Group notes after MJI accidentally came across:
http://www.oceancontrols.com.au/SMC-005 (Installed and working on cable car)
http://www.oceancontrols.com.au/SMC-011 (Purchased; may be used on optics puller)

How Many Microsteps are Needed?


At a maximum required cart rate of 4mm/s of OPD, this corresponds to 1mm/s cart velocity. According to North (2007, PhD), the wheel diameter is 20mm, meaning a 62.8mm circumference, and a frequency of 0.016 revs/s. To ensure that microstepping noise is above 30 Hz (twice the cart resonance frequency, where mechanical vibrations do not transmit well), there has to be at least 1884 micro-steps per revolution. The ocean controls 005 drive has up to 10,000 steps per revolution, so exceeds this requirement by a factor of more than 5. Each step is a very large 25 microns of OPD, but this is a small fraction of the piezo range, and a step of 2.1 microns driving a 15 Hz resonance is only a small fringe velocity for the piezos to control. But clearly, in this case, half the maximum speed would then drive a micro-stepping resonance. Lets actually write the math down. The cart velocity at which the 15Hz resonance is driven is:



The semi-amplitude of a single step in microns is 2 times the wheel circumference divided by the number of steps:



At 102,400 steps per rev (the maximum for SMC-011), this is 1.2 microns, and the cart velocity at which this resonance is driven is 0.037 mm/s. Our requirement really should be less than a radian of phase of fringe motion in 10 milli-sec, which means a semi-amplitude of 0.1 microns. Making an educated guess that the servo decreases the amplitude by a factor of 5 but the mechanical resonance increases it by a factor of 5 (conservative), we'd still have 0.1 microns after the servo locked and we're out by a factor of 12 from this requirement. Something seems wrong with this... i.e. how could such a system work at < 20 or > 50 microns/s, but not in-between.

What about slew speed? We really want to go 10m of OPD per minute, i.e. 0.66 revs/s or 33 Hz going to the stepper. So filtering out 15Hz wouldn't work unless the filter could be turned off during slew... something very awkward/likely impossible anyway because we're talking about filtering out the high-current signals to the stepper.

What about fancy features of the SMC-011? These include a "pulse smoother" and ways to tune an internal current servo loop. MJI is (obviously) very tempted to just try and see what happens...

According to SMO's thesis, the existing system uses 8 bits for the cable car and 16 bits for the optics car translating into 256 micro steps per step for the cable car and 65536 for the optics. This means 51200 micro steps per revolution for the cable car and 13,107,200 for the optics car (the design goal having been ~1/4 of the metrology resolution). The Ocean controller has 102,400 steps/rev. The optics car is currently running at a much smaller micro-step size. However, it is MJI's understanding that the DAC doesn't have a bit depth to match the number of micro-steps (8 or 12 or 14 but not 16 bits), and when WJT and MJI discussed this in detail one SUSI trip, they concluded that removing 4 bits (dividing the number of microsteps by 16) would have no affect of the performance. Of course, this is still ~1,000,000 steps, and a further factor of 10 from the current optics cart step size. [MJI is wrong. The DAC has 16 bits. However, whether such fine step resolution is really needed is questionable.]

Coming from the other direction, with 102,400 steps/rev, the step size is ~614nm or 34 times the metrology resolution. While this is easily within the piezo range, BAW is worried that the movement will not be smooth enough resulting in a jump of more than half a wavelength every time it steps. It is possible that it will be mechanically damped enough to not be a problem, but I just don't know. One way to increase the number of steps per revolution but to still take advantage of a commercial motor controller is to put in a gear box. The downside of this is that it starts to limit the top slewing speed as the maximum rate for the Ocean controller is 200kHz. It would be ideal if we could change the step resolution on the fly (requires RS-232 communications). Also, the output shaft size and position on the gearbox is not correct for the wheel requiring more mechanical work.

Digital or Analog?


A commercial microstepping controller with much bigger microsteps solves many of our new controller issues, and also replaces the stepper motor amplifier circuits. Digital signals are probably superior for the motor control as the step size is very small meaning very small changes in analog signal levels (microvolts) leading to the conclusion that it would be very susceptible to electrical noise if we were to try to send analog signals through the cable. All that is needed is a low-noise PCI to piezo amplifier signal transmission, or possibly a DAC built in to the new PCI card as BW is investigating. Analog signals might make more sense for the piezos as the minimum step size as it is currently set-up is 2 mV. This is still small, but I think manageable with differential signals.

