• State Resolved
  • Locked Locked
  • Replies 8 replies
  • Answers 1 answer
  • Subscribers 32 subscribers
  • Views 2472 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

TMP 006 Noise Reduction Modes

Alex Behnaz
Prodigy 215 points
Other Parts Discussed in Thread: TMP006

Hello All,

I recently purchased a TMP 006 evaluation module, tested it, and have been reading up on these forums & the datasheet. The end application is external body temperature measurement, so extremely low noise is critical. Some of my test results are shown below.

Figure 1: The response curve of the TMP 006 when going through cycle 23-25-27-30-33 degrees C

Figure 2: Zoomed in on noise at 30 degrees C step

 

These tests were performed under the following conditions:

- 4 SPS ("1 average")

- IR source is a calibrated black body rated for calibration of thermal cameras

- TMP 006 evaluation board powered and connected by USB to laptop

- TMP 006 evaluation board placed approx 1.5 inches from the black body in free room conditions (no coverings, exposed to external room environment)

 

I am interested in figuring out if there is some way that I can reduce this noise *without* filtering (analog or DSP). My questions are as follows:

1) On the evaluation kit, how is the filtering done? I assume that the TMP 006 simply takes multiple readings, and the software then does an average.

2) How well is the USB power supply filtered? I know that USB power tends to be very poorly regulated. I was thinking of using a ferrite bead and TPS7A49xx ultra low noise regulator to feed the TMP-006. Will a high-quality power supply with very low noise and high PSRR significantly improve noise response?

3) I understand from earlier posts that better readings can be obtained by shielding the sensor from external drafts (air currents) with say a plastic bag. Will this affect the sensor's performance in any way, or will it simply improve it by reducing noise?

I plan to construct a serious scientific solution soon. Any input on this would be appreciated.

Thanks,

Alex Behnaz

  • 0
    TI__Mastermind 27715 points

    Hello Alex,

    I'll first answer your three questions and then offer some other recommendations.

    1) On the evaluation kit, how is the filtering done? I assume that the TMP 006 simply takes multiple readings, and the software then does an average.

    Actually, the averaging is performed inside the TMP006. The software simply calculates the object temperature result based on the data in the TMP006 sensor voltage and ambient temperature registers. Selecting a slower conversion rate that takes more samples can potentially cut your noise in half. See the table extracted from the TMP006 data sheet below.

    There is a way to reduce errors/noise associated with transients in the local temperature by using a simple correction algorithm; the new version of the TMP006EVM software (very soon to be released) will include this feature. I've attached a few slides which describe the transient correction procedure in more detail.

    2) How well is the USB power supply filtered? I know that USB power tends to be very poorly regulated. I was thinking of using a ferrite bead and TPS7A49xx ultra low noise regulator to feed the TMP-006. Will a high-quality power supply with very low noise and high PSRR significantly improve noise response?

    The power supply on the TMP006 is provided by the SM-USB-DIG by default, and the SM-USB-DIG uses the TPS76333 regulator to generate the +3.3V rail off of the +5V USB power line. The TPS76333 is designed for low dropout and low power, not low noise, and I observed AC noise of 120mVpp at its output under normal operation (see the scope capture below). The TPS7A49xx will definitely give you a cleaner power rail, but the benefit to the TMP006's performance will most likely be negligible. The PSRR of the TMP006 is given as 0.1°C/V, so our observed 120mV of noise would then cause an error of 0.012°C. This value is almost 3x less than the resolution of the part at 0.03125°C. 

    3) I understand from earlier posts that better readings can be obtained by shielding the sensor from external drafts (air currents) with say a plastic bag. Will this affect the sensor's performance in any way, or will it simply improve it by reducing noise?

    The TMP006 is indeed sensitive to air flow across the sensor and in the general area of the PCB on which it is mounted, since air flow causes local temperature transients and can also cause a temperature gradient to develop between the TMP006 die and the PCB which results in increased sensor offset voltages and therefore object temperature errors. 

