Description
This paper develops an optimal frequency-weighted strategy to design anytime controllers that can react to changing computational resources. The selection of the weighting function is driven by the expectation of the situations that would require anytime operation. For example, if the anytime operation is due to occasional and isolated missed deadlines, then the weighting on high frequencies should be larger than that for low frequencies. Low frequency components will have a smaller change over one sample time, so failing to update these components for one sample period will have less effect than with the high frequency components. Additional considerations explored in this paper are stability analyses, architectural issues, and transient management. Two examples are included that demonstrate the methodology.
