Linear Power Amplifier Digital Predistortion Technology Research

In modern communication system, linear modulation schemes with high spectrumefficiency, such as QPSK, QAM and8PSK etc., have been widely adopted. Theenvelope of the RF signal varies with time. Thus the peak-to-average ratio (PAR) ishigh. The major drawback of the linear modulation techniques is the requirement for RFlinear power amplifier (PA). Intermodulation distortion (IMD) rapidly degrades whenthe signal peaks approach amplitude saturation region, thus requiring some backoff ofthe average power level. However, lower efficiency is obtained as the average power isdecreased. Some high efficient amplifier structure, such as Doherty, is popularlyadopted, together with lineanrizers. Digital Predistortion (DPD) is one of the most costeffective ways among all linearization techniques.Firstly, we introduce the methods of analyzing nonlinear PA. Baseband PA modelcan be derived from RF PA model. We then introduce some baseband PA models, suchas Wiener, Hammerstein and Memory Polynomial etc. We also intrudce the averagepower variation effects on the PA behavior.We focus on studing predistortion parameter extraction process. That is the powerlevel of the data for predistortion model extraction must be normalized to the samepower level of the original input so that the extracted parameters can be directly used inthe predistorter after model extraction. Fistly, we introduce parameter extractionprocedure of indirect learning sructure, then show the parameter extraction process oftwo types of expected gain that is maximum gain type and maximum power type.Finally, We propose the generalized parameter extraction model, and detail discussion ispresented.We also introduce the effect that PAR at predistorter output port is higher thanoriginal input signal. This leads to that the average power of predistorted signal is lowerthan that of training signal. The average power variation on the input of the PA affectsits behavior considerably. This causes mismatch between the predistorter’s nonlinearfunction and the actual inverse PA behavior which greatly depends on its input averagepower. The average power variation will degrade the DPD performance. One solution of the problem is iterative DPD, another one is synergetic Crest Factor Reduction (CFR)and DPD. The design of complex QR dcomposition using FPGA is also intruduced.Finally, we present the predistortion evaluation board based on the Long TermEvolution (LTE) standard. Some design details of signal generator, interpolation filters,digital upconverter (DUC), predistorter and RF mixer modules are presented. Themeasured results of DPD and results analysis are also presented. Measurement resultsshow that about13dB ACPR improvement is obtained when DPD is added.