Advanced non-linear model for accurate prediction of harmonically terminated power amplifier performance

This paper presents recent advances in the state-of-the-art of neural network modeling of microwave FET devices. Enhanced accuracy of the adaptive knowledge-based neural network (AKBNN) model is shown by comparing predicted load-pull performance of the device to measurements in an automated harmonic load-pull system. Test devices are a 1.2 mm HFET measured at 2.2 GHz, and a 4.8 mm pHEMT at 8.4 GHz. Modeled versus measured comparisons include power-added efficiency and output power under fundamental frequency and second and third harmonic frequency tuning. The effectiveness of this modeling approach for the design of high-efficiency power amplifiers operating in Class-E or Class-F modes is discussed.