Positioning with Inertial Sensors and Aid Sensors when loosing GPS Fix

Abstract

This thesis designs a multipurpose navigation unit. The unit is equipped with a GPS which is the primary navigation sensor, when the GPS satellites are available. If the GPS satellites becomes unavailable, a inertial system with aid sensors is used for navigation. A unit that collects data from a GPS, gyroscopes, accelerometers, magnetometers and a barometer has been developed in a project leading to this thesis. The software development was not completed. The software responsible for communicating between the ADIS16405 sensor, which contains three gyroscopes, three accelerometers and three magnetometers, and the microcontroller was completed in this thesis. The software responsible for communicating between the barometer and the microcontoller was optimized in order to prevent timing errors. The menu system has been extended and sensor data is now saved in different files. Navigation equations has been found and is augmented with noise states. The augmented equation is used as the process equation in the Unscented Kalman filter. The augmented noise states is an approximation of flicker noise from the gyros. It was found that a first order Gauss-Markov process was the noise process which gave a change of covariance over time that was most similar the flicker noise of the tested noise processes. A filter for the gyroscope output is designed. This filter is intended to remove flicker noise from the gyroscope measurements. Measurements equations for the magnetometer and barometer is also found. All sensors have been calibrated and the error model coefficients has been estimated. The gyroscopes and accelerometers have less noise than specified in their datasheet. Because the lack of a calibration equipment the magnetometer has not been calibrated accurately. The inertial system has been simulated by Monte Carlo simulations to quantify the error sources. The gyroscope noise is the main error source, and noise from the sensors that provide data to the initial attitude calculation is the second largest error source. The standard deviation of the position error is approximately 65 m after 60 seconds when the system is stationary. This is found by using Monte Carlo simulations. It would be interesting to find how much this error could be reduced by using a better gyroscope. We assumed that the better gyroscope had on tenth of the noise compared to the estimated noise values from the real gyroscope. This reduced the standard deviation of the error to approximately 13 m after 60 seconds. Simulations show that the standard deviation of the position error could be reduced from 65 m after 60 seconds to approximately 20 m after 60 seconds by using magnetometers and barometer in addition to the inertial sensors. The real world tests have a very large position deviation. It is likely that this is caused by unmodeled effects, like magnetic disturbance. The Unscented Kalman filter also contributes to the error by not propagating the mean of the process equation correctly.