Research On Adaptive Predistortion For High Bandwidth Wireless Communication System

Power amplifiers (PAs) are indispensable components in wirelesscommunication systems, but are inherently nonlinear. The nonlinearity willdistort the signals, which causes the out-of-band spectral regrowth and in-banddistortion. Therefore, it is necessary to perform PA linearization. Among all theexisting linearization techniques, adaptive digital predistortion (DPD) is themost promising, which has been gained widespread popularity. This thesisfocuses on the study of DPD technique for high bandwidth wirelesscommunication system.First of all, this thesis introduces the existing common used PA models andpoints out the corresponding application field. Secondly, two basic identificationstructure of DPD are described, namely, the direct learning structure andindirect learning structure, then their advantages and disadvantages areexplained. Based on the indirect learning structure, three classical parameteridentification algorithms of DPD are derivated and analyzed, including LSalgorithm, LMS algorithm and RLS algorithm, the performance of the algorithmsin reducing out-of-band spectral regrowth, complexity, convergence and stabilityare also analyzed. Furthermore, considering the problems such as the increasingof bandwidth of modern wireless communication system, uneven distribution ofsignals and feedback branch contain measurement noise that lead to thenon-stationary characteristics of signals, the nonlinear identification algorithmsfor high bandwidth non-stationary signals are studied, including the traditionalNLMS algorithm and GNGD algorithm. A robust adaptive variable step-sizeLMS algorithm is then proposed, simulation shows that comparing with thetraditional identification algorithm, the new algorithm can greatly reduce theout-of-band spectral regrowth, at the same time, it has strong stability. At last,for the performance degradation because of the measurement noise, thesuppression of measurement noise for DPD system is studied, this dissertationput forward a new DPD scheme, this scheme can reducing the effect ofmeasurement noise to parameter identification effectively. Simulation alsodemonstrates the superiority of the novel scheme in reducing out-of-band spectralregrowth and convergence.