'Stealth' filtering with reduced order observations

Communications resources in modern automated systems have to be shared between different users and purposes. A process filtering task performed by a dedicated processor and integrating measurements delivered by a remote information source may in particular be subject to data flow limitations. The authors examine some problems related to these measurement data reductions in a linear filtering algorithm, focusing on linear observation compression schemes. A global approach gives a sufficient algebraic condition to admit reduced-order measurement vectors. Compression policies, dynamically optimized under specific criteria, are proposed. Remaining generic problems and possible extensions of such an approach are also discussed.