Modeling of a Low-Noise Amplifier with a Recurrent Neural Network

Macromodelling of nonlinear circuits particularly those operating at RF frequencies is of importance and interest specifically in systems where the total simulation time is constrained. Recurrent Neural Network (RNN) is a suitable approach for generating an accurate model for high frequency and nonlinear circuits. Indeed, with having such a model at hand, the circuit designers have the potential to grasp the system level verification way faster than the conventional simulators. In this paper, a low-noise amplifier (LNA) is used as the nonlinear circuit, and a RNN is designed to generate the model. The LNA is configured in LTSPICE, and its simulation data is used as the input of the RNN model via Python programming. The RNN is able to train the ANN without the knowledge of circuit's electronic structure details, and just by having the input-output waveforms of the original nonlinear circuit. The modeling based on the proposed RNN has the same accuracy compared to the original nonlinear circuit. Another feature of the proposed RNN structure is that it can also model the dynamic behavior of nonlinear circuits. In addition, models based on recurrent neural networks are much faster than models based on conventional simulation tools such as LTSPICE. It is worth mentioning that this modeling approach can be extended to a variety of applications, including medical, to have a quick prediction of the vital signals.