A Study Of Algorithms And Fault Diagnosis For Precise Point Positioning

Precise point positioning (PPP) is an important subject in GPS. The paper focuses on the theories and algorithms of PPP. The main works are shown as follows: 1. Kalman filtering is used to process the static data and the dynamic data. To the different state, the different dynamic models are given in Kalman filtering, and examined by several data in the different state. The reliability and accuracy of dynamic PPP will be improved if the immobile distances of GPS receivers in the aircraft as contraints are considered. 2. The variance component estimation is used to adjust the proportion of weights between phase observations and pseudorange observations. In PPP, the proportion of weights for two kinds of observations is a constant during the computation, so it is not consistent with the fact, it will slow down the convergence speed of ambiguities. Simplified Helmert variance component estimation is employed to adjust efficiently the proportion in the real time in order to improve the accuracy of PPP and the convergence speed of ambiguities. 3. The information of troposphere delay in CORS is used in PPP. When it rains or snows, the common model of troposphere will get the higher delay. The sequence of troposphere delay in the position of the receiver will be interpolated by the information of troposphere delay of CORS, and is used to replace the adjustment estimated by troposphere model. So the accuracy of positioning will be improved. 4. The algorithms to fix the ambiguities are applied to dynamic PPP. When ambiguities are estimated as the state parameters, it obviously enlarges the dimensions of the matrix, and is disadvantageous to the convergence of ambiguities. Firstly fixing single difference ionosphere ambiguities can not only reduce the dimensions of the parameters, but also improve the convergence speed and the accuracy of position in the initial time. For noises of GPS data always are not Gaussian noises, the least square is used to weaken the influence of colored noises on the base of ambiguities fixed. Because the covariance matrix of the state noise is given by experience which always has bias, it can affect the contribution to the positioning result. The adaptive factor is used to modulate the proportion between the covariance matrix of predicted state vector and the covariance matrix of observational noise in order to make full use of dynamic model.5. To the linearization of nonlinear functions and non-Guasssian noise of GPS data, particle filter is employed to reduce the loss of accuracy. The number of particles in particle filter relates with dimensions of the state vector. If dimensions of the state vector are reduced, the number of particle can decrease. Particle filter based on fixing single difference ambiguities can not only reduce the number of particles, but also improve the positioning accuracy in the paper. Important sampling with Kalman filtering can optimize particles for degeneracy of the particle occurs in particle filter. To get higher accuracy of the particle, particle swarm optimization on the base of important sampling with Kalman filtering can heighten accuracy of particles. Mean shift algorithm is used to improve the efficiency of particle filter which is a shortcoming for particle filter.6. The prior information of ionosphere is considered in Kalman filtering for single frequency PPP. The residual of ionosphere delay has strong influence on the result of positioning as grid ionosphere model only can eliminate 60% delay. Ionosphere delay is estimated again as parameter to further improve accuracy of ionosphere delay on the base of ionosphere model predicting delay with observations.7. Detection and diagnosis of failures in PPP is the important segment. It is difficult that robust estimation can come true for there are redundant observations in PPP. In the paper, interacting multiple models based on the ratio of the probability of failure models is proposed to improve efficiency of diagnosis. And particle filter is used deal with observations with outliers. Two neural networks are designed to diagnose failure to get higher efficiency of correction as neural network has strong approach ability and pattern recognition ability.