Behavioral Modeling Of RF Power Amplifiers Using Extreme Learning Machine

With the recent upsurge of the wireless communication system,especially in advanced commercial applications such as cellphone,WLAN and navigation system,as well as traditional military applications such as radar,electronic countermeasures equipment and intersatellite communication,as one of the most important components in the wireless communication system,the radio frequency(RF)front end of wireless transmitters has been attracting more and more attention.Based on it,radio frequency power amplifiers(RF PAs)are playing more and more critical roles as one of pivotal components in the RF front end.In a variety of applications of the RF PAs,the power amplifier(PA)introduces nonlinearities when it operates near maximum output power.However,the nonlinearities will have a direct effect on the power and efficiency of the PAs.Hence,it becomes a key point to analyze and predict the nonlinear characteristics of RF PAs.And the behavioral modeling is used to accomplish the work.Therefore,modeling the behaviors of RF PAs is the key for the success to predict the performance of the applications.Nevertheless,the progresses of PA behavior modellings have been strongly relying on the advances of mathematical modeling.In recent years,due to the fast operation and good modeling effect,an advanced modeling technique called extreme learning machine has been extensively used in many good applications.But it has not been introduced in the behavioral modeling of circuit.Thus,this paper will do a detailed study on the application of the algorithm in the direction of power amplifier behavior modeling,in order to analyze and predict the nonlinear characteristics.The main contents are as follows:This paper introduces nonlinear characteristics and common behavior models of RF power amplifiers.And a more advanced model named extreme learning machine is introduced for nonlinear characteristics of RF power amplifiers.Then,according to different memory effects of PAs,concrete modeling methods based on extreme learning machine algorithm are given.In addition,this paper presents a new piecewise extreme learning machine modeling method,which further improves the accuracy of the model based on the extreme learning machine.The modeling method of piecewise extreme learning machine divides the modeling area into several sub-areas.Each sub-area is modeled using the extreme learning machine algorithm.This paper gives a detailed process,which is a theoretical preparation for the establishment of the piecewise extreme learning machine model.The paper validates the modeling methods mentioned in this paper by two different examples of RF power amplifiers,and illustrates the feasibility of the modeling method of extreme learning machine and the piecewise extreme learning machine.Based on Monte Carlo simulation,the modeling results of extreme learning machine are compared with those of other commonly used artificial neural networks.The results show that the former is faster than the latter by three orders of magnitude.