Research on multi-state behavior model characterization method for traveling wave tube amplifiers

In recent years, artificial intelligence technologies represented by deep neural networks have been applied in the construction of behavioral models for power amplifiers. High precision nonlinear fitting can meet the requirements of characterizing the behavioral models of power amplifiers, but it is still only applicable to a single working state. With the in-depth study of the input-output characteristics of traveling-wave tube amplifier (TWTA), the output signal is affected by multiple factors such as the frequency of the input excitation signal and temperature changes. Therefore, it is urgent to study how to construct a multi-state behavior model for TWTA based on deep neural networks. This article proposes a multi-state neural network modeling method for TWTA. This method introduces embedded encoding vectors to characterize the various working states of TWTA, and innovatively constructs a multi-state behavior model by adding skip connections to avoid the problem of gradient vanishing. The experimental results show that compared with traditional methods, this method can construct multiple operating states that characterize TWTA without losing model accuracy as the model size increases.