Efficient Baseband Digital Predistortion Techniques For Linearizing Power Amplifier By Taking Into Account Nonlinear Memory Effect

Baseband digital predistortion (DPD) techniques for linearizing power amplifiers (PA) with memory effects are investigated in this thesis. Firstly, the relevant elements concerning PA linearization are introduced, such as PA nonlinearity behavior, its influence on the communication systems, its modeling and characterization... Then some existing linearization techniques are presented, such as power backoff, feedforward, feedback, linear amplification with nonlinear components and DPD. DPD is the most promising linearization technique. After that, the implementation architecture and identification algorithms of DPD are described. In this thesis, four DPD methods are proposed. The first method (MP/LUT DPD) is to combine a memory polynomial (MP) model and a simple non-interpolated LUT. Both the amplitude and phase of the predistorted signal are calculated by LUT. The second method is to add linear interpolation technique to MP/LUT DPD. The third method improves the second one by using a quadratic interpolation technique to MP/LUT DPD. The fourth method is to combine MP DPD and feed forward neural network. The principal innovation is that the training samples of the neural networks are the predistorded signal obtained by MP DPD. Finally, simulation results and experimental results are given and analyzed. The proposed methods provide different trade-off between the linearization performance, time efficiency and complexity.