Power amplifier linearization-using an indirect-learning-based inverse TDNN model

Abstract

In this article, an effective digital predistortion procedure of data acquisition, analysis, modeling and linearization for high power amplifiers is presented, based on tapped delay neural networks (TDNN). Memory effects of high power RF amplifiers were identified and modeled using baseband signal analysis. The output RF signal of the amplifier was converted down to the baseband,in-phase (I) and quadrature (Q) signals, which were sampled for modeling and linearization in the digital domain. Behavioral modeling was carried out using a well-known neural network algorithm including delay taps in order to consider the memory effects of high power amplifiers. Based on the behavioral model of the power amplifier, an indirect learning process for linearization was constructed, using an inverse TDNN model. Compared to the memoryless predistortion process without tapped delay lines, the proposed predistortion method showed an approximately 15 dB better adjacent-channel leakage ratio (ACLR), which was about 60 dBc for a WCDMA downlink signal.