THE PROBLEM OF 1000 SUPERCOMPUTERS
Towards Truly Reliable Automation
Modern automation needs computation of reliable mean values from measurements that are in real-time coming into a computerized system from many different data sources. All such data are influenced by both the random and systematic errors of various signals from radio-beacons, satellites, radars, lidars, sodars, environmental in-situ measurements etc. Thus, the most reliable mean values can only be computed by weighting all the measurements by correct weights describing accuracy and true representation.
What Problem?
The famous theory of Kalman Filtering (KF) was developed in 1960 for obtaining the best possible (i.e. optimal, correct) weights. Thus, every modern navigation receiver has built-in KF processing inside. It has also turned out that these optimal weights are necessary to be used by every truly reliable Guidance, Navigation or Control (GN&C) system. However, these optimal weights are the more tedious to compute the more measurements have to be made of use by the KF processor of an automated system. In fact, all large sophisticated systems like those used for numerical weather prediction are already exploiting so much observational data that their KF computations are facing the problem of "1000 desktop CRAYs working in tandem" (Gal-Chen et al., 1988: Bull. Am. Met. Soc., Vol. 71, No. 5, May 1990, see page 684).
What Solution?
An extremely fast way of computing these correct weights was discovered at the Finnish Meteorological Institute in Helsinki in 1989. The solution is known worldwide as the Fast Kalman Filtering (FKF) method. The FKF method also provides objective accuracy estimation based on observed internal consistency of all available measurements. FKF has already been patented in 43 countries through the international patent applications of PCT/FI90/00122 and PCT/FI93/00192. In addition, some 50 countries are being included through PCT/FI96/00621. The FKF solution is semi-analytical. Thus, it is highly improbable that any simpler and faster alternative can exist.
What Applications?
FKF represents the fastest truly reliable computing method for sophisticated GN&C systems as it allows the operational accuracy to be computed in real-time by R. C. Rao's method of MInimum-Norm-Quadratic-Unbiased-Estimation (MINQUE) based on observed overdetermination of signals and sensors. Growing demands for the best possible safety of automation will be satisfied only by using FKF as other precise computing methods are leading to uses of supercomputers that may even have to work in tandem. Automation will be increasingly based on FKF processing in applications like: navigation receivers of next-generation mobile phones, electronic stabilization of cameras, process control, autonomous vehicles, environmental and weather forecasting systems, Asynchronous Transmission Mode (ATM) data-communications, various remote-sensing or tomographic measurement systems, statistical calibration of various observing systems, etc.
New Opportunities!
One may only guess the full importance of the FKF patents that are now offering new opportunities to small/great countries/industries for manufacturing high-technology products under worldwide market protection. Thus, the products can be safely developed without the risk that all investments are torn into pieces by market forces from savage price competition.