Behavioral Modeling of a Multimode Power Amplifier Using Perceptrons with AM-AM and AM-PM Complex-Valued Activation Function

Abstract

Three-layer perceptron (TLP) is one of the approaches to the behavioral modeling of radio frequency (RF) power amplifiers (PAs) for wireless communication systems. A low-pass equivalent PA behavioral modeling should be able to represent the AM-AM and AM-PM characteristic curves, besides generating only useful contributions. This work shows that adapting a complex-valued activation function toencompass the strict characteristics from the AM-AM and AM-PM curves at the same time fulfilling the bandpass constraint gives better results in terms of network complexity and accuracy. Besides that, a novel TLP-based model for a multimode PA is proposed. This type of PA has characteristics that add difficulty to its inverse modeling. A modification in the ascendant method is also proposed for application with TLP-based PA models. This method simplifies the models in terms of operations needed for implementation. In a scenario of same number of network parameters, a significant reduction in modeling error is achieved when a complex-valued TLP that accounts for both AM-AM and AM-PM is used instead of a complex-valued TLP that accounts only for AM-AM. Such reduction is quantified by up to 5.73 dB (-35.53 dB to -41.26 dB) improvements in normalized mean square error (NMSE) metric. Moreover, when applying the ascendant method a 67% reduction in active parameters is achieved (36 to 12 active parameters), deteriorating the NMSE by less than 0.5 dB.