A new variant of the indirect learning architecture for the linearization of power amplifiers

Abstract

The indirect learning architecture (ILA) is the most commonly used technique for the identification of digital pre-distorters for power amplifiers (PA). A critical issue in ILA is the selection of the normalization gain used to synthesize the predistorter function. In this paper, we investigate the effects that the normalization gain has on the average output power and linearity of PAs. Moreover, we propose a new variant of the ILA that eliminates the need of a normalization gain inside the iterative loop. Experimental results show that the selection of the normalization gain affects the average output power and consequently the linearity performance of the linearized PA. If the normalization gain is not chosen correctly, the average output power of the linearized PA will differ from the average output power obtained before DPD. It is experimentally shown that the proposed ILA variant can maintain the same average output power before and after DPD. Consequently the proposed ILA variant simplifies the linearization process and allows proper evaluation of the DPD performance.