Dynamic Behavior Modeling of Analog Circuits Based on Deep Operator Network

Analog circuits are widely used in audio amplification, computing, power amplification, radio communication, etc. It is crucial to design a reliable circuit health monitoring system. Neural networks (NNs) are general-purpose approximators of continuous functions, however, a little-known but powerful finding is that a neural network with a single hidden layer can accurately approximate any nonlinear continuous operator. The general approximation theorem for operators extends point-topoint mapping to function-to-function mapping, forming the deep operator network (DeepONet). The zero-state response of an analog electrical circuit is equivalent to a convolution operator. This paper aims to use DeepONet to model the physical dynamic behavior of the second-order band-pass filter analog circuit, complete the function mapping from the input end (excitation) to the output end (zero-state response) of the circuit, and then learn the entire filtering function of the circuit. After the network model is trained, we can obtain the output function value of any time point of the circuit for any input of the circuit, to achieve real-time monitoring of the circuit, which is of great significance for improving the system reliability.