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.

EK-TM4C123GXL: Hardware problem with ADC generating noise?

Prodigy 120 points

Replies: 19

Views: 5842

Part Number: EK-TM4C123GXL

The non-profit I work for (Smoky Mountain Scientific, smokymtsci.com) develops and sells low cost, open source instruments for science education.  One of our instruments incorporates a Tiva-C launchpad that outputs a pwm voltage and measures voltage and current passed in electrochemical reactions.  I recently purchased 100 Tiva launchpads from Mouser and I am having trouble with noise in the analog read channels for about 1/4th of the boards.  We have been building and selling these instruments for a couple of years now and have never had this problem before.    To test the boards, I send a pwm signal through a simple low pass filter (a resistor and a capacitor) then read the output on two channels (A11 and A8).  The attached figures show examples of output from a good (file Board46) and a bad (file Board37) boards as displayed on our user interface.  I put the apparatus together on a grounding mat and then plug the usb connector into my computer to flash the microcontroller and do the testing.  I have posted a description of this on the 43oh web site (forum.43oh.com/.../ and have contacted Mouser about it.  Mouser said that I needed to bring the problem to the e2e forum before they could do anything.  Any recommendations?
Thanks,

Jack

  instruments for

 

19 Replies

  • What is the error in counts?

    Robert
  • In reply to Robert Adsett:

    Looking at the COM37 figure, it appears that the noise on the y-axis is ~ 50 mV. There is a threefold conversion between the measured and reported values (to account for amplification in the final circuit), so the noise at the ADC would be ~17 mV. So, 0.017*4095/3.3 ~ 21 counts. That may not seem like a whole lot, but I am averaging 64 measurements to get each point. Therefore, the noise in each measurement is pretty large.
  • Jack,

    The "non-profit" mention is honorable, and from your description, the project seems quite interesting!

    However, they do publish these terms for a reason...

    Document link here: www.ti.com/.../sszz027m.pdf

    The actual suggestion here, while not that sweet, is to design a board for the purpose, with proper analog signals care.

    Regards

    Bruno

  • In reply to Bruno Saraiva:

    Dang, I was unaware of that.
    While I would like to migrate this project to be free of the development board, I was trained as a chemist and I do not posses the knowledge and experience to develop a stand alone board. If someone reading this thread is willing to guide me through that development effort, I can be reached at summers at wcu dot edu.
  • In reply to Jack Summers11:

    Jack Summers11
    the noise at the ADC would be ~17 mV. So, 0.017*4095/3.3 ~ 21 counts. That may not seem like a whole lot,

    21 counts isn't a lot. Two points from the data sheet to note

    First total error over range

    Datasheet

    ET Total unadjusted error, over full input range  ±30.0 LSB

    Second for noise, single ended

    Datasheet

    SNRS Signal-to-noise-ratio, Single-ended input, 60  dB

    Even the typical noise figure is just 65 db (~10.5bits). Differential input gets as high as 72dB typical (~11.7 bits)

    Then you have to consider whether your reference is stable.

    Robert

  • In reply to Jack Summers11:

    Jack,
    Being trained as a chemist, you do show some good knowledge of signal analysis already!
    Don't get too desperate, I've been educated as a mechanical engineer (that degree after which wives assume you will be able to fix a modern car if it suddenly stalls on the road shoulder)... And yet, here I am talking (reasonably correct) EE matters.
    It won't be less than a good year to be able to design by yourself a schematic/circuit from scratch with proper quality. But the very Launchpad manual has all its schematics and even board layout (I believe the latter can be downloaded somewhere). They are built in Eagle (lots of design software out there...)
    Regards
    Bruno

  • Might you have presented a "key" clue - which thus far - sits/kicked (roadside)    [clearly awaiting M.E. Bruno]    while bruised & abandoned?

    Jack Summers11
    ... having trouble with noise ... for about 1/4th of the boards ... AND building/selling these for a couple of years - have never had this problem!

    So - "What's changed" - either on (your) end or this vendor's?     For now - we must accept your report as "true & properly reflective of reality" - yet that proves (not always) the case...

