Bandwidth-Scalable Digital Predistortion Using Multigroup Aggregation Neural Network for PAs

Abstract

A multigroup aggregation neural network (MGANN) model for bandwidth-scalable digital predistortion (DPD) is proposed. The MGANN model introduces a multinetwork structure based on the characteristics of neural networks (NNs) to broaden the bandwidth application range and eliminate the updates online. The proposed structure combines the input layer and the first hidden layer into multiple networks retrieved by means of inertia coefficients. In addition, to improve modeling accuracy, a new input vector is used by introducing the product term of I/Q components and the amplitude of the signal. The experimental results indicate that the proposed model can significantly improve the adjacent channel power ratio (ACPR) within the range of 20–200M with an average of 12.1 dB compared with traditional GMP models when using a fixed set of parameters.