Noise Modeling and Compensation of Low-Cost IMU Based on Allan Variance-Transformer Model

Abstract

Low-cost inertial measurement units (IMUs) were widely adopted in navigation, robotics, and consumer electronics due to their cost-effectiveness and compact size, but their high noise levels severely degraded measurement accuracy. A novel noise modeling and compensation method was proposed in this paper, combining Allan variance analysis with a Transformer model. The noise model was constructed using static gyroscope data, and online compensation was achieved. Experimental validation was performed based on static gyroscope data extracted from CSV files, demonstrating the feasibility of the proposed method in noise modeling and filtering. The method not only utilized Allan variance to reveal noise characteristics but also leveraged the Transformer’s sequential modeling capability to improve accuracy, providing a new solution for low-cost IMU noise compensation.