Improved methods for navigation data fusion
Traditionally, data fusion in modern onboard navigation systems is performed using stochastic filtering algorithms. To design such algorithms, signals and errors of navigation sensors should be described as random sequences or processes; this, in turn, generates a need for identification of their models. The latter are often obtained by the methods based on the calculation of power spectral densities, correlation functions, and Allan variances. However, all these methods involve long measurements, empirical approach to the identification of the model structure and complicated accuracy estimation, this being their disadvantages.
In this regard, the activities of the International Integrated Navigation and Attitude Reference Systems Scientific Laboratory include:
– development of optimal identification algorithms based on nonlinear filtering approach using the Kalman filter bank (Figure 1). Such algorithms allow identifying the model structure corresponding to the maximum likelihood and obtaining optimal Bayesian estimates of its parameters for any time interval;
– development of computationally effective algorithms for estimation of polyharmonic signals with unknown parameters. These suboptimal algorithms are based on spline interpolation and the method of artificial measurement. As a result, the estimates are obtained by simplified transformations instead of using filter bank, which significantly reduces the amount of computations;
– development of federated filtering methods, which allow the solution of navigation problem to be distributed between several data processors. It enables a more flexible approach to design of navigation systems with modular construction.
Solution of these problems provides efficient algorithms for data fusion in onboard navigation systems for vehicles of different classes.
Figure 1. Algorithm of identification flowchart.