Connect with us

15ps and falling

Discussion in 'Electronic Design' started by colin, Jun 24, 2007.

Scroll to continue with content
  1. colin

    colin Guest

    Im measuring the time difference between two signals from two opto encoders
    basicaly on the same shaft, at the moment ive got down to a peak at 15ps.
    averaged over a few days, but I assume this is still mostly noise.

    I need to get down to a noise floor of 0.1ps. although im only running at
    1krpm I hope to run at 10krpm,
    but need to do some balancing and some more noise insulating.
    if only I could run at 100krpm but my mech engineering abilites are a bit

    this should give 1/10 the error for the same mechanical error,
    although the vibration might not help much,
    but the increased number of samples helps reduce the noise by averaging.

    I also need to look at different bearings again.

    remagnetizing the motor rotor with a sinewave instead of squarewave pattern
    helped a fair bit.
    and using an accelerometer to get it level to within 500ug helped too.
    although it tends to wander off being level througout the day,
    maybe I need to set it up in an optics lab or something but that is unlikly,
    a better table would help that wasnt on wooden floorboards,
    but not sure what I can do about that.

    Colin =^.^=
  2. whit3rd

    whit3rd Guest

    Yow! Most optical sensors are capacitive output impedance, with
    photomultipliers and maybe avalanche diodes the exception. Signal
    to noise is gonna kill it at small timescale.

    The classic gear-tooth-sensor with a ferromagnetic gear tooth
    also has field-diffusion-rate limits, BUT a brass gear variant using
    diamagnetism should hold up to fast toggling. That'd be what
    I'd be looking for. If the gear teeth were uniform enough, you can
    phaselock and get to serious frequencies that way.
    There was a brief time (NeXT anyone?) when magnetooptic disks
    seemed like a good idea. But, the optic readout couldn't keep up
    with magnetic readout technologies, and all fast disks now are
  3. colin

    colin Guest

    yeah but the dominant noise seems to be of mechanical origin.
    bearing play, torsional vibrations, general flexing etc.
    so just lots of samples to overcome this.

    even the interval is only measured with a 40mhz clock too,
    the time is averaged with 4000 pulses per rev.

    the faster the speed the lower the time error for the same mechanical error
    in angle.

    Colin =^.^=
  4. whit3rd

    whit3rd Guest

    There's some state-of-the-art bearings (hydraulic bearings; google
    for "moglice") that might do some good, but the ultimate
    is levitation. Rogers Ritter and Jesse Beams (at U. Va) have
    made working systems, should be easy to find in the physics
  5. colin

    colin Guest

    Yeah I was thinking about magnetic bearings too,
    probably making some.

    its hard to be sure what proportion of the error is originating where.

    Colin =^.^=
  6. MooseFET

    MooseFET Guest

    I PMT plate looks like a very high resistance in parallel with a
    capacitor. The capacitance limits the frequency responce.
  7. colin

    colin Guest

    I have 5.76e9 samples per day so I get a lot of the resolution from
    3 days will give me about 0.1ps for 25ns resolution per single pulse.

    the mechanical noise is a lot worse than this atm.
    the noise from comparing the two channels from one encoder is quite low.

    Colin =^.^=
  8. MooseFET

    MooseFET Guest

    Yes, that sort of method is quite a good way to improve the
    measurement. I have had a few troubles when doing this so I will
    suggest them as things you need to watch out for.

    If the experiment and the measurement system can crosstalk, the really
    high frequencies from the experiment can sneek into the clock circuit
    of the measurement system. This makes the two try to lock together
    and modulates the time base.

    The measurement assumes that the timing of the pulse slides to all
    posible times in the cycle of the timebase. This is almost always
    true but sometimes gremlins will get in and make it not true. These
    gremlins only seem to act when the lights are off over night.
  9. colin

    colin Guest


    fortunatly there is only one clock. I have considered using an atomic clock.
    although the electronics is actually rotating,
    atomic clocks are affected by magnetic field,
    crystal clocks are affected by gravity/tilt,
    although the clock error would have to be quite large to give a significant
    error in the reading that wasnt canceled out by other means.
    The BLDC motor is driven by the same clock that measures the pulse,
    however the phase is not fixed, and the speed varies by a small error,
    so this should hold true, plus the optical discs have a small but detectable
    unintended eccentircity,
    so this gaurantees a reasonable slide, eqv to several pulses.

    I did have a problem with day light though, a small amount geting through to
    the optos was enough to introduce a ~1ns error, especialy when the sun shone
    in through the window at dawn.

    I have come accross all sorts of problems on the way so far ...

    Colin =^.^=
Ask a Question
Want to reply to this thread or ask your own question?
You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.
Electronics Point Logo
Continue to site
Quote of the day