Study of Spiking Neural Network-Based Regressor on Applications in Digital Predistortion for Power Amplifiers

Abstract

Digital predistortion (DPD) technology linearizes power amplifiers (PAs) so that they can operate in a high-efficiency region. The estimation of the coefficients for the DPD model is therefore crucial but resource-intensive. Meanwhile, spiking neural networks (SNNs) are considered to have advantages in energy efficiency due to their ability to closely mimic the activity of human brain neurons. In this paper, we introduce a novel approach to estimate the coefficients of the DPD model. This approach uses a regressor with a structure that combines SNN and artificial neural network (ANN), along with two different loss functions. The proposed method is evaluated with datasets measured on a strongly nonlinear PA with a peak output power of 200W. The results show that we can achieve a good fit of generalized memory polynomial (GMP) based DPD coefficients with the proposed method. To the best of our knowledge, this is the first time that the SNN is used in computing DPD coefficients. This study offers valuable insights into the potential of SNN-based wireless communication technologies.