Signal Processing for Micro Inertial Sensors

摘要

In the development of the guidance and control packages for unmanned vehicles, it is highly desirable to have inertial measurement sensors which are small, inexpensive, low power, reliable and accurate. New technological advances in the design and construction of micro inertial sensors, such as accelerometers and gyroscopes, have much promise in providing small, inexpensive, and low power devices; however, much improvement in the reliability and, especially, the accuracy of these micro devices is still necessary. Further major improvements in these two properties will probably not be accomplished in the near future, thus it will be necessary to use special signal processing methods to provide the accuracy. One way which has been proposed to improve the accuracy, and concurrently the reliability, of micro sensors is to use many, perhaps one hundred or more, micro sensors on a single chip (or a few chips) and using statistical methods to combine the individual outputs of these sensors to provide an accurate measurement. One method of performing such a combination is through an extended Kalman filter (EKF). A standard application of an EKF to an array of gyroscopes would involve at least six state equations per gyroscope and the number of covariance equations would be in the order of the square of the product of six times the number of gyroscopes. Obviously, the curse of dimensionality very quickly limits the number of sensors (gyroscopes) which can be used. Even if the EKF for each individual gyroscope is uncoupled from the rest, the number of covariance equations is of the order of the number of gyroscopes times six squared. This can still lead to a formidable computational burden. In this paper, a new technique of applying an EKF to this problem of combining many sensors is proposed. By using the common nominal model for each of the micro sensors and developing a single EKF, improved accuracy is achieved by a single EKF with the dimension of one sensor.