Study On Algorithms Extension In GPS/INS Integrated Navigation

This dissertation mainly focuses on algorithms of data processing in GPS/INS integratednavigation which includes the principles of GPS/INS integrated navigation. After analyzing ofthe influence of the prior covariance errors in standard Kalman filter, the influence ofobservation outliers and the dynamic model errors in GPS/INS navigation is discussed. Theprecise point positioning (PPP)/INS integration and use of time differencing carrier phase(TDCP) in improving the velocity estimation are also studied. The main works andcontributions are summarized as follows:1. The principles of GPS/INS integrated navigation are summarized including the principlesof INS navigation and GPS navigation. An overview on the common reference frames andtransformations in inertial navigation is provided. The data processing of INS in EarthCentered Earth Fixed (ECEF) frame and the filter models of GPS/INS integration are alsoprovided.2. The principles of Kalman filter and adaptively robust filter are summarized. The influenceof the prior covariance errors to the standard dynamic Kalman filtering is discussed. Theinfluence expressions of the prior covariance matrix errors including the covariance matrix ofstate parameters, dynamical model errors and measurement noises are deduced. The resultshows that there is a balance between the prior covariance matrixes, which guarantee theKalman filter to achieve an optimal result. The more precise of the measurement, the smallerof the covariance matrix of measurement noise is. If the dynamical model is low precise, thefilter result can be improved from augmenting the covariance matrix of the predicted states.3. Generally there are not enough redundant observations to estimate all the state parametersin GPS/INS navigation, thus if the measurements exist errors, it’s not proper to control theirinfluences by using robust filter. In order to eliminate or depress the influences of themeasurement errors, an interacting multiple model algorithm is proposed. A model set,including both normal and exceptional observational models, is established. The Markovtransition probabilities are used to switch one model to another. When observations arenormal, the normal observation model is used chiefly, otherwise the exceptional observationmodel works. The navigation result is weighted fusion of all the outputs in the model set. The Markov transition probabilities take important part in the algorithm for model choosing. It isdifficult to choose which model works if the probabilities are too small, which are calculatedby the predicted residuals. So a simple formulation is given for probability calculation, whichcan avoid the confusion about the model choosing.4. Kalman filter is widely used in the area of kinematic positioning and navigation. However,it does not have the ability to resist the influence of measurement outliers. Hence itsperformance is easy impacted by the observation outliers or kinematic state disturbing. Inorder to guarantee the reliability of the navigation with precise dynamic model, a model set,which contains many different observation models, is established. An improved Kalmanfiltering, in which the design matrix of the observational model is substituted by itsexpectation is proposed to control the influences of the measurement outliers. An integratedGPS/INS navigation example is given to show that the modified Kalman filtering algorithmworks well.5. In GPS/INS integrated navigation, the IMM algorithm and Kalman filtering with randomparameter matrices are proposed to depress the influence of measurement outliers. Similarly,the both algorithms also can be used to get rid of the dynamical model errors if thecorresponding model set is established properly. However, the state prediction is mainly relyon the mechanization equations of initial navigation system, and it is not easy to establish areliable dynamical model set. So an adaptively Kalman filtering with random parametermatrices is proposed, in which the adaptive factor and random parameter matrices are used tocontrol the influence of the dynamical model errors and measurement outliers respectively.6. In the tightly coupled GPS/INS integrated navigation, the information of the GPSmeasurements are not using sufficiently. Actually the carrier’s dynamical information can beextraction from GPS observations. A novel tightly coupled GPS/INS integration consideringthe GPS dynamical information is proposed. This algorithm can increase the redundant in thepredicted states of position and velocity from GPS CV model, and also can enhance thereliability of the prior predicted states.7. An improved time differencing carrier phase (TDCP) velocity estimation algorithm in thetightly coupled GPS/INS integration has been proposed. Velocity is estimated using twodifferent methods. One employs the Doppler measurements and the derived Doppler measurements from TDCP to improve the velocity estimation; and the other one takes use ofthe derived Doppler measurements as virtual measurements in observation model. Then thepartial state discrepancy is used to estimate the velocity by using robust estimation.Experimental results shown that the precision of the velocity from both methods can beimproved effectively.