep neural network (DNN) is utilized to extract the curve from the data, thereby eliminating the noise and detecting the initial point and the phase jump points according to the projection of the data on the curve. Experimental results suggest that the learned curve is in good consistency with the ground-truth curve. Meanwhile, the time-varying electron density and the collision frequency can be accurately diagnosed using the learned curve.