40m QIL Cryo_Lab CTN SUS_Lab TCS_Lab OMC_Lab CRIME_Lab FEA ENG_Labs OptContFac Mariner WBEEShop
  SUS Lab eLog  Not logged in ELOG logo
Entry  Thu Feb 2 15:11:44 2012, Giordon Stark, DailyProgress, Coating Q, NEP of Experimental Set up [FIXED] NoiseMeasurements.pngExperimentalSetup.pngExperimentalSetup.pdf
    Reply  Thu Feb 2 18:59:20 2012, Zach, DailyProgress, Coating Q, front-end noise budget CQ_frontend_electronics_NB_2_2_2102.png
Message ID: 399     Entry time: Thu Feb 2 18:59:20 2012     In reply to: 398
Author: Zach 
Type: DailyProgress 
Category: Coating Q 
Subject: front-end noise budget 

I took the liberty of making my own plot. This was both in the interest of time and also to better illustrate how the plot should be made by example:



  • "NEP" is a relatively useless term. Manufacturers use it so they can report the smallest possible intensity noise level for a given electronic noise level. Also, it would be in W/rHz, not V/rHz.
    • It is useful to plot the intensity noise level for our particular setup (i.e., wavelength) alongside the electronic noise level in volts. This is best accomplished by putting another axis on the right side with the proper scale factor with respect to the other one. For this, you use the PD's transimpedance gain [V/A], its responsivity [A/W], and your brain [?].
    • The only tricky part, really, comes when you have to plot the manufacturer's quoted noise level. For this, you have to go from their NEP [W/rHz], to the corresponding electronic noise level [V/rHz], using the proper responsivity at the peak wavelength. In this case, the peak responsivity (@ lambda = 970 nm) is 0.65 A/W, while for us (@ lambda = 633 nm), it's ~0.41 A/W. A good check to make sure you've done all the calibrating correctly is to make sure that the manufacturer's spec that you've plotted---in W/rHz on the right side---is equal to their NEP number times the ratio of the two responsivities.
  • I'm not sure how you got things to end up out of whack in yours, but the calibration should be simple: take the raw voltage noise level we measured with the spectrum analyzer, then divide by our preamp gain of 500. This is the input-referred noise level, as we have discussed. That means that even though the input noise level of the spectrum analyzer is actually higher than the dark PD noise in an absolute sense, it actually looks much lower, since we've amplified our signal with the SR560.
  • The trace names should be clear and make sense:
    • PD1 dark: The electronic noise measured out of the first PD with no laser light on it
    • PD2 dark: The electronic noise measured out of the second PD with no laser light on it
    • PD1 - PD2, dark: The differential noise of the two PDs, with no laser light on them.
      • Notice how this is roughly sqrt(2) times the level of the previous two traces. This is what we expect from the incoherent sum of the two random signals.
    • Dark noise floor: The self-noise (shorted input) of the spectrum analyzer, while on the input range setting we used for the dark measurements.
      • We saved this as "PD1NFL", but since the input range never changed throughout the first 3 measurements, and the noise spectrum of the analyzer is stationary, this is a valid noise floor measurement for all three.
    • PD1 - PD2, balanced: The differential noise of the two PDs, with the laser on and the DC difference zeroed out (hence "balanced").
      • Notice how this measurement is significantly noisier than all the others. This means that, despite the large common-mode rejection we get of intensity noise from the laser (since the PDs are balanced), there is still a strong enough differential component that we see noise from the laser. This is a crucial thing to take home.
      • As of right now, this level (~25 nV/rHz --OR-- ~50 pW/rHz) sets the ultimate noise floor of the measurement in this configuration. Once we have figured out how to calculate an achievable signal size, we can calculate the sort of SNRs we can expect. This requires the following knowledge:
        • Using our ESD, to how large of a strain amplitude can we drive up the samples' modes?
        • How do we get from a strain amplitude to a rotation of polarization?
        • How do we get from a polarization rotation to a differential power signal in W at the PDs? (This last one is easy, and you should work it out ASAP)
    • Bright noise floor: Measured in the same way as the dark noise floor, but using the input range setting we used for the measurements with the laser on
    • Preamp noise (SR560): The well known---and visually re-measured---high-gain-setting input noise of the SR560 preamp (4 nV/rHz above ~10 Hz)
    • PD manufacturer spec (PDA36A): Calculated in the above-described way using the "NEP" figure. It seems strange that our actual measurement should be so far below this (~10x, not 100x), but it turns out that this is an over-estimated number by the manufacturer to save their a$$es. Depending on the quality of the 3rd-party components that go in, some units might be closer to this level than others. Considering the transimpedance gain of 1510 V/A and knowing that the opamp used is an AD829, the measured output noise value of ~7.5 nV/rHz is totally reasonable.

Also, some general plotting tips:

  • Use a linewidth of 2 for your plots. It gives them an air of confidence
  • The same applies for the fonts. The axis ticks, etc, should be ~12-14pt, the labels should be 14-16pt, and the title should be closer to 20pt. Otherwise, you just can't read anything. Legend font sizes depend on what you can fit in.
  • Use a grid so that you don't have to break out a ruler to interpolate
  • Set your axis limits so that you make good use of the plot space. All that whiteness is just a waste. Dynamic range rules.

Try to let this stuff soak in, as we'll probably have to make this whole measurement again. Next time, you will plot it!

Another thing: our whole noise budget will get a little more complicated once we add in the lockin, but this analysis of the front-end (the very place where physics meets measurement apparatus) is extremely important.


[Giordon, Zach]

 The last post we discussed the noise measurements with a noticeable peak around the 50kHz. This time, we've set the preamp gain to 500 and retook the measurements. The attached image shows the result of said measurements.

One thing we noticed is that the photodetector noise is roughly two orders of magnitude smaller than our best estimates for this. Zach thinks it's not real, I think it's pretty awesome. Here some explanations of the measurements.

  • A-B output means that the output comes from the preamplifier on an AC coupled setting amplified at 500. This either came with the laser on or off. We fixed the range here and then ran a noise floor measurement where we grounded the input.
  • PDA and PDB mean Photodetector A and Photodetector B as listed in the experimental set up drawing in an eLog post some days ago. PDA has a Noise Floor (NFLR) measurement like before as we noticed that the range of it was significantly different from the A-B output range [this is not true of PDB].
  • Theoretical PD and Theoretical Preamp are values highlighted from the rough noise calculations made about a week ago. (Same place as the inclusion of the experimental set up).

I've also noticed that the plot doesn't make complete sense to me. For example, the noise floor of PD1 seems to be higher than PD1 itself - but maybe Zach can re-explain to me the difference between fixed range and auto range here.

For lazy people, I've reattached the experimental set up and noise calculations.



ELOG V3.1.3-