Physics Informed Spiking Neural Networks: Application to Digital Predistortion for Power Amplifier Linearization

Abstract

Recently, new emerging techniques of neuromorphic hardware render spiking neuron networks (SNN) promising as an energy-efficient solution for artificial intelligence (AI). With the idea of physics informed neural network, the structure can be simple while training data can be light. However its application in RF telecommunication system is still challenging. This paper, as the first time in the literatures, proposes a solution of SNN-based digital predistortion (SNN-DPD) for linearization of RF transmitters, such as power amplifiers (PA). A two-layer SNN is deployed in frequency domain to process the spectrum of the stimulus for a predistorted signal. The proposed technique is experimentally validated on a test bench with a real PA of different bias voltages. We also test the proposed SNN-DPD for multi-band linearization. The proposed method reaches the best performance of traditional DPD methods while owing advantages of the SNN, such as low power consumption and good biomimicry for AI.

Keywords

• Artificial intelligence (AI)
• digital predistortion
• frequency domain
• linearization
• physics informed
• power amplifiers