Improving the inertial navigation system (INS) error model for INS and INS/DGPS applications

In this thesis, different approaches are investigated for improving inertial error modeling to obtain better accuracy in SINS stand-alone and SINS/DGPS applications. The SINS error model contains deterministic as well as stochastic errors. Position, velocity and attitude errors are usually modeled as deterministic errors while the SINS sensor residual biases are often modeled as stochastic errors. The current SINS deterministic error model is obtained by linearizing the SINS mechanization equations and neglecting all second-order terms. The SINS stochastic biases are often represented by a first-order Gauss-Markov process. To improve SINS error models, both error types are handled in the thesis.; Different stochastic processes for modeling SINS sensor errors are discussed. The actual behavior of SINS sensor random errors is investigated by computing the autocorrelation sequence using long data records. Autoregressive (AR) processes are introduced as an alternative approach in modeling SINS sensor residual biases. Different methods for the optimal determination of the AR model parameters are studied. Compared to the other discussed random processes, results showed that the implementation of AR models improves the results by 40%--60% in SINS stand-alone positioning and by 15%--35% in SINS/DGPS applications during DGPS outages.; De-noising SINS sensor measurements using wavelet decomposition is presented as a method to cope with random noise. Wavelet de-noising is performed on static SINS data for an accurate estimation of the AR model parameters and for the determination of autocorrelation sequences. De-noising is applied on kinematic SINS data to reduce position errors. Testing results showed that the positioning performance using de-noised data improves by 55% in SINS stand-alone positioning and by 35% during DGPS outages in SINS/DGPS applications. In addition, a combination procedure using SINS data de-noising together with AR modeling of sensor errors is performed. This gives a further improvement of 10%--45%.; For the SINS deterministic errors, another error model is derived that considers all second-order terms. Errors computed by the linearized current SINS error model and the new derived second-order error model are compared using kinematic data. The results show that none of the second-order terms has a significant effect. To improve positions obtained during DGPS outages in SINS/DGPS applications, two different bridging methods are considered, backward smoothing and SINS parametric error modeling. In the thesis, the backward smoothing equations are modified while the SINS parametric error model is developed. When applying either one of the bridging approaches during DGPS outages, position errors are decreased by 85%--93%.