Temperature-Bias Compensation of Low-Cost Inertial Sensors – Possible or Pipe Dream?

Abstract

Navigation using low-cost inertial sensors costing less than £1 each is generally considered impossible. With various measurement error contributions, the velocity and position estimates from these sensors drift exponentially with time. By simulating the sensor, we show how the zero bias error is the most serious contributor. The zero bias is known to change with temperature due to dissimilar thermomechanical characteristics of materials in the sensor’s construction and others have shown this trend to be nonlinear, exhibit hysteresis and unique to each sensor. This is a problem because it suggests error compensation by modelling (software level), or sensor redundancy (hardware level) will be ineffective. From temperature experiments on three of the same low-cost sensors, we show that temperature-bias responses are indeed unique and nonlinear but may be opposing between sensors. Furthermore, we show that one can get lucky and obtain a sensor with an axis that is relatively insensitive to temperature. This is encouraging because it supports the idea that an inertial measurement unit comprised of an array of inertial sensors can be fused to provide higher accuracy measurements than a single sensor operating alone. Lastly, we identify a threat to this idea we call temperature shock and suggest how it can be avoided. While the contributions of this work are intended to improve the accuracy of human position tracking, their impact extends to any field where lengthy periods of position tracking under Global Positioning System (GPS) denial is required.