Behavioral Modeling Of RF Power Amplifier Based On Recursive Polynomials

Behavioral modeling of RF and microwave power amplifiers is of great interest in system levelanalysis, simulation and design of modern wireless system, especially when excited by widebandlarge signals as in 3G, WiMax and LTE systems. Tremendous research efforts have been paid forthis topic in the last decade. However, despite some recent encouraging advancement, the accuratemodeling of RF power amplifiers driven by wideband signals with high peak-to-average powerratios (PAPR) is still deemed a complicated task.The fundamental difficulty for the RF power amplifier modeling under large wideband signalexcitation is the simultaneous existing of the nonlinearity and the memory effects with differentlengths. Popular approaches include the Volterra series, the memory polynomial, the neural networkand the neuro-fuzzy system, etc.This paper describes and discusses the nonlinearity and distortion caused by distortion of RFPA, the memory effects and how to build behavioral model, and then, analysis the currently existedmodeling methods and their advantages and disadvantages.The main contribution of this paper is the introduction of orthogonal least squares algorithm todetermine the structure of the recursive polynomial power amplifier model. Compared to Volterraseries and polynomial model, recursive polynomial structure has fewer parameters. Meanwhile, therecursive polynomial power amplifier model can be identified by linear identification procedure,avoid the nonlinear optimization or training process, reduce the amount of computations comparedwith neural network model. Since it is unlikely that all the terms contribute identical to the modelbehavior, this paper introduces the orthogonal least squares algorithm to determine the structure ofthe recursive polynomial power amplifier model. The use of recursive structure and OLS makes themodel simpler. Finally, this paper imposes the steady-state data as constraint condition to make themodel more accurate. At the last of paper, several RF PAs are chosen for simulation and test, Firstly,exact the datas to train the recursive polynomial power amplifier model built in MATLAB. Thendifferent tests of the model will be made, including the transient and steady-state waveform, thespectrum, the power compression curve envelop waveform, etc From the results, we can see thegreat match between the model and real data, the most important is that the model has simplerstructure and fewer parameters. This model will be useful in communication system research.