Trigger memoization in self-triggered control
Indranil Saha, Rupak Majumdar · 2012
Self-triggered implementations of controllers have been proposed as an alternative to traditional time-triggered implementations. In a self-triggered implementation, the control task computes the actuator signal as well as a triggering time that specifies the next time instant at which the control task should be run. Self-triggered implementations have the potential to decrease communication costs and CPU requirements over time-triggered ones, e.g., by running the steady-state plant in open loop for long intervals if there is no disturbance. We show that commonly claimed gains for self-triggered implementations are too optimistic. The analysis of most self-triggering algorithms ignore the execution times for computing the trigger times. We show, using implementations of several self-triggering algorithms proposed in the literature on common embedded platforms, that the execution time to compute the trigger time can be non-negligible compared to the trigger times, and may even be higher than the trigger time itself, rendering a naive implementation infeasible.