| To solve the contradiction between linearity and efficiency of radio frequency(RF) power amplifiers (PA), linearization techniques are well developed. Digitalpre-distortion (DPD) is the mainstream of linearization techniques for its advantages,such as low cost, good effect and flexible structure.In this thesis, the DPD technique is studied and a DPD system is implemented inFPGA. Firstly, upon deeply understanding the characteristics of PA and the principle ofDPD, the author thinks that the study of DPD consists of three parts, including learningstructures, predistorter models and adaptive algorithms. Indirect learning structure isused widely because it can design the predistorter without modeling the PA. Based onindirect learning structure, two DPD schemes are designed, simulated and verified. Theone scheme is lookup table (LUT) predistorter using least mean square (LMS)algorithm, and the other scheme is memory polynomial predistorter using recursiveleast square (RLS) algorithm. According to MATLAB simulation results, advantagesand disadvantages of the two schemes are compared. Moreover, the VSG-FSA (VectorSignal Generator-Frequency Spectrum Analyzer) test platform is constructed to verifyeffects of the two schemes with off-line training.Finally, DPD based on LUT is implemented without controller in FPGA. Thehardware platform and the software one of DPD system are introduced. The keymodules of DPD system, including LUT, LMS, data copy, time delay estimation, phaseestimation and1/G (desired gain) estimation et al, are also described in detail.Additionally, debug process of DPD system is presented. After debugged successfully,a small signal amplifier and a high PA are used to test the effect of DPD. Themeasurements are analyzed and the advice of improvements is given. IM3is improvedby25-30dB under two-tone testing; ACPR is improved by17-18dB under WCDMAtesting. |
PDF Full Text Request