Implementation And Application Of HPA Predistortion Using The Principle Of Neural Network

Because of the inherent characteristics, HPA may cause the signal amplitude(AM/AM) and phase distortion(AM/PM) which will lead to expansion of the spectrum and interference of adjacent channel signal in the process of amplification in CMMB repeater.In recent years, as the further study of neural networks, people has payed more and more attention to the neural network predistortion technique. The purpose of the paper is to use neural network predistortion to effectively reduce the nonlinear distortion of the HPA in CMMB repeater, and improve the efficiency of the HPA.Firstly, the research background and the present research status were summarized. It indicated that the digital predistortion had great potential to develop in theoretical and practical research. Then, the basic knowledge of the nonlinear characteristics of HPA, the basic principle of digital predistortion and the HPA behavioral models were introduced briefly. Furthermore, the Saleh model was chosen as the HPA behavioral model in this paper.Secondly, several frequently-used learning algorithms such as GD, GDM, CGF, BFG and LM algorithms were presented. Considering the convergence speed and complexity of these algorithms, LM algorithm was chosen as the learning algorithm of the neural network models in this paper. In addition, in order to eliminate the over-fitting phenomenon of LM algorithm, bayesian regularization algorithm was introduced. Then, a novel neural network(FIR-NLNNN) was proposed, which can respectively process the memory effect and the nonlinearity of HPA. Simulation results show that the proposed method can effectively eliminate the memory effect and nonlinearity of HPA and reduce ACPR(adjacent channel error power ratio) by about 30dB. Compared with the traditional neural network(RVTDNN) under the same performance conditions, the coefficient of FIR-NLNNN is reduce by about 50%. Similarly, the times of multiplication and addition in the iterative process of FIR-NLNNN are reduced by 75%.Based on the MATLAB simulation, the FIR-NLNNN model and gradient descent algorithm were implemented by System Generator and synthesized by ISE. The results show that the Slices resources consumed by FIR-NLNNN are about 67% of those consumed by RVTDNN, and the DSP resources used by FIR-NLNNN are about 65%of those used by RVTDNN. Simulation results show that the HPA model using FIR-NLNNN is good for fitting the nonlinearity and memory effect of HPA, and the HPA predistortion using FIR-NLNNN can effectively eliminate the memory effect and nonlinearity of HPA and reduce ACPR by about 20dB.At last,the CMMB repeater station system was constructed completely. Compared with the feed-forward signal, the feedback signal had some delays. The synchronization of the signals was the premise to realize the HPA predistortion. In this paper, the right loop-delay value was estimated using the improved correction method detecting pseudo random numbers. Furthermore, in order to build the test platform of the hardware, the test signal of CMMB was designed and implemented.