Autonomous Calibration of MEMS Gyros in Consumer Portable Devices

You Li, Jacques Georgy, Xiaoji Niu, Qingli Li, Naser El‐SheimyIEEE Sensors Journal 2015 · DOI · arXiv

Abstract This paper presents a real-time calibration method for MEMS gyroscope sensors in consumer portable devices. The proposed method operates autonomously without requiring external equipment or specific user intervention. By utilizing multi-level constraints—including pseudo-observations, accelerometer and magnetometer measurements, and quasi-static attitude updates—the algorithm achieves reliable and accurate calibration under natural user motions. Walking tests conducted with Samsung Galaxy S3 and S4 smartphones demonstrate that the method effectively reduces gyro bias even in challenging motion modes (e.g., dangling, pocket) and indoor environments with significant magnetic interference.

Introduction

Low-cost Micro-Electro-Mechanical Systems (MEMS) inertial sensors are essential for navigation in consumer electronics. However, these sensors are inherently noisy, suffering from significant run-to-run biases and thermal drifts. In indoor environments where Global Navigation Satellite System (GNSS) signals are unavailable, navigation relies on "dead reckoning," where sensor errors accumulate rapidly, leading to substantial positioning inaccuracies.

Traditional calibration techniques are impractical for average consumers as they often require the device to be perfectly static or rotated through specific orientations on high-precision equipment. This research seeks a "set it and forget it" solution: an algorithm that runs in the background and calibrates sensors while the user performs everyday activities like walking.

Methodology

The core of the solution is a Kalman filter that integrates several novel constraints to estimate gyro biases in real-time:

  • Multi-Level Constraints: The system uses a hierarchy of data rather than a single reference.
  • Pseudo-Observations: When minimal position change is detected over short periods, the filter introduces "pseudo-velocity" (zero) and "pseudo-position" (constant) updates.
  • Tightly-Coupled Sensor Fusion: To avoid mathematical singularities at 90-degree tilts (common in Euler-angle systems), the model uses raw accelerometer and magnetometer readings directly.
  • Quasi-Static Magnetic Field (QSMF) Updates: Indoor environments often contain magnetic interference. Instead of seeking an absolute magnetic heading, the algorithm identifies periods where the local magnetic field is stable (even if distorted). These stable windows are used to calibrate the vertical component of the gyro bias.
Magnetometer readings and calibrated references in outdoor vs. indoor environments.
  • Quasi-Static Attitude Updates (QSAU): The system detects brief moments of relative stability during a walk to further refine bias estimates.

Experimental Setup and Datasets

The algorithm was evaluated using real-world walking tests with Samsung Galaxy S3 and S4 smartphones. The researchers tested various motion modes and environments to ensure robustness:

  • Motion Modes: Handheld, phoning, dangling (in hand), pocket, belt, and backpack.
  • Environments: An outdoor sidewalk with minimal interference and an indoor university building chosen for its significant magnetic perturbations.
Sensor data across various motion modes including handheld, phoning, and dangling.

Key Results and Insights

The algorithm demonstrated impressive convergence and accuracy across all tested scenarios:

  • Bias Reduction: Initial gyro biases were reduced to under 0.1 deg/s outdoors and under 0.15 deg/s indoors.
  • Convergence Speed: Most sensor errors converged within the first 30 to 50 seconds of motion.
  • Robustness: Even in high-dynamic "dangling" and "pocket" modes, the system maintained high accuracy, with RMS errors around 0.17 deg/s.
  • Exploiting Natural Motion: A key insight is that natural human movement provides enough variety in orientation to make all gyro axes observable over time, eliminating the need for specialized rotation equipment.

Conclusion

This autonomous calibration method provides a promising path for improving consumer-grade navigation. By turning environmental challenges—like magnetic interference—into calibration strengths through the QSMF approach, the system remains reliable in complex indoor settings. Future work will focus on improving computational efficiency to preserve mobile battery life and integrating thermal models to handle temperature-induced sensor drift.

Summary updated

Abstract

This paper presents a real-time calibration method for gyro sensors in consumer portable devices. The calibration happens automatically without the need for external equipment or user intervention. Multilevel constraints, including the pseudoobservations, the accelerometer and magnetometer measurements, and the quasi-static attitude updates, are used to make the method reliable and accurate under natural user motions. Walking tests with the Samsung Galaxy S3 and S4 smartphones showed that the method provided promising calibration results even under challenging motion modes, such as dangling and pocket, and in challenging indoor environments with frequent magnetic interferences.
BibTeX
@article{li2015autonomous,
  author = {You Li and Jacques Georgy and Xiaoji Niu and Qingli Li and Naser El‐Sheimy},
  title = {{Autonomous Calibration of MEMS Gyros in Consumer Portable Devices}},
  journal = {IEEE Sensors Journal},
  year = {2015},
  doi = {10.1109/jsen.2015.2410756},
}
Figure