Neural network modeling of the input/output characteristics of a high power backward wave oscillator

Abstract

Summary form only given. A neural network model has been used to characterize the input/output behavior of the Sinus-6 electron beam accelerator-driven high power backward-wave oscillator (BWO) at the University of New Mexico. Since this microwave source uses a 10 ns duration electron beam that is pulsed from one shot to the next, and since the sampling interval in the experimental data is not fixed, a static, continuous neural network model was used to fit the data. Simulation results showed that such a simple nonlinear model is sufficient to accurately describe the input/output behavior of the Sinus-6-driven BWO and that the fitted output waveforms are essentially noiseless. The model used to describe the Sinus-6-driven BWO consists of one input and two outputs. A computer-generated trigger pulse initiates the dosing of a pressurized spark gap switch which leads to a high voltage pulse V propagating down a transmission line leading to the microwave system. The system S consists of an electron gun (also referred to as the anode-cathode (A-K) gap) and a microwave tube. The electron gun generates an electron beam current I for a given accelerator voltage V and A-K gap setting. This current propagates through the microwave gun which consists of a slow wave structure immersed in a strong axial magnetic field. The two outputs from system S are the radiated microwave power (y/sub 1/) (or beam-to-peak envelope power conversion efficiency (z/sub 1/)) and the radiated frequency (y/sub 2/). Our recent work has focused on enhancing our simple model by introducing both the accelerator voltage and beam current (or, more accurately, the spark gap switch pressure and A-K gap setting) as physical inputs to our system S. This more realistic mathematical model will facilitate the design of a controller to maximize both the microwave conversion efficiency and power.