Research On Power Amplifier Behavior Modeling Method Based On Generative Adversarial Network

With the advancement of the times,Internet technology is more and more integrated into people's daily lives.With the requirement of network video,audio and digital communication technology in communication transmission speed,the requirements are becoming higher and higher.The wireless communication system technology has also undergone earth-shaking changes.As one of the indispensable and important components in the wireless communication system,the RF power amplifier,when the RF power amplifier works near the maximum output power,it will produce serious nonlinear characteristics,which will affect its output power and efficiency,At the same time,it will show a memory effect.Accurately modeling the behavior of RF power amplifiers for its nonlinearization and predistortion research and its implementation is of great theoretical significance and great significance for promoting the in-depth development of green communication theory,energy saving and emission reduction.Research on practical application value.At present,a data modeling method called generating an adversarial network,because it can retain the features related to the training data from the noise,after training the network generator and discriminator,you can generate data samples that are similar to the real sample data distribution As more and more variants of generative adversarial networks are proposed,they have achieved a good effect on the stability of training,the speed of convergence and the effect of generating data,which have been applied to modeling in many fields.But it has not been applied to the modeling of RF power amplifier.This article first introduces the nonlinear characteristics of RF power amplifiers and commonly used modeling methods,highlights several commonly used behavioral models of memory RF power amplifiers,analyzes the advantages and disadvantages of various models,and then introduces the generation of adversarial networks.A concept and design idea,starting fromthe original GAN to the GAN network variant to introduce the mathematical model and algorithm flow of the network.In view of the nonlinear characteristics of RF power amplifiers,two modeling methods for generating adversarial networks(GAN)are introduced.The generator in the GAN network uses a recurrent neural network(LSTM)structure.The output of the network is not only related to the current input,but also It is related to the network output of the past moment,and can be used to describe the memory effect of the RF power amplifier.Based on two different GAN network variants,WGAN-GP and SNGAN,respectively modeled the behavior of the RF power amplifier,and analyzed the experimental results of the two models,which confirmed the modeling of the power generation method introduced in the paper Feasibility.The experimental results of the two GAN variant modeling methods show that although the model is not as accurate as the traditional power amplifier model,there is room for improvement.Both models can describe the nonlinear characteristics and memory effects of the power amplifier.SNGAN is superior to WGAN-GP in terms of model accuracy,training stability and convergence speed.