Error Compensation And Extension Ofadaptive Filtering Theory In GNSS/INS Integrated Navigation

This dissertation mainly focuses on the theories and algorithms of positioning and attitude determination in GNSS/INS integrated navigation which includes the principles of GNSS/INS integrated navigation, the random errors analysis and modeling of INS, the initial alignment and kinematic attitude alignment, error compensation of function model and random model, the adaptive filtering algorithms in GNSS/INS integrated navigation and INS navigation algorithms during GNSS outages. The main works and contributions are summarized as follows:1. The principles of GNSS/INS integrated navigation are summarized including the principles of INS navigation and GNSS navigation, the filtering models of GNSS/INS integrated navigation and the differences and advantages of open loop and close loop for error modification.2. The INS random errors are analyzed and processed systematically. Several kinds of random errors are analyzed and compared by power spectral density and Allan variance. And the random errors of three kinds of gyros are analyzed and the main characteristics and the corresponding coefficients are achieved. A method of determining the scale of wavelet multiresolution analysis automatically is presented and used to de-noise the high frequency noise. And ARMA models are used to fit the low frequency noise and the order and parameters are determined for the three kinds of gyros.3. In the refined initial alignment, a new method of SINS initial alignment based on the minimum of cumulated error is presented for the situation of short time. Real experiments are carried out to show that the new method can get good results when the initial static time is very short.4. A new algorithm considering attitude update in GNSS/INS integrated navigation for land vehicle is presented. Then the principle of yaw angle determined by GNSS velocity is introduced and the pitch and roll angles of low-cost INS in land vehicle are analyzed. By the actual data calculation, the precision of yaw angle determined by GNSS velocity is comfirmed and it is shown that the new algorithm can improve the navigation accuracy greatly compared with the Kalman filtering based on position and velocity.5. Three methods of direct estimation, estimation based on wavelet transform and Allan variance estimation are given to determine the power spectral density of white noise in INS random errors. By the calculations, it is shown that direct estimation and estimation based on wavelet transform can not recognize the white noise and other noises while Allan variance estimation can do. However, Allan variance estimation can not estimate the power spectral density when white noise buries in other noises. GNSS/INS integrated navigation results also show Allan variance can reflect the charactoristic of the white noise in INS random errors.6. Autocorrelation and parcorrelation parameters of INS errors show that AR model espectially ARMA model can reflect the low frequency errors better than GM model. Based on AR(3) model and ARMA(3,3) model, the dynamic model function and process noise random model are deduced. Real experiment results show compared with GM model, AR model or ARMA model can improve navigation precision, espectially ARMA model.7. In GNSS/INS real-time tight integration, the phase smoothing code method is introduced into the GNSS/INS tight integration. By the calculation, it is shown that when cycle slip occurs few times during a period, the method of phase smoothing code can improve the precision of GPS/INS integration greatly. But if cycle slip occurs many times during a period, the precision of GPS/INS integration is improved slightly.8. A new method of velocity determination by derived pseudoranges is presented. The formulas of derived pseudoranges are given and the precision of the velocity is analyzed. By two actual experiments, the accuracy of velocity determination by derived pseudoranges is compared and analyzed. It is also shown that in urban areas velocity determination by derived pseudoranges can ensure the reliability.9. In loose integration and tight integration, in order to avoid degrading the accuracy of reliable parameters by single adaptive factor, classified adaptive factors based on predicted residuals and selecting the parameter weights filtering. The new adaptive factors give position, velocity, attitude, gyro and accelerators errors parameters different weights for their charactorics. Compared with standard Kalman filtering and the adaptive filtering with single factor, the new algorithm can improve the navigation precision.10. A new factor based on partial state discrepancy is developed. The formula of the new adaptive factor is deduced. And the differences of the three factors including single adaptive factor and classified adaptive factors based on predicted residuals are compared and analyzed. By the calculation, it is shown that all adaptive factors constructed by the predicted residuals and partial state discrepancy can resist the influence of the dynamic model errors when no outliers exist in measurements, and the precision of their navigation are almost equal. But if outliers exist in measurements, the adaptive factor based on the predicted residuals can not identify the state model errors and the observation errors while the adaptive factor based on partial state discrepancy can resist the influence of the outliers, so the latter navigation precision is prior to the former navigation precision.11. A two-step adaptive robust Kalman filtering based on the observability of the parameters is presented. First the process of the tight integration is given and then the formulas and the approches of the new method are deduced and analyzed. Finally an actual calculation is given. It is shown that campared with tight integration, the two-step adaptive robust Kalman filtering can control the disturbances of the state and the outliers of the observation. And the navigation precision does not decrease while the integration period becomes longer and the INS errors become bigger. The INS errors can be rightly estimated and the precision of attitude angles is improved.12. In tight coupled GNSS/INS integration by using Kalman filtering, the main error includes not only the dynamic model error (INS error), but also the errors caused by the poor geometry of GNSS satellites or short of GNSS satellites. An extended adaptive Kalman filtering algorithm is presented based on the adaptive filter. The new algorithm can not only resist the influence of the dynamic model errors but also control the influence of the errors caused by the poor geometry of GNSS satellites by adjusting the coefficient matrix of the predicted states. An actual computation example shows that the new algorithm can degrade the influence of the two kinds of errors and improve the precision of navigation.13. When GNSS occurs outages in land vehicle navigation, the Odometer observations are added to the measurement equation to improve the system observability and navigation accuracy based on velocity constrains. And a new method is implemented on the position modification. By calculation, it is shown that compared with INS navigation, the precision of position and velocity are improved further.14. When GNSS outages occur, INS errors will accumulate quickly. Then error model modification is given. By the actual calculation, error model modification can compensate INS errors more effectively and improve the precision of INS navigation alone obviously.