Behavioral Modeling And Digital Predistortion Of Nonlinear Power Amplifiers With Memory Effects

Power amplifiers (PAs) are intrinsically nonlinear devices and also exhibit memory effects when applied to wideband wireless communication systems. The nonlinearities as well as the memory effects of PAs cause distortions to the transmitted signals, which decrease the qualities of communication systems by increasing bit error rate and also produce spectral regrowth interfering with the adjacent channels. In order to analyze, estimate and mitigate the effects of nonlinear PAs with memory effects on the wireless communication systems, the research on behavioral modeling and digital predistortion (DPD) of PAs has become a hot topic in the field of wireless communication.In recent years, many researchers all over the world have carried out a lot of researches on the behavioral modeling and digital predistortion of PAs, and various behavior models have been proposed. These works contribute significantly to addressing the key issues relating to wireless communication and promoting the development of wireless communication systems. However, with the rapid and sustained development of modern communication techniques, both the inner structures and the working conditions of PAs are becoming more and more complex. Accordingly, it’s much more difficult to accurately describe the nonlinearities and memory effects of PAs. In fact, most of the existing behavioral models have certain shortcomings, so it’s necessary to conduct further research to improve the performance of the behavioral models.Studies on behavioral modeling and digital predistortion of nonlinear power amplifiers with memory effects focuse upon the construction of behavioral models and parameter identification algorithms. In this paper, further researches are carried out to improve the performance of behavioral models of PAs and DPD. The contents of the paper are as follows.1. Memory polynomial (MP) is a reduced version of Volterra series and consists of only the diagonal terms of Volterra kernels, which limites its ability to precisely describe the memory effects of power amplifiers (PAs). In order to enhance the modeling performance of memory polynomial by improving its memory structure, this paper proposes a novel behavioral modeling approach which incorporates the most effective cross-terms in memory polynomial (MPM) by using an ad hoc evolutionary algorithm (EA) in conjunction with an effective parameter identification method. This new proposal takes advantage of the relative simplicity and high efficiency of MPM for PA behavioral modeling, whereas incorporation of the cross-terms further enhances modeling performance. Experiment and simulation results demonstrate the superior performance of the proposed model.2. One problem associated with polynomial based power amplifier behavioral modeling and predistortion is numerical instability, which arises from bad conditioning of the regression matrix. In order to address this problem, a robust coefficient identification algorithm is proposed, which is based on Tikhonov regularization and QR decomposition of the regression matrix. This algorithm can both alleviate the numerical instability problem and limits the peaks of the predistorted signals, and so the digital predistorter based on the proposed algorithm is robust.3. An orthonormal Hermite polynomial based neural network (OHPBNN) is proposed for accurate power amplifier (PA) modeling and digital predistortion (DPD). The proposed OHPBNN exhibits much better approximation performance than the feedforward neural networks employing sigmoidal function in hidden neurons. We then enhanced the OHPBNN PA model by takeing into account the characteristics of the amplitude modulation to amplitude modulation and amplitude modulation to phase modulation distortions which were incorporated into the behavioral model by means of using two set of dynamic fuzzy weights.4. Local model networks based power amplifier behavioral modeling and digital predistortion was proposed. Local model networks have the advantages of simplicity, adaptation and flexibility, etc., and is capable of model various complex dynamic nonlinear systems. Local model networks can be adaptively constructed by heuristically partioning the input space. However the parameter identification with weighted least squares (WLS) only result in a sub-optimal behavioral model. To abtain an optimal model, the overall parameters were also updated at each iteration. Whatmore, by clipping some neglectable local models for each input, a parsimounious local model network is obtained.