Connect with us

Temp sensor "accuracy"?

Discussion in 'Electronic Design' started by Jon Slaughter, Dec 28, 2007.

Scroll to continue with content
  1. What do they mean by accuracy and resolution here?

    It says +-2.0C accuracy then a resolution of 0.0625C.

    Does this mean I can compensate for the error in the accuracy by using a
    more accurate device? So say I'm measuring a 100C source and its reading 98C
    then I know I'm off by -2.0C and I can just compensate for that in all my
    calculations? Is there any issues with this(like maybe drift)?

  2. Also does anyone know of any similar devices with better accuracy?
  3. Bruce Varley

    Bruce Varley Guest

    The spec doesn't talk about drift, it's not an easy aspect to specify, are
    you talking over an hour, or a week, or 20 years? And do you expect the
    devices to remain stable despite possible adverse conditions such as
    transient temperature extremes or vibration?

    In my fairly extended experience with industrial temperature measurement, if
    you want to really hold accuracy around the 1C mark, you have to *work* at
    it, whatever the spec sheets say. That includes calibration back to primary
    sources regularly. So think about how much accuracy you really need, it'll
    cost you one way or the other.
  4. John Larkin

    John Larkin Guest

    Thinfilm platinum RTDs are available as little ceramic slabs with
    leads, or as surface-mount things that look like resistors. They are
    very accurate, typically around 0.1C maybe. But you have to measure
    their resistance accurately.

    Somebody makes digital (spi serial readout) temp sensors that are 0.5
    C accurate, I think. National? ADI? Can't remember.

  5. Yes, you can compensate the error to some extent. However it would be very
    naive to expect the long term accuracy of better then 1C from a
    semiconductor sensor.

    Drift, self heating, aging.
    If better accuracy is required, you need RTD.

  6. Robbo

    Robbo Guest

    0.5C accuracy - but it's 1-Wire, not I2C. Do you neet to measure 100C or was
    that just a nominal value?

    "The DS18S20 Digital thermometer provides 9-bit centigrade temperature
    measurements and has an alarm function with nonvolatile user-programmable
    upper and lower trigger points. The DS18S20 communicates over a 1-Wire® bus
    that by definition requires only one data line (and ground) for
    communication with a central microprocessor. It has an operating temperature
    range of -55°C to +125°C and is accurate to ±0.5°C over the range of -10°C
    to +85°C. In addition, the DS18S20 can derive power directly from the data
    line ("parasite power"), eliminating the need for an external power supply."
  7. Robert Baer

    Robert Baer Guest

    First, assume that compensation is not linear.
    Second, assume that no two sensors are alike.
    You can depend on those assumptions until precise measurements WRT
    them indicate otherwise.
    "Resolution" in this case is like having 5 digit data from a 3 digit
    (repeatable? non-repeatable?) source.
    Here, with that kind of resolution, it is fair to assume "repeatable"
    within the "resoultion".
    BUT, if you need something like that, then you must verify that from
    a reasonable standard.
  8. whit3rd

    whit3rd Guest

    Accuracy and resolution (sometimes called precision) have the normal
    meanings here. The 'accuracy' is limited by high-speed test and
    calibration conditions at the factory, while the resolution is limited
    by expected
    non-temperature environmental factors (stress, age), and noise.
    No, not really; you can CALIBRATE it using a more accurate device,
    though. Your calibration will take place when the first weeks of
    aging have completed, so might be more effective than the production-
    calibration was.

    In my experience, calibration (of resistors and capacitors) lapses
    soldering the little components down. So, you don't want to buy the
    gizmo with good calibration, so much as have a plan to calibrate it
    after it completes its stressful insertion/soldering phases.

    Most semiconductor thermal sensors use logarithmic I/V characteristics
    of a planar junction, and have a good fit to the 'ideal' curve,
    precision about
    .001 degree K, with only a single (scale) factor to be trimmed to
    complete the calibration.

  10. Guest

    "Interchangeable" thermistors from Yellow Springs Instruments,
    Betatherm, Thermometrics and so forth are available with accuracies up
    to +/-0.05C - though a thermistor that accurate is going to cost
    around $50. Parts with an accuracy of +/-0.2C are widely available at
    a tenth of that price or less.

