摘自 Dr. Bernhard Hofmann-Wellenhof, Dr. Klaus Legat, Dr. Manfred Wieser (auth.)-Navigation_ Principles of Positioning and Guidance-Springer-Verlag Wien (2003)
Types of redundancy
An essential feature of sensor fusion is the presence of redundant information, i.e., more information than required to solve a defined task is availableabout a given process. After Beyer and Wigger (2001: Sect. 2.4.5), four types:
-
Parallel redundancy arises by using several identical sensors or devices.Voting systems directly compare the signals of the sensors to get aunique solution.
-
Complementary redundancy arises iftwo or more sensors with differentphysical operation principles and varying characteristics are used. Thesensors complement each other in the way that the advantage of the onecould be the disadvantage of the other and vice versa. As an example,the combination of inertial navigation and GNSS may be considered.
-
Dissimilar redundancy occurs in case of two or more nonidentical sensors which do not fully complement each other. A typical example isthe integration of GNSS and Loran-C: both systems provide positionfixes based on RF techniques but differ in terms of system architecture,signal structure, etc.
-
Analytical redundancy is based on a predefined knowledge of the system models. This knowledge may refer to kinematic modeling withrespect to the measurement environment, e.g., in case of a line-basedtrajectory on a given network; map aiding discussed in Sect. 13.4.5belongs to this category. As far as the dynamic model is concerned,preknowledge of velocity and acceleration limitations may be given.
Updating process
The multisensor technique requires appropriate methods of updating the navigation solution by redundant information. Several methods may solve this task:
• Signal blending (averaging) is usually applied in case of parallel redundancy. When using several sensors of different quality, weighted averaging is applied. Signal blending does not take into account a dynamic model.
• Filtering tries to achieve a more realistic processing of the signals by involving a dynamic model of the motion. In case of conventional filtering, stationary stochastic covariance models are used for the updating process.
• Optimal filtering employs time-variant stochastic covariance modelsand is achieved by Kalman filtering which is commonly applied for updating the state vector gained by multisensor navigation systems.