40m QIL Cryo_Lab CTN SUS_Lab TCS_Lab OMC_Lab CRIME_Lab FEA ENG_Labs OptContFac Mariner WBEEShop
  40m Log  Not logged in ELOG logo
Entry  Thu Nov 17 15:17:16 2016, gautam, Update, IMC, MCL Feedback MCLerror.pdfMCLtest.pngYarmCtrl.pdf
    Reply  Wed Nov 23 01:13:02 2016, gautam, Update, IMC, MCL Feedback MCL_plant.pdfOLG.pdfMC_armSpectra_X.pdfMC_armSpectra_Y.pdf
       Reply  Wed Nov 23 15:08:56 2016, rana, Update, IMC, MCL Feedback 
          Reply  Fri Feb 3 00:36:52 2017, gautam, Update, IMC, MCL Feedback - framing the problem 
             Reply  Fri Feb 3 11:40:09 2017, rana, Update, IMC, MCL Feedback - framing the problem 
                Reply  Mon Feb 6 17:03:41 2017, gautam, Update, IMC, MCL Feedback - simulink model updated mc40_v1.pdf
                   Reply  Mon Feb 6 18:20:08 2017, Koji, Update, IMC, MCL Feedback - simulink model updated 
                      Reply  Wed Feb 8 19:13:02 2017, gautam, Update, IMC, MCL Feedback - TF measurements 
                         Reply  Thu Feb 9 23:35:34 2017, gautam, Update, IMC, MCL Feedback - TF measurements mc40_v1.pdfCMboard_OLTF_fit.pdfFSSFast_OLTF_fit.pdfPCdrive_OLTF_measured.pdfdata.zip
Message ID: 12793     Entry time: Fri Feb 3 00:36:52 2017     In reply to: 12637     Reply to this: 12795
Author: gautam 
Type: Update 
Category: IMC 
Subject: MCL Feedback - framing the problem 

Rana motivated me to take a step back and reframe the objectives and approach for this project, so I am collecting some thoughts here on my understanding of it. As I write this, some things still remain unclear to me, so I am leaving these as questions here for me to think about...


  1. The PSL is locked to the IMC cavity - but at frequencies near 1 Hz, the laser frequency is forced to follow the IMC cavity length fluctuations, even though the free-running PSL frequency noise at those frequencies is lower. This excess is also imprinted on the arms when locked to the IR. We would like to improve the situation by feeding back a portion of the MC PDH error signal to the cavity length actuator to stabilize the MC cavity length at low frequencies. Moreover, we would like this loop to not imprint additional control noise in the arm control signals, which is a problem we have observed with the existing MCL loop. 
  2. The borader goal here is to use this project as a case study for designing the optimal loop and adaptive feedback. Can we come up with an algorithm, which takes
    • A model of our system (made with measured data where possible)
    • A list of our requirements (e.g. in this case, frequency noise requirements in various frequency bands, smooth crossovers between the various loops that enable locking the PSL to the IMC cavity and avoid injecting excess control noise into the plant)

and come up with the best loop that meets all our rquirements? What constitutes the "best" loop? How do we weight the relative importance of our various requirements? 

Proposed approach:

For the specific problem of making the MCL feedback loop better, the approach I have in mind right now is the following:

  1. Build a model of the 40m IMC loop. Ultimately the performance of the loop we implement will depend on the transfer function from various additive noise sources and disturbances in the feedback loop (e.g. electronics noise) to the output (i.e. laser frequency). Building an accurate model will allow us to quantify the performance of the proposed control loop, and hence, optimize it with some algorithm. I did some work on a simplistic, purely analytical model of the two MC loops (MCF and MCL), but Rana pointed out that it is better to have something more realistic for this purpose. I have inherited his Simulink models, which I will now adapt to reflect the 40m topology. 
  2. Come up with a list of requirements for the MCL controller. Some things that come to mind:
    • Reduce the arm control signal spectral amplitude below 20 Hz
    • Not increase the arm control signal spectral amplitude above 20 Hz
    • Crossover smoothly with the FSS slow temperature control loop and the MCF loop. 
    • What factor of suppression are we looking for? What is achievable? Once I build the model, it should shed some light on these..
    • Is the PMC a more stable frequency reference than the NPRO crystal at low frequencies? This measurement by Koji seems to suggest that it isn't (assuming the 1e4 product for the NPRO free-running frequency noise)..
  3. Once we have a model and a satisfactory list of requirements, design a control loop that meets these using traditional techniques, i.e. desired tracking error in the control band of 0.1-20 Hz (is this possible? The model will tell us...), gain and phase margin requirements etc. But this need not necessarily be the optimal controller that meets all of our requirements
  4. Optimize the controller - how? Can we define an objective function that, for example, rewards arm control signal suppression and penalizes injection of control noise, and just fminsearch in the [z,p,k] parameter space of the controller? Is there a smarter way to do this?
  5. Can this algorithm be adaptive, and optimize the controller to adapt to prevailing seismic conditions for example? Is this the same as saying we have a model that is accurate enough for us to predict the response of the plant to environmental disturbances? 

My immediate goal is to have the Simulink model updated.

Thoughts/comments on the above will be appreciated...

ELOG V3.1.3-