Application of Neural Networks to Reduce Distortion of RF Signals in Switch Mode Power Amplifiers

Abstract

The paper covers the problem of reducing distortion of switch mode power amplifier (SMPA) output voltage using a neural network. Signal distortion in SMPA is mainly caused by transistors’ parameters dispersion. In multi-cell SMPA being widely used in envelope tracking power supplies (ETPS) significant effect on signal distortion is also produced by deviation of parameters of combiner elements. To reduce signal distortion an adaptive adjustment by neural network of the width of control pulses’ being applied to the transistors’ gates can be used. Using this approach for lowering even harmonics in the RF carrier generated by SMPA, the 2nd harmonic was reduced by 10$\ldots$15 dB. The created neural network showed its ability to operate with both a linear approximation of the rise and fall time intervals of transistors and a quasi-sinusoidal one. This allows us to propose the admissibility of the neural network use in case of rather complex approximation of SMPA output voltage at the switching time intervals. Proceeding from the need to increase the efficiency of the PAs, the use of switch modes is of increasing interest.