Artificial Neural Network-Based Complex Gain Technique for Digital Predistortion of Power Amplifiers

Abstract

This paper proposes an artificial neural network (ANN)-based complex gain technique for digital predistortion (DPD) of power amplifiers (PAs). Existing look-up table (LUT)-based complex gain approach is one of the accurate DPD methods for PA linearization. However, this LUT-based approach consumes lots of memory and the linearization procedure is relatively slow due to the frequently search of the LUT. To address these issues, the proposed ANN-based complex gain technique uses ANN to learn the complicated relationships in the LUT and replaces the LUT with the trained ANN model, effectively minimizing the number of coefficients in the memory and avoiding indexing the LUT. The proposed technique decreases the memory requirement and speeds up the linearization procedure, while maintaining high linearization performance for PAs. The proposed technique is illustrated by two examples, i.e., a Freescale PA and a Doherty PA. The results show that the proposed technique has good linearization capability comparable to the existing LUT-based complex gain approach and can save data storage room as much as 99.65% in the Freescale PA example and 99.87% in the Doherty PA example.