Neural Network Based Linearization of RF Power Amplifiers using In-Situ Device Temperature Measurement

RF Power Amplifiers are still a bottleneck and challenging topic for future mobile basestations. Efficiency, bandwidth and flexibility are parameters that can be potentially enhanced. Nonlinearities are one of the most challenging characteristics and not yet fully described or known. They directly influence the performance and especially the efficiency of Power Amplifier (PA) devices. While static nonlinearities can usually be measured and evaluated, dynamic non-linearities like thermal or aging memory effects can be hardly measured and compensated in time. This work shows a method to compensate thermal memory effects in a 20 W GaN Class ABJ Power Amplifier by processing the in-situ temperature of the transistor measured close to its channel. The device temperature is used for a neural network based linearization approach.