DIGITAL FILTERING USING SIMULATION MODELS

A VARIETY OF OBJECTS ARE TRACKED ON ARMY TEST RANGES, INCLUDING: SLOW-MOVING GROUND VEHICLES, HIGH PERFORMANCE AIRCRAFT, MISSILES, CANNON BULLETS, ORBITAL VEHICLES AND AIRBORNE DEBRIS. VIDEO AND RADAR SENSOR SYSTEMS SUPPORT TRACKING. ACCURATE, REAL TIME TRACKING IS CRITICAL IN SUPPORTING RANGE SAFETY CALCULATIONS AND PERFORMANCE EVALUATION. CURRENTLY OBJECTS ARE TRACKED USING POLYNOMIAL FILTERS (E.G. ALPHA-BETA FILTERS). POLYNOMIAL MODELS ARE USED BECAUSE THEY: ARE EASY TO CONSTRUCT, ARE COMPUTATIONALLY EFFICIENT, AND DO NOT REQUIRE DETAILED DESCRIPTIONS OF THE TARGET'S DYNAMICS. THE PRICE OF THIS SIMPLICITY IS THAT THEY RELY ON THE DATA TO CONTINUALLY CORRECT THE STATE ESTIMATE FOR UNMODELED EFFECTS. SINCE POLYNOMIAL FILTERS IGNORE SIGNIFICANT (AND KNOWN) DYNAMICS, THEY MUST BE MADE MORE RESPONSIVE TO CURRENT DATA (I.E. USE LARGER FILTER GAINS) THAN FILTERS BASED ON MORE ACCURATE MODELS. POLYNOMIAL FILTERS WILL BE CORRESPONDINGLY MORE RESPONSIVE TO NOISE IN THE DATA AS WELL. WE PROPOSE, IN PHASE I, TO IDENTIFY A LIMITED NUMBER OF GENERAL PURPOSE MOTION MODELS, BASED ON STOCHASTIC, DIFFERENTIAL EQUATIONS. WE WILL BUILD A SYSTEM FOR RAPIDLY SELECTING AND IMPLEMENTING FILTERS BASED ON THESE MODELS. WE WILL USE DATA COMPRESSION TECHNIQUES, AND THE MODIFIED EULER METHOD FOR SOLVING DIFFERENTIAL EQUATIONS NUMERICALLY, TO REDUCE THE COMPUTATIONAL LOAD.