    Thermistors aren't all that easy to use. The calibration curve is non-
    linear - see the Steenhart-Hart fitting function - and, because they
    are negative temperature coefficient devices, their resistance goes
    unstable if you dissipate too much heat in the part. In my experience
    10 uW is okay, 100uW can be risky. But their resistance decreases by
    roughl 4% for every degree Celcius that the junction gets warmer,
    which makes them gratifyingly sensitive.

    Platinum resistance sensors are an order of magnitude less sensitive,
    and semi-conductor sensors are appreciably noisier.
  11. Its not digital, but the LM135 Kelvin sensors can be adjusted to within +/- 1°C
    with a simple trimpot.
    Also converts easy to Celcius:
    Celcius = 25 + 100 x (voltage - 2.982)
    Range is -55°C to +150°C
    Rest depends on your ADC resolution and math.
  12. Chuck

    Chuck Guest

    One issue not addressed is understanding
    what you mean by the temperature of
    something. Common objects are likely to
    reveal different temperatures on their
    surfaces and their interiors depending
    on location. With crude instruments this
    often isn't noticed. But with fractional
    degree resolution, it may.

    Imagine using a micrometer (albeit a
    large one) to measure the length of a
    piece of firewood. You have abundant
    accuracy and precision, but misleading
    results unless you map the entire end
    surfaces, a process that could even
    change the measurements.

  13. Henry Kiefer

    Henry Kiefer Guest

    Of good performance is the SMT160 from
    Can be PWM or analog out.

    - Henry
  14. neon


    Oct 21, 2006
    acuracy of +/- 2% means just that at 98c it could be 98 to 102 no guaranty of anything in between. can you adjust to it sure but then again buy one that is accurate to +/- to whatever you require. to adjust it to read 100c you must have another instrument that you can trust to read 100 c acurately resolution means that you will be able to tell the temperature to withing .0625 C at 98 it can read 98.0625c to 102.0625c
  15. RRogers

    RRogers Guest

    We are getting medical thermistors with a claimed accuracy of .05degC
    for about $2.00 at a quantity of 4,000. Unfortunately I am on
    vacation and can't supply the name of the manufacturer. The reason I
    said medical is that we only require accuracy at 37degC; I don't have
    the information for other temperatures.

  16. DeadCat

    DeadCat Guest

    For the fairly non-demanding task of measuring air temperature in my garden,
    I've used some of these:

    They have a quoted accuracy of ±0.5°C and a resolution of 0.0625C. The
    accuracy figure presumably only applies at the point it leaves the factory.
    I don't know whether the difference between the reported and actual
    temperatures varies long-term, but the difference seems to be fairly
    constant on a timescale of minutes, because the graphs of temperature in my
    garden (see below) vary fairly smoothly, rather than jumping about in a
    ±0.5°C range.
  17. Phil Hobbs

    Phil Hobbs Guest

    The main problems with IC temperature sensors are the low thermal
    conductivity of the package and the high thermal conductivity of the

    Plastic has a thermal conductivity in the ballpark of 0.1 W/m/K, whereas
    copper's is 400 W/m/K, so it's really really hard not to wind up
    measuring the temperature of the circuit board rather than the air
    temperature, and even if you manage that somehow, their thermal time
    constant is many seconds.

    Yellow Springs International (YSI) sells glass bead thermistors with
    time constants in the 100 ms range and interchangeability at the 0.2 K
    level. Glass bead thermistors are very sensitive, and also very
    stable--a few ppm per year for the best ones, as long as you don't
    stress the package.


    Phil Hobbs


    Phil Hobbs
  18. Henry Kiefer

    Henry Kiefer Guest

    You can set the pcb to equilibrium temperature if the copper space is
    big enough on the pins.

    The main problem is mechanical stress in the semiconductor package.
    Other problem is the testing time the manufacturer allows having
    financial benefit.

    You can expect 0,1K drift in a couple of years for semiconductor
    temperature sensors.

    - Henry
  19. Phil Hobbs

    Phil Hobbs Guest

    What do you mean by 'equilibrium temperature'? It sounds as though
    you're trying to make the problem worse rather than better. Bigger pads
    just increase the thermal forcing from the board.


    Phil Hobbs
  20. Henry Kiefer

    Henry Kiefer Guest

    I'm not sure what you mean. If you want to measure board temperature
    then increase pad area, if you want to measure the air temp then
    increase the sensor surface with a heatsink for example.

    I think this is logical for an engineer.

    - Henry
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