Attempt to summarize cart testing


Motivation: fringe visibility numbers should go here.

Air conditioning


The A/C affects the 50Hz vibration in the cart. With the cart tracking an astromod with approximately 0 speed, there is a clear difference between A/C on and A/C off:
Tracking ~0m/s
A/C
RMS
50Hz peak
100Hz peak
on
3.18
3.05
0.31
off
1.84
0.36
0.10
This is consistent with the accelerometer measurements. However, the 50Hz spike does not disappear completely until the power to the cart is switched off. Finally, there is no 50Hz signal when the cart is powered off even with the A/C on. It seems this signal is being contributed as line noise on the mains power (possibly through the optics car pusher motor). The A/C does introduce lower frequency vibrations either to the cart or the metrology lasers with peaks at 8.9Hz, 14.3Hz, 15.4Hz, 24.4Hz, 32.7Hz, 48.5Hz:
PLC power off
A/C
RMS
50Hz peak
100Hz peak
on
1.15
0.00
0.00
off
0.69
0.00
0.00
And for completeness, the PLC powered, but not tracking:
Not tracking
A/C
RMS
50Hz peak
100Hz peak
on
1.47
0.37
0.23
off
1.03
0.10
0.12

50/100Hz vibration


Apart from the moderate effect of the A/C and possibly sheer randomness, this seems to be a more-or-less constant signal not strongly affected by the speed of the cart. There may be a stronger effect at low speed. I propose that it is delivered to the cart via the optics pusher motor power amplifiers which would be consistent with the AC signal I measured on the oscilloscope on what it supposed to be a DC power supply.
50Hz.png
--
100Hz.png

15 Hz vibration


This is believed to be a mechanical vibration of the optics car. It has also been measured with the accelerometer. It depends on the tracking speed of the cart, but it is not a linear dependency (which makes sense if certain speeds are closer to resonances than others). The RMS in the error signal seems to be linearly correlated with the 15Hz signal which seems to be the dominant vibration.
15Hz.png
--
rms.png
--
15Hz_rms.png
The dominant frequency in the mechanical vibration is 12.2 Hz. This should be slow enough for the piezos to manage, but the amplitude appears to be too large and/or the delays in the servo loop are too large.
12_2_Hz.png


48.8Hz vibration


There is a 48.8Hz (and possibly other frequencies) signal that grows almost exponentially with speed. This is clearly related to the mechanical vibration (4x12.2Hz).
48_8Hz.png

22.6 Hz vibration


There is a sharp 22.6Hz peak in all of the data when not tracking even when the cart is powered off (it was a bit higher at 23Hz in June). It seems to be a sinusoid with an amplitude of +/-2 steps (total 4/64 of a wavelength peak to trough). The piezos appear to take this out when tracking . . . actually, I think it is simply masked by the "dead zone" in the piezo servo loop. Possibly metrology signal variation? Counting electronics variability? Vibration from office A/C which cannot be remotely disabled? Anyway, the amplitude is small enough that it is not really something to worry about.

Red Herring Warning


I "discovered" a 12.5Hz sawtooth when the cart was stationary after tracking. The metrology oscilloscope traces looked rock solid, but the sawtooth persisted even when the cart was powered off. This was clearly related to the 80ms clock. What it is is the AV68K remembering its last astromod data which contained a position and velocity. So it updates the expected position every 2ms based on the velocity creating a false ramp in the error signal, and then every 40ms, it reverts back to the defined position creating a jump back.

Piezo Gain/FB and Damping


There is some evidence that the FB and damping are not optimally set. When tracking at 1 mm/s, the RMS in the delta error signal has a minimum around 0x43 whereas the default FB is 0x30. It is not clear, however, that the curve is consistent at different speeds; when tracking at 0.1 mm/s, the delta RMS goes up slightly when the FB is increased (more complete testing might be worthwhile). The RMS in the raw error signal continues to go down as FB is increased, but I suspect that over correcting is helping to smooth out the 12 Hz signal at such high gains. Mathematically, the ideal gain should be about 0x40 as there is a normalization factor of 0x20 in the opcon code as well as a spurious division by 2. The code also uses a piezo steps to metrology wavelength conversion of 74. Julian North reports in his thesis that he measured 64. The additional correction would reduce the mathematically ideal gain to 0x37.
RMS_fb.png

--
RMS_fb_tot.png
There is also evidence that reducing damping helps smooth out the 12 Hz vibration as it responds quicker. The raw error signal RMS goes down significantly at the expense of increasing the delta error signal RMS. Do I really need to make more plots?