Linearization of Power Amplifier using Digital Predistortion, Implementation on FPGA

by Erik Andersson

Institution: Linköping University
Year: 2014
Keywords: Digital predistortion; Power amplifier; Nonlinear models; FPGA; Linearization; Jamming; Engineering and Technology; Electrical Engineering, Electronic Engineering, Information Engineering; Embedded Systems; Teknik och teknologier; Elektroteknik och elektronik; Inbäddad systemteknik; Signal Processing; Signalbehandling; Elektroteknik; Electrical Engineering
Record ID: 1365808
Full text PDF: http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-112258


The purpose of this thesis is to linearize a power amplifier using digital predistortion. A power amplifier is a nonlinear system, meaning that when fed with a pure input signal the output will be distorted. The idea behind digital predistortion is to distort the signal before feeding it to the power amplifier. The combined distortions from the predistorter and the power amplifier will then ideally cancel each other. In this thesis, two different approaches are investigated and implemented on an FPGA. The first approach uses a nonlinear model that tries to cancel out the nonlinearities of the power amplifier. The second approach is model-free and instead makes use of a look-up table that maps the input to a distorted output. Both approaches are made adaptive so that the parameters are continuously updated using adaptive algorithms. First the two approaches are simulated and tested thoroughly with different parameters and with a power amplifier model extracted from the real amplifier. The results are shown satisfactory in the simulations, giving good linearization for both the model and the model-free technique. The two techniques are then implemented on an FPGA and tested on the power amplifier. Even though the results are not as well as in the simulations, the system gets more linear for both the approaches. The results vary widely due to different circumstances such as input frequency and power. Typically, the distortions can be attenuated with around 10 dB. When comparing the two techniques with each other, the model-free method shows slightly better results.