    We are currently performing a series of tests with various poly-IR and metallic covers and lenses in order to find the best solution to protect the device, but in the meantime any sort of enclosure you can use to shield the EVM from air currents (while still retaining line of sight between the TMP006 and the black body) should help reduce your noise.

    Other recommendations/questions:
    1. Have you performed a calibration on your system in order to determine S0, or are you using the default value on the EVM? Using the proper S0 is crucial to achieving the best accuracy and noise performance. See Section 6, "Calibrating the System" in the TMP006 User's Guide (http://www.ti.com/lit/ug/sbou107/sbou107.pdf) for more information on how to perform a two-point calibration and calculation of S0.

    2. Your current test setup does not involve any local temperature regulation. Stable local temperature is a must for achieving the best possible accuracy and noise performance. For our internal testing and characterization we use temperature controllers by Silicon Thermal (http://www.siliconthermal.com/).

    3. Are you using any sort of clamp or other hardware to keep the TMP006EVM in place? We've noticed that that shiny metallic objects, such as screws used to mount our test board, can cause IR scatter that increases noise. We've begun using two-sided thermally conductive tape to mount our TMP006 test boards and have found that our noise has improved significantly. Other methods, such as painting the offending objects matte black, work as well.

    Best regards,

    Ian Williams
    Linear Applications Engineer
    High Performance Linear and Sensing Products 

    TMP006 Transient Correction.pdf
  • 0 in reply to Ian Williams
    Prodigy 215 points

    Ian,

    Thanks for the quick reply.

    1) Have you performed a calibration on your system in order to determine S0, or are you using the default value on the EVM? Using the proper S0 is crucial to achieving the best accuracy and noise performance.

    I simply adjusted the So value to match my known temperature. I don't really see how this would affect noise. It looks like it affects accuracy more than anything else.

    2) Your current test setup does not involve any local temperature regulation. Stable local temperature is a must for achieving the best possible accuracy and noise performance.

    In the final device, the patient will be wearing the system. I won't be able to regulate the temperature in any way.

    3) Are you using any sort of clamp or other hardware to keep the TMP006EVM in place? We've noticed that that shiny metallic objects, such as screws used to mount our test board, can cause IR scatter that increases noise.

    I have clamps that are covered in thermally absorbing material to prevent scatter.

     

    Beyond these three points, I also have a question about the internal hardware:

    On pg. 10 of the datasheet it says: 1 LSB = 1/32°C = 0.03125. Is this over the full range of -10 C to 30 C? Is it possible to change the hardware to adjust the dynamic range to include a certain pre-known temperature range? For example, if I am interested in the 20 C to 30 C range, can I modify the amplifier gain (or something else) to increase sensitivity?


    Thanks,

    Alex

  • 0 in reply to Alex Behnaz
    TI__Mastermind 27715 points

    Alex,

    You asked:

    "On pg. 10 of the datasheet it says: 1 LSB = 1/32°C = 0.03125. Is this over the full range of -10 C to 30 C? Is it possible to change the hardware to adjust the dynamic range to include a certain pre-known temperature range? For example, if I am interested in the 20 C to 30 C range, can I modify the amplifier gain (or something else) to increase sensitivity?"

    The resolution of the part does not change over the temperature range. I'd also like to clarify that the -10°C to 30°C range is the optimal range of TOBJ - TA, not the range of target object temperature. For optimal performance in your limited measurement range, I'd recommend a two-point calibration at the endpoints of your span of interest - in this case with TOBJ of 20°C to 30°C. The S0 term modifies the internal amp gain to compensate for any IR signal losses, whether they are a result of field of view limitations, target object emissivity, attenuation due to any lens or cover material used, or any other system variables.

    "In the final device, the patient will be wearing the system. I won't be able to regulate the temperature in any way."

    In this case, you will see the target object temperature result vary as the local temperature changes. I'm curious as to the mechanics/application of your end system; will the user be wearing a small device where the TMP006 is installed such that the sensor is pointing towards the user's skin? If so, what sort of distances are involved?