    These questions spring from that "key clue:"

    • How sure are you that this "issue" had NOT occurred at a similar frequency - in the past?      (i.e. were your "past" tests (really) able to detect - and alarm - such issues?)
    • What's changed on your end?      Input analog circuitry, test protocol, added or altered product software, test set-up, board handling (throughout all venues of your production/test protocol)
    • What's changed on the vendor's board end?     Ideally - even a "non-profit" - should maintain a series of, "Golden Boards" (pristine boards - which passed all tests - and then are safely placed w/in "Quarantine" so they may (later) be "compared" against boards "failing or w/issues."
    • Should you have your own "good" boards - should you not rigorously "A-B" test/inspect/compare/contrast those w/your new boards - which fail?
    • Similarly - cannot you compare "good" vendor boards - just as listed above - w/those vendor boards which have "issues?"
    • Has your "test set-up" continued - unchanged - between past dates of "success" - and today's "issues?"     Might the "C" of your R-C (LPF) have degraded - changed value?
    • Beyond your (mention) of the R-C (LPF) all here are, "in the dark" as to the extent of your (unique) processing electronics - which pass signals to the vendor's LPad.     May I suggest that the vendor's board is less likely to "deviate" than is yours?     (due to the volume & controls placed upon vendor's production - yours stand "unknown.")

    It should be both simple & quick to - one at a time - swap boards "In-Out" - and "lock the issue down" to:

    • your unique boards  (most likely)
    • vendor's boards   (unlikely)
    • both your & vendor's boards   (most unlikely)

    Identifying the "source" of your problem is Job #1.     Once identified - corrective action may commence.      (btw - how kind of Mouser to "shunt you here" for rescue.      Should they not - at minimum -  "offer dinner" to Robert, Bruno & myself before "ordering up" our combined, time/effort?)

  • In reply to cb1_mobile:

    One more item of note (I cannot (now) "edit" the above post) - might your issue reflect the "reality" of "More & Stronger" RF signals in your near vicinity? Cell-phones, Tablets, IOT - even a (stronger) local radio or TV station - may induce such noise!

    Again - the value of proper, "A-B" Testing - will best shine great light upon the "reality" of your issue...

    And...yet another (obvious) item I earlier missed: "Reverse the board's connected to Ports 37 & 46 - even the difference between Ports must be held, "suspect."

  • In reply to Robert Adsett:

    Thanks for the input Robert.  I am not sure that I follow what you are saying.  Total error has to include both systematic and random error, and systematic error would not contribute to the problem shown in the figure.  Not sure what is meant by single ended noise.  I am not using an external reference and assume that the internal references in the launchpads should be similar.     

    That notwithstanding, you have pushed me down what may be a profitable direction.  To check the noise on the ADCs I made a voltage divider with two 20 KOhm resistors, plugged one end into VCC, the other into GND and tested A8 and A11 from the center of the voltage divider.  I modified an Energia sketch to read the two ADCs and output to the serial port every few ms. After a few seconds I stopped the scrolling text and copied the values into excel.  I used excel to determined the mean and standard deviation of the data.  To my surprise, all of the boards I tested (both those that performed well and those that did not) gave data with a standard deviation of ~13 counts.       From this, I surmise that the problem is not with the ADCs. 

    I changed the test circuit back to the low pass filter and tested the effects of changing the pwm clock speed and the delay between readings.  So far, I have not found the factor that differentiates the two sets of boards. 

    To complicate matters further, some of the boards that originally tested bad are now showing the desired behavior (perhaps the software did not load properly the first time?). 

    I dont know what is going on.  I will keep testing and let you know what i find.

    Thanks again.

    Jack

  • In reply to cb1_mobile:

    Hi cb1_moble. I test every copy of the instrument before I ship it. The test protocol involves attaching a test resistor across the electrode leads, ramping the voltage across the resistor and observing the resulting I versus V plot. I take a screen shot each time and send it via email to the customer. Like you, I originally suspected that the problem was with our boards. After considerable testing, I removed our board and tested the Launchpads with a passive low pass filter circuit. Since the problem persisted in the absence of our board, I concluded that the problem lay with the launchpad ADCs. Subsequent testing raised doubts about that. Further tests will hopefully bring clarity.
    regarding a couple of your points:
    "Similarly - cannot you compare "good" vendor boards - just as listed above - w/those vendor boards which have "issues?"". That is exactly what I am trying to do.
    "Might the "C" of your R-C (LPF) have degraded - changed value?". This seems unlikely. The cap would have to selectively fail for testing with the 'bad' boards and not fail for the 'good' boards. Also, the original test was with the board that we make.
    "... all here are, "in the dark" as to the extent of your (unique) processing electronics - which pass signals to the vendor's LPad. " As I said, this is open source stuff. Check the description of the hardware here: publiclab.org/.../new-wheestat-hardware. I will provide eagle files if you would like.

    Jack

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.