Silicon field effect transistor radio frequency switch harmonic prediction method based on dynamic space mapping

Abstract

The invention discloses a silicon field effect transistor radio frequency switch harmonic prediction method based on dynamic space mapping. A nonlinear capacitance equivalent circuit of the radio frequency switch is established as a rough model; an input signal and an output signal thereof are obtained through harmonic balance simulation, a time domain dynamic neural network is established, and a prediction network is obtained through training of the signals obtained through harmonic balance simulation; and an input signal and an output signal of the input signal are actually tested, a fine model is established through a prediction network, the signal obtained through actual testing is used for training, and harmonic prediction is achieved through the trained fine model. According to the method, the advantages of high precision of a fine model and high speed of a coarse model are combined, any capacitance effect and non-quasi-static effect missing in static neural space mapping are predicted, and harmonic waves can be accurately predicted compared with an equivalent circuit model.

Description

Technical field [0001] Harmonic prediction method of the present invention belongs to the field of microwave linear modeling, in particular, it relates to a method of harmonic prediction based on silicon-on-insulator dynamic space map neural network field effect transistor (SOIFET) RF switch. Background technique [0002] RF switch is an important part of a typical front-end module, commonly used in cellular phones as an interface between the outside world and the internal RF transceiver and a baseband integrated circuit interface. Driven by Moore's Law, the traditional complementary metal oxide semiconductor technology can not meet the modern switching IC industry for higher integration, higher speed and higher power requirements, silicon-on-insulator field effect transistor with its superior performance It was originally developed for digital applications. However, silicon-on-insulator field effect transistor tends to have higher nonlinearity, when driven by a large excitation ...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the problems existing in the background technology, the purpose of the method of the present invention is to provide a method for harmonic prediction of a silicon-on-insulator field-effect transistor radio frequency switch based on a dynamic space mapping neural network for the current situation that the accuracy of the rough model is not enough. The equivalent circuit model has achieved significant performance and efficiency improvements. Compared with the static space mapping network and the direct use of the neural network to predict the output signal of the switch, the accuracy rate has also been improved.

Images