Behavioral Modeling Of GaN RF Power Amplifier Based On ELM

In wireless communications,RF power amplifiers play an important role as an indispensable core module in the RF front-end.The reason for this high-profile role of PAs is not only the activity of PA to boost the radio signal to a sufficiently transmission power level,but also that it is the major source of signal distortion and spurious signal generation,harmonics and inter-modulation products.Quantifying and predicting the impact of these detrimental effects of PA nonlinearity on the transmitted signals requires accurate models of PAs.Additionally,an essential first step is to model PAs nonlinearity accurately in studying a linearization technique(e.g.digital pre-distortion)for PAsThe behavioral model is an important means to characterize the nonlinear characteristics of power amplifiers,and has long been widely concerned by radiocommunication scholars.Artificial neural networks are the most widely used in the behavior modeling of power amplifiers because of their powerful learning capabilities.Nevertheless,given that ANNs can be trained to fit arbitrary nonlinear relationships theoretically.ANNs of different types have been successfully applied for behavioral modeling RF PAs,including RNNs,and DNN.However,these models have complex multi-layer neural networks that require manual intervention and relatively long training.Recently,a single hidden layer feedforward networks(SLFNs)called extreme learning machine(ELM)was applied the behavior modeling of power amplifiers.Thanks to the nature of ELM,there is no need for much manual intervention to determine the network structure and adjust the parameters.Besides,ELM is dramatically faster than other ANNs in training neural networks,because of lack of iteration in the algorithm of ELM.However,the accuracy of ELM modeling is too large in repeated experiments.Furthermore,ELM require large number of hidden nodes in the network construction due to the process of randomly generating the input parameters in the initial step.To solve the drawbacks mentioned above,an Evolutionary Extreme Learning Machine(E-ELM)is introduced to model the behavioral performance of RF PAs.This paper has done a detailed study on behavioral modeling of RF Power Amplifier based on E-ELM.The main contents are as follows:Firstly,in order to solve the problem of large range of model accuracy in ELM modeling experiment.Introduce Differential Evolution(DE)into E-ELM to search for the global optimal input weight and bias of ELM.The model is built with the data of class AB,class E and class F amplifiers,and the modeling and prediction ability of E-ELM for different amplifiers is verified.Secondly,this paper explores the design and optimization of E-ELM assisted Doherty power amplifier.Using CREE 10 W Ga N HEMT transistor and based on E-ELM algorithm,the model of carrier amplifier and peak amplifier was established,and the model was used to guide the matching and optimization of Doherty output partial fundamental impedance.Finally.a 2.4GHz Ga N Class-F^-1/F Doherty power amplifier was designed.It is proved that the behavioral modeling can guide and assist the design of Doherty amplifier.Based on the circuit test,the applicability of E-ELM algorithm to Doherty amplifier is further verified.