Self-adaptive Filters for the Integration of Navigation Data

Abstract

The traditional non-adaptive Kalman filter includes models of the error characteristics of the navigation aids in use and such filters are very successful, so long as their model assumptions approximate to the true error characteristics sufficiently closely. However, for any filter there will be times when the environment changes and one or several aids will have errors which are not consistent with the assumed error models. It is necessary to consider carefully the sensitivity of the filter to such changes and, where a significant reduction in performance ensues, modifications to the filter are necessary.This paper introduces a Kalman filter which monitors the behaviour of internal variables to detect and characterize any model imperfections. The filter will then adapt its internal model of the environment accordingly. The discussion is restricted to the development of a navigation filter for integrating dead reckoning (EM log and gyrocompass) and Omega data. The principles are the same for any filter and details regarding similar analysis involving the use of other aids, for example Satnav and Decca, have been developed in a similar way.Before implementing any filter it is necessary to understand the behaviour of the measurement errors. For the dead reckoning and Omega aids this behaviour is described in section 2, while section 3 outlines a filter for integrating these aids and introduces the problems of model imperfections.