Use of Statistical Signal Properties for Adaptive Predistortion of High Power Amplifiers

Abstract

One of the key issues of Digital Radio Mondiale (DRM) is green broadcasting. For wide area coverage, the use of high-power transmitters is essential. However, the applied transmission technology based on Orthogonal Frequency Division Multiplexing (OFDM) results in non-linearities in the emitted signal, low power efficiency, and high costs of transmitters. Digital predistortion is a promising scheme for power amplifier (PA) linearization. This paper presents an efficient approach to estimate the parameters of a digital predistorter based on adaptive filtering with direct learning architecture (DLA). A well-known algorithm for identifying and tracking the time-varying parameters of an unknown system is the recursive least squares (RLS) method with exponential/directional forgetting. In this paper, the efficiency of both exponential/directional forgetting techniques is investigated for different degrees of PA nonlinearities. On this basis, a new hybrid technique based on statistical properties of the PA input signal is proposed. The evaluation results show that for both scenarios, the statistic-based forgetting technique not only provides better accuracy but also is more robust against high PA nonlinearities.