Comparison of direct learning and indirect learning predistortion architectures

Abstract

Power amplifiers in a communication system are inherently nonlinear. Digital predistorters can compensate these nonlinearity effects. In this paper, two memory polynomial predistorters including direct and indirect learning architectures are compared with each other. To the best of our knowledge, no similar comparisons have been published. Both of these architectures are special cases of the self-tuning control. We have modeled predistorters and analysed nonlinear effects of a power amplifier and their digital compensation by using Matlab¿. Simulation results show that the memory polynomial model has convergence problems at large amplitudes and also problems of accuracy of representation. We observed that the results of the compensation depend also on the amplitude, not only on the frequency. The results of the linearisation show that the direct learning architecture achieves a better performance in almost all cases.