    Best regards,

    Ian Williams

  • 0 in reply to Ian Williams
    Prodigy 215 points

    Ian,

    As always, thanks for the quick replies.

    Yes, the sensor will be placed in an enclosure (probably a fabric of some sort) and the TMP006 will be aimed towards the patient's skin. I suspect that the distances involved will be in the range of 1-5 mm.

    I am still doing tests on the setup. I have been able to get a good response for oscillations with 1C amplitudes. I am hoping to resolve down to 0.01 C without too much noise. I don't need a perfect time domain waveform as much as I need to have the frequency domain peak standout from the noise floor. Do you have an idea of when the new EVM software will be out? Do you think that the filter on the local temperature will siginficantly improve the response?

    Thanks,

    Alex

     

  • 0 in reply to Alex Behnaz
    TI__Mastermind 27715 points

    Hello Alex,

    I am waiting on a colleague to make some other additions to the software, but I expect him to finish within the next couple of weeks. I plan to create a short blog post on the main page of the Temperature Sensors area to notify the community when the update goes live, since many people are using TMP006EVMs and will benefit from the new version.

    As for whether the algorithm will significantly improve the response of your application, I believe it will help but I cannot say to what degree. Like I mentioned previously, the new transient correction algorithm has the most value in applications such as monitoring IC temperature, where spikes in activity/load can cause rapidly changing local temperatures. I'd imagine that the skin temperature of a human has a much slower rate of change. Either way, you should see some improvement since in your system since the extra averaging tends to improve the noise response in all applications.

    Best regards,

    Ian Williams

  • 0 in reply to Ian Williams
    Prodigy 10 points

    Hello Ian

    I am also interested in using this device for body temp measurements.  I suspect that Alex might have the same application in mind - it's a type of neurofeedback known as pirHEG, where HEG = HemoEncephaloGraph.  One measures the small temperature variations on the forehead, these are produced by blood flow and brain "activity".  For this application the absolute temperature is not so important, but one needs to track the relative movement with a VERY high resolution.  The current systems can resolve down to 0.01 farenheit... so a system resolution of 0.01 deg celcius is about the minimum required.  I am of the opion that the TMP006 will just not be good enough for this application, simply because the LSB = 0.03125 deg C.

    The reason for my writing is this: I have read the TMP006 is the first in a series of products...  On my which list would be a device where one could perhaps change the gain of the sensor amplifier, either by means of an external resistor or a PGA controlled via I2C.  What sort of future devices are on the horizon?  Elsewhere I have read that you do NOT plan to make devices available where the gain and offsets will be changed?

    Regards

    Stefan Jung

  • 0 in reply to Stefan Jung
    Prodigy 215 points

    Stefan,

    I had a conversation with TI engineers. It seems like their next generation is focusing on creating a sensor that essentially automates the calculations and spits out a temperature.

    I tested the TMP006 under "scientific conditions" and found that the major issue was temperature resolution and noise. I am thinking of doing further tests, but right now I don't think the TMP006 is designed for precision measurements like the ones we need. I have noticed that thermopiles do tend to be quite noisy, and that larger area devices seem to alleivate some of that noise.

    Perhaps TI might consider adding these design objectives to one of the family depending on market demand.

    Alex

  • 0 in reply to Ian Williams
    Genius 5430 points

    Ian;

    These questions posted about measuring skin temperature are so numerous--together with other moderate-temperature thermometry questions-- that it is clear that the decision to use a cover that cuts off the longer wavelength response of this detector needs rethinking. Marketing should have thought of this and insisted on a longer wavelength response.

    Look at the Rayleigh-Jeans curve of radiation emission vs wavelength. Not much is radiated from an object at skin temperature. You are competing with microbolometers and thermocouples but those do not have gain or signal processing. Change the window so that TI can offer a version with enhanced long wavelength response.