Behavioral Modeling of Nonlinear Power Amplifiers Using Spiking Neural Networks

In this paper, we propose a novel way for power amplifiers (PA) modeling using spiking neurons. The rate of neurons firing spikes is a nonlinear function of its excitation current. Taking the firing rate as the output and the excitation current as the input of a one-layer spiking neuron network, we build up a PA behavioral model with low nonlinearity order to mimic its strong nonlinearity. The results of modeling two Doherty PA show that the proposed method can reach better performance but with lower computational complexity compared with traditional methods. This is the first time that the nonlinearity property of spiking neurons are used for processing such nonlinear signals. Future work is to develop a complete system for the training of the spiking neural networks and to explore the application of spiking neural networks on real-time PA linearization.