Michael Lemmon Dept. of Electrical Engineering University of Notre Dame

1
Breaking the Paradigm of Periodic Hard Real-time
Feedback Control

• Periodic hard real-time guarantees are difficult
  to enforce in mesh wireless communication
  networks.
• Is it possible to effectively control systems
  without periodically triggered tasks? YES!
• Event-triggered Schemes have been developed that
  ensure asymptotic stability and L2-stability.

2
Event-triggered Feedback
Introduce switching rule to ensures the
dissipative inequality is always satisfied
for t?rk , fk1 )

• Full state feedback controller
• L2 Storage Function
• Dissipative Inequalityensures specified level
  of disturbance rejection

triggering thresholds
error er(t) x(t)-xr
fk1
rk
fk
rk1
rk2
fk2
JOB2
JOB1
3
Event Triggering in Networked Control Systems

• Event-triggered Feedback is not practical for
  distributed control of networked systems.
• This suggests we use a self-triggered approach
  to event detection

Event Detection requires a function of
neighboring states exceed a given threshold.
Self-triggered requires central agent to
broadcast the next sampling instant Neighbors
then schedule their replies
4
Self-Triggering for H2 Performance
Self-triggering requires task/agent to predict
its next release time. For LTI systems it is
possible to obtain practical bounds on task
period/jitter that assure L2-stability This
provides a way of adapting real-time resources
to what is actually happening in the application.

5
Real-time Challenges for Self-triggered Feedback
over Networked Control Systems

• Design of Distributed Controllers under
  Self-triggered Feedback
• robustness to missed and delayed feedback
• Asynchronous feedback and global clock
  synchronization
• Self-triggering uses bounds acceptable levels of
  task jitter and task periods. How can we use
  these bounds for message scheduling?
• generalizations of elastic task model paradigm
• How do we schedule competing requests?How does a
  node decide which request to satisfy first?
• How do we uniquely characterize priority in a
  distributed system?
• Adaptive reallocation of real-time resources and
  control effort in response to changes in
  user/run-time/control environment.
• Moving away from periodic hard-real time feedback
  control
• Scheduling constraints that are application-aware
  (control).