1. This forum section is a read-only archive which contains old newsgroup posts. If you wish to post a query, please do so in one of our main forum sections (here). This way you will get a faster, better response from the members on Motherboard Point.

Faster temperature measurement when temperature compensating?

Discussion in 'Embedded' started by Robert Scott, Nov 8, 2006.

  1. Robert Scott

    Robert Scott Guest

    You could calibrate "on the fly", but the accuracy of your calibration would
    suffer. You can model the sensor as a 1-pole (simple RC) filter. If you take a
    sample before a step change and several samples during the response to a step
    change, you might be able to find a best-fit exponential curve for that data.
    Extrapolate its asymtotic value and you have it. It would be good to make sure
    your samples cover at least one time-constant of the exponential approach so
    that you know you have a real sample that is at least 63% of the way to the true
    asymtotic value.


    Robert Scott
    Ypsilanti, Michigan
     
    Robert Scott, Nov 8, 2006
    #1
    1. Advertisements

  2. Robert Scott

    HJ Guest

    How can I speed up the procedure when I temperature calibrate unit X? Must I
    wait for the temp to stabilize at a given temperature before measuring or
    is it possible to measure "on the fly" as the temp increses and not waiting
    for it to stabilize?

    The temp sensor is internal to the device (in the same IC package).
     
    HJ, Nov 8, 2006
    #2
    1. Advertisements

  3. Robert Scott

    Arlet Guest

    If the temperature is rising, and your reference thermometer reacts
    quicker or slower to this change than the device you're calibrating,
    you will introduce an error. Whether this will be acceptable depends
    on the test setup, the rate of change, and the accuracy you want. For
    modest accuracy requirements it may work.

    If you want to get an idea of the error, first do a proper calibration
    where you let the temperature stabilize, and then do an "on the fly"
    comparison.
     
    Arlet, Nov 8, 2006
    #3
  4. It depends on what accuracy do you need.

    Must I
    The temperature gradients of several degrees C are typical. It is very
    difficult to get the gradients to 0.1C or below.

    Vladimir Vassilevsky

    DSP and Mixed Signal Design Consultant

    http://www.abvolt.com
     
    Vladimir Vassilevsky, Nov 8, 2006
    #4

  5. This will provide for the accuracy of no better then several degrees C.
    It is very difficult to have a consistency and a control for all kinds
    of the temperature gradients at the board level.

    Vladimir Vassilevsky

    DSP and Mixed Signal Design Consultant

    http://www.abvolt.com
     
    Vladimir Vassilevsky, Nov 8, 2006
    #5
  6. For a single lag T = To + (Tf - To) * (1 - exp((t-t0)/tau))

    Solve for Tf* and if you know tau (which should be fairly consistent
    if the assembly is consistent and cleaned well) and you can predict Tf
    given arbitrary t and To. As Robert says, you want t-t0 to be at least
    tau to have any hope of getting a good extrapolation, and of course
    the temperature difference has to be large in comparison to the
    resolution and stability of your measurement.


    * Tf = To * (T - T0)/( 1 - exp((t-t0)/tau))

    If it's much more than a single lag, which depends on the physical
    configuration, you'll probably have to do some kind of numerical fit
    and I have more doubt as to the ability to improve things enough to be
    worth the trouble.


    Best regards,
    Spehro Pefhany
     
    Spehro Pefhany, Nov 8, 2006
    #6
  7. Robert Scott

    Robert Scott Guest

    Not necessarily. It is possible that the sensor might be taken from a storage
    area that is already not too far from the temperature of the calibration
    chamber. In that case you would not have to deal with "all kinds of temperature
    gradients". Instead you would have a fairly repeatable starting condition. The
    on-the-fly extrapolation would only be used to make a relatively small
    correction. The accuracy of that extrapolation will be nearly proportional to
    the difference between the storage area temperature and the calibration chamber
    temperature. Without knowing this difference you cannot conclude that the
    extrapolation can never be more than "several degrees C".

    Of course if the starting condition is totally unknown, then accuracy will be
    poor. But if this calibration is part of a manufacturing process that is going
    to be done over and over again, then several improvements can be made. One is
    to make the starting condition as uniform as possible as described above.
    Another method was suggested by an earlier responder. That is to subject a few
    units to the extra-long stabilization time in order to develop a generic model
    for the extrapolation. There is no reason to treat each unit as a brand new
    problem. Knowledge gained from a few extended calibration trials can be used to
    refine the lag model.


    Robert Scott
    Ypsilanti, Michigan
     
    Robert Scott, Nov 8, 2006
    #7
  8. You see, Robert, I had to deal with this problem before.


    In that case you would not have to deal with "all kinds of temperature

    It all depends on what place on the shelf in the calibration chamber is
    the particular unit, how many units do you have in the chamber, and even
    what is the weather today (since the efficiency of the cooling/heating
    depends on the conditions outside). As for the board level problems, a
    single LED creates ~50mWt of heat, which is enough to create a gradient
    of 1C in the area of 1 cm.


    The
    Usually is not worth it, since it is very dependent on everything.
    Just select a reasonable heating/cooling speed and hold time.


    Vladimir Vassilevsky

    DSP and Mixed Signal Design Consultant

    http://www.abvolt.com
     
    Vladimir Vassilevsky, Nov 8, 2006
    #8
  9. That depends entirely on how precise you want the measurement to be.
    The longer you wait, the better the chance that you won't have any
    systematic error in the final result.

    Now, if you know your device and its environment really well, it can be
    possible to do a "ballistic" measurement, i.e. observe the "flight path"
    of temperature over time, to extract data like thermal capacitance and
    resistance between the thermometer and the point where you really want
    to know the temperature, so you can predict arithmetically where this
    path will eventually end. But given that you're trying to "calibrate"
    something, it's rather unlikely that this can work.
     
    =?ISO-8859-1?Q?Hans-Bernhard_Br=F6ker?=, Nov 8, 2006
    #9
    1. Advertisements

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.