Application of a Q-Improved Kalman Filter in Non-Gaussian Noise Sensor Signal Acquisition

Abstract

In many fields of modern engineering activities, such as industrial automation, robotics, and intelligent vehicles, the reliable acquisition of sensor signals is crucial. The traditional Kalman filter, with its strong computational power and mathematical rigor, is usually considered effective when the sensor noise is Gaussian distributed and has constant statistical properties. Compared to traditional state estimation methods, in real sensor environments, examples of non-Gaussian, heavy-tailed, and impulsive noise are often more prevalent. This paper introduces an improved Kalman filter for non-Gaussian noise environments commonly encountered in sensor network-based applications. Robust innovation domain processing and adaptive statistical transformation can simultaneously eliminate the impact of outliers while mitigating heavy-tailed interference. The evaluation results using synthetic and real datasets indicate that the modified method has higher estimation accuracy and lower sensitivity to non-Gaussian outliers. Compared to classical and more stable robust extensions, this system can achieve higher accuracy for instruments with stringent real-time requirements. It also has stronger adaptability (to noise variations). From a scientific and engineering perspective, the improved Q Kalman filter provides a solid foundation for the stable operation of sensors in uncertain and harsh environments, as well as for precise measurements in rich environments.