Research On The Predistortion Technology Of Wideband Power Amplifiers

The demand of higher spectral efficiency forces wireless communication systems employing modulation and transmission schemes which have high spectrum efficiency (such as M-QAM, OFDM, WCDMA etc.).The signals of these schemes have the same features including high envelop fluctuating and high peak-to-average ratio.When these signals pass through PAs,the nonlinearities of PAs cause both a distortion of the signal and an increased out-of-band output spectrum, which leads to a rise in bit error rate and adjacent channel interference.To obtain high efficiency and avoid severe distortion caused by PA nonlinearities, a linearization technique is required. Among linearization techniques of PAs, predistortion technology attracts more attentions due to its low complexity, high stability,wide bandwidth and good performance.With the development of the wireless communications, the data rate increases continually, which leads to the increase of bandwidth of PAs. For wider bandwidth applications, PA memory effects can no longer be ignored.This paper conduct a deeply research on the predistortion technology with memory. The research includes structure of predistortion system, the behavioral models of PAs,predistortion technology for PAs with memory, the determination method of the nonlinearity order K, the memory order Q, and expected linear gain etc.Following four aspects are originality innovation discoveries:1.The single feedback path offline learning structure and two feedback path offline learning structure are proposed based on the research of predistortion structure.In the two feedback path offline learning structure, the parameters of the predistorter can be directly extracted from an offline system identification process.This eliminates the usual requirement for a closed-loop real-time parameter adaptation, which dramatically reduces the implementation complexity of the system.Two feedback path offline learning structure increases the cost of the hardware due to the additional feedback path.To solve this problem, a single feedback path offline learning structure is proposed. Both of the proposed learning structures are suitable for several implementation structures.2.A dynamic memory polynomial predistortion algorithm is presented based on the research of the behavioral models of PAs. Compared with the traditional memory polynomial predistorter, the dynamic memory polynomial (DMP) predistorter can reduce the number of the coefficients efficiently (up to 65%) without decreasing the linearization performance.Simulations results show that the proposed DMP predistorter is robust and performs well.Experimental results show that ACPR achieved by the DMP predistorter is almost the same as that achieved by the traditional memory polynomial predistortier (difference in ACPR is less than±0.5dB),however, the number of DMP predistorter coefficients is reduced by 46.7%.3.The order-recursive least square method is extended to complex number applications, and is used to determine the nonlinear order K and memory order Q of a predistorter. A KQ order-decision method is proposed to find the optimal nonlinearity and memory order of predistorters corresponding to a PA with unknown characteristics.This method makes a predistorter more intelligent and universal because it can adjust the K and Q parameters adaptively depending on different PAs. Finally, the proposed approach is verified by simulations and experiments.4.A gain-adjustable predistortion algorithm is proposed.The range of the expected linear gain and the optimum gain is also detailed. This algorithm can change the gain of PA by adjusting the expected gain of a predistortion algorithm. Moreover, when a PA has a large input, the linearization performance can be improved by decreasing the expected gain. Simulations and experimental results validate the algorithm.