Study Of Behavioral Modeling And Predistortion Techniques For Class D Power Amplifier

Class D power amplifiers (CDPAs) are important components in electronic system, and CDPAsare inherent nonlinear characteristics. The frequency spectral growth beyond the signal bandwidth iscasued by CDPA’s nonlinearity， which would interfere with adjacent channels. This thesiscomprehensively studies the following contents including CDPA’s specific circuit; CDPA’s nonlinearcharacteristics and its distortions; CDPA’s memory effect and its form; CDPA’s behavioral modelingand predistortion techniques.Firstly, the specific circuit of the acoustic transmitter is provided. The circuit mainly consists offive sub-module circuits, which are pulse width modulation circuit; dead-time control circuit; CDPAcircuit; matching network and transducer. The realization methods and simulation results of fivesub-module circuits are given respectively.Secondly, the concept of CDPA’s nonlinear characteristics, all kinds of distortion products causedby nonlinearity, memory effect and its forms are introduced. The CDPA’s nonlinear characteristics,including AM-AM and AM-PM characteristics, envelope characteristics, harmonic distortion,intermodulation distortion and transition characteristics, are simulated and analyzed using Pspicesoftware. All nonlinear characteristic curves obtained from Pspice software have proved that CDPA isthe nonlinear system. Analysing the major factors, including dead-time, MOS tube gate-sourcecapacitance, power supply ripple and matching network, which mainly affect CDPA’s nonlinearcharacteristics.Thirdly, two improved neural network model are proposed, which are the improved BP neuralnetwork (IBPNN) model and the improved Elman neural network (IENN) model. Some conclusionscould be known by analyzing two improved neural network model simulation results. First of all, theIBPNN model could guarantee convergence accuracy and have fast convergence rate, which meansthat the model results could be solved by one-step weight determination. Then, the error function ofIENN model could be accurate to10-8when the iterative times is100, which means that theconvergence precision of IENN model is high and its convergence rate is fast. At last, the simulationresults show that these improved neural network models could well describe CDPA’s nonlinearcharacteristics; what’s more, the IENN model also well depicts CDPA’s memory effect.Lastly, the digital predistortion techniques realization method is provided.