Allan variance analysis of the H764G stochastic sensor model and its application in land vehicle navigation

The Global Positioning System/inertial measurement unit GPS/IMU integration provides an efficient sensor configuration suitable for land and airborne vehicle navigation, or georegistration of imaging sensors used for mapping purposes. However, in GPS-challenged environments, where loss of lock due to interference, jamming, strong multipath or direct line-of-sight blockage, IMU becomes the stand alone navigation provider. In such situations, the performance degrades rather quickly (depending on the grade of the inertial sensors) due to the lack of external measurement update that supports the estimation of the IMU error states. Consequently, the strapdown inertial navigation algorithm uses the last estimated error states (at last GPS epoch) for the following navigation solution calculation. This may result in significant navigation errors, if the IMU error states were not sufficiently resolved before the GPS loss of lock, or their stochastic models are not appropriate; and the effect will be more pronounced for longer GPS gaps and lower-end IMUs. Therefore, assuring better estimation of error states is desired for the IMU-based navigation systems, which may be facilitated by better understanding of the sensor error stochastic properties. In this paper, the stochastic properties of the H764G sensors are derived using the Allan Variance method; the details of the analysis and the resulting model parameters are presented, and the navigation performance of a GPS/IMU land vehicle navigation system is evaluated using the "customized" set of parameters, the manufacturer's specifications and a heuristic method of EKF (Extended Kalman Filter) tuning. The results indicate that the rate white noise is the main error source of the Ring Laser Gyro (RLG), while the accelerometers also display rate random walk for the X axis, rate ramp for the Y axis and bias instability error for the Z axis. The 7-hour data set analyzed here was collected in static conditions with the sampling rate of 256Hz. The manufacturer's specifications are: (1) both gyro and accelerometer have bias and white noise error, and (2) all three axes of gyro and accelerometer, respectively, have the same error characteristics. To assess the impact of the Alan Variance-determined noise coefficients on the navigation results, two field tests performed at The Ohio State University (OSU) in 2008 and 2009 were used. Two dual-frequency GPS receivers (one kinematic and one static, serving as base receiver) and H764G INS collected data with the rate of 1Hz and 256Hz, respectively. The data were processed by the AIMS-PRO integrated software, developed at OSU; the primary modules of the AIMS-PRO EKF are (1) loose integration and (2) tight integration of GPS, IMU, pseudolite, as well as terrestrial and airborne laser scanning and image-based navigation data. Differential dual-frequency carrier phase GPS and loose coupling strategy were adopted in the test described in this paper. Three parameter sets, derived by the Allan Variance analysis, the manufacturer's specifications and heuristic method of EKF tuning, were used for comparison, including an independent experiment and statistical test. The statistical comparison of the results indicates that under short GPS outage, no significant difference between the three solutions exists. However, there is a major improvement in the solution generated with the set of parameters introduced by the heuristic method of EKF tuning. After introducing a 5-minute GPS gap the average position error decreased from 64 m to 33 m, as compared to the other two solutions.