Augmented radial basis function neural network predistorter for linearisation of wideband power amplifiers

Abstract

An augmented radial basis function neural network (ARBFNN) is proposed for modelling and linearising a wideband Doherty power amplifier (DPA) with strong memory effects and static nonlinearity. To evaluate the performance of the ARBFNN, a 51 dBm DPA and a 25 MHz mixed test signal were used in modelling and linearisation measurement. Compared with the memory polynomial (MP) model and the real-valued time-delay neural network (RVTDNN), the ARBFNN is highly effective, leading to 3 and 5 dB improvements in the normalised mean square error. More importantly, the ARBFNN predistorter represents a significant improvement over the RVTDNN and MP in the suppression of the out-of-band spectral regrowth. In addition, the ARBFNN has a similar linearisation capability as the generalised MP model, but has much better numerical stability. © The Institution of Engineering and Technology 2014.