Optimal Feedback Design
Goal of project: Design and test various algorithms to create an optimal feedback controller.
On this page, a Simulink plant model will be provided, which algorithms will try to "control". The plant will occasionally be upgraded, so that we can test out more and more sophisticated algorithms. This page will also provide a cost function that algorithms will need to minimize, and a scoring rubric so that we can compare algorithms.
Your algorithm should look at the plant, do some sysID to figure out what the plant looks like, then self-design an optimal controller.
Plant Model
Round 1: 2 single pendula, one pole for cavity.
- Seismic noise (I'll provide a filter through which you can pass white noise, to create seismic noise).
- Simplant will have optics (cavity pole), mechanics (pendulums) and electronics noise.
- Local damping is static - algorithms should only try to control cavity length.
Upgrades to do in the future:
- Double, then triple or quad suspensions - Brett may have a model we can use.
- Other noise sources, ex. shot noise.
- Maybe add changing seismic noise, so that we can test optimal day/night transitions.
- Other degrees of freedom, especially potential low frequency cross-coupling (algorithm should figure out the cross-coupling).
- Useful at low-frequency, high-gain region, where transfer functions are hard to measure.
- Add ability to tweak damping control as well as length control.
Cost Function
Ideas, although should start simpler than these:
- Controller doesn't inject noise above thermal noise limit of IFO.
- Differential motion sensed by main laser is minimized.
Scoring
Elements of score:
- Convergence time
- Accuracy
- Minimization of cost function
Results of Algorithms
- When you make and test an algorithm, post your results here (on a subpage, then list it here)!