Model-Based Feature Extraction for Mid-Course Discrimination

A main objective of BMDO is to reduce the dependency of the optical mid-course discrimination algorithms on a priori information. Existing algorithms use average intensity, modulation, signature trends and frequency content as features, which are very sensitive to both geometry and training assumptions and impose large storage requirements on an operational system. Also features like average intensity may be easily masked using simple countermeasures. PRA is proposing to expand upon the physics model and estimation procedures developed in Phase I to extract, in real-time from optical signatures, dynamics-based features such as coning angle, angular momentum vector, spin rate and precession rate. These features reduce the dependency of discrimination on a priori information, make discrimination less susceptible to countermeasures and also simplify the training process. The estimator uses a physics model to iterate on a set of dynamics-based parameters until the sensor intensity measurements are best matched.