Nonlinear modeling of GaN Doherty power amplifiers using radial-basis function neural networks

A gallium nitride Doherty power amplifier (GaN Doherty PA) was designed for 2.5 GHz WiMAX band, and a radial-basis function neural network (RBFNN) model is proposed for predicting this amplifier´ nonlinear characteristics. Comparison of AM/AM, AM/PM, PAE and Pout curves between the RBFNN model and circuit simulation are given. After 125 epochs, the convergence of this RBFNN model becomes slower and the training error reaches to a lower value (below 1% error level). The results indicate that the proposed RBFNN model can reproduce the nonlinear transfer characteristics well.