Output of AD8436

Hi everyone,

we are designing a true rms circuit using the AD8436 (see attached pdf).

The Circuits uses a wide input range of AD8436 to measure input voltages from 0-250V RMS with different input frequencies  An external opmamp circuit adjusts the values to the one specified in AD8436's datasheet. The Circuits works but it does not deliver the right values.

The output of the TRMS device always delivers a much to low value. When I remove the 3.3uF capacitor of the low pass filter I can see that the peak value of signal corresponds to the expected RMS value. But as soon as I activate the filter again the value drops to much. The datasheet states 0.5% error but I reach 2%-3.6% depending on the input voltage. The higher the input voltage is the higher the error.

We've already checked the whole circuit and tried to use the sallen key- filter but the result stays the same. I also tried to exchange capacity values without success. Simulating the circuit using the pspice model does not show the effect.

But when I look through the datasheet I recognized the sentence: "Fully specified for 300mV RMS input". Does that mean that every picture is only valid for that option ...? What if the input (as in our case) rises upt to 1.77V RMS? In other datasheets like the AD636 or the RMS- to DC- Application guide I can find a diagram showing the error in correlation of the input voltage. Does the AD8436 not have a correlation like this? When I refer to the effects I've seen in our circuit I would say yes but I can't find any information on this.

Could anyone explain these effects?

What is wrong in our circuit or what do I have to take in to account to get a reseult with higher accuray?

Any help or suggest is very welcomed

Regards Christian

  • Hi everyone,

    in the meantime we could solve the problem.

    There were two effects explaining that issue:

    1. the input signal which had a low amplitude HF- signal overlayed which comes out of the signal source. Up to

        know we're not sure where it comes from but anyhow it is there,

    2. The bandwidth of AD8436. Bacuase of the bandwidth of AD8436 ist was impoosible for the device to evaluate

        the HF- ripple as well. But the prroblem was that we wher comparing that value to a device which can do this

        so there was a big gap in measurement values.


    We added additional 100pF Capacitors in parallel to the input amplifier which is located before the input of AD8436 to filter out HF ripple of the signal. After modification values are comparable to another device not evaluating the HF- ripple as will.

  • 0
    •  Analog Employees 
    on Sep 13, 2012 4:09 PM


    I responded to this on 9/11, using the email address at the bottom, signed Christian. I received what I think was an 'out of office' message, but I don't read German very well. First, I'll copy my original response:

    Hi Christian,

    I’m using you email contact information because my EZ path seems to have a ton of delay and messaging is very frustrating. You can go ahead and attach this to the blog if you wish.

    I’m the apps engineer for the AD8436. I printed out your schematic, BTW thanks for including it. It really saves a lot of time and avoids headaches deciphering text and I wish everybody did the same. The AD8436 is trimmed and tested in production at 300mV, however your error numbers do sound excessive. I have some suggestions you could check out that might help. I too have see funny errors with rms-dc converters, however I have always found a solution so I’ able to pass a few on to fellow travelers

    First I need to get the input conditions straight in my mind. You noted an input voltage up to 250Vrms, I assume this is sine wave? This would be 707V pk-pk, or 353V (rounding) pk. You attenuate .00145x or about 69:1 with the 510k/15k differential attenuator, dropping the input to 5.12V pk per side, or 10.24V pk-pk. Have I got this right so far?

    Without taking onto account source loading (the 100k resistors are driven by slightly less than 7.5k each side), your diff-amp circuit U600A provides a gain of 5. So for a 250V rms signal you end up with 50V at the output of the diffamp, quite impossible with a 12V supply. Is there additional attenuation not included on the schematic?

    So before we go further please straighten me out, somewhere I must have made an error to come up with such a ridiculous result.

    Also, I haven’t seen a .info extension, is this something new?



    Now to respond to the latest::

    I'm happy to see that you've discovered a spurious input and I hope this turns out to be the real issue. It's very easy to fix, with the by-pass cap as you have done, so......... nice work!

    I have a couple of other suggestions as well. I noticed you're using an 47uF/805/X5R style capacitor as an averaging capacitor. I haven't investigated X5R multilayer caps capacitors but the small footprint, low voltage and large value (47uF) raise some flags.

    Size: Unless you're trying to convert frequencies < 2 Hz, you don't need such a large value. This size will just stretch out you settling time, for line frequency applicaitons (which I assume you have here) 4.7uF to 10uF is plenty to reduce the conversion to around 0.1%. You will find that external low-pass filtering is far more efficient, either passive with a 3.3uF at pin-OUT, or using the 2-pole SK option. I can suggest values for that approach as well. I suspect you might have chosen your cap values based on a prior ADI device, but in any event the internal ting achitecture is different in the AD8436 from prior parts.

    You didn't spevcify whether your error was positive or negative. If positive, the error would be consistent with the spurious input condition you found, if negative, it might have another source, nemely, board contamination. This problem has become much worse with the use of lead-free RoHS compliant electronics, as the fluxes used are hard to remove and the cleansing materials have to be maintained. In our lab, I found that hand=washing is not always effective. We now have a proper small scale board assembly area with up-to-date board washing facilities and it makes all the difference. I now specify the use of this wonderful capability and have had no more coard contamination issues.

    Getting back to the CAVG component topic, the sensitivity of this circuit has a lot to do with the capacitor drive circuit, which happens to be a current source. If there is any leakage path, either on the bopard or through tha capacitor the resistance (even several M-ohms) creates a shunt for the dc-current component. The current source provides a  fixed current, so the shunt path creates a small output error, usually negative.

    I've tried checking capacitors with electrometers and I think this works, but I also have a large selection of film caps, which I consider the gold-standard in capacitor quality factors. They also have very low dielectric absorption, which can cause a drift or repeatability issues. For a short while I used an 1000uF electrolytic as a low-pass filter to knock out ripple. This sort of works, and I thought at first the voltage source drive would mitigate capacitor errors but I learned about dielectric absorbtion the hard way! I now have 100uF worth of film caps for my lab measurements. Of course, this isn't needed for real applications.

    One final word, Kemet has some high temperature parts designed for automotive applocations - the rated temperature is 150C and they come in large sizes. I've tried a coiple of these and they look pretty good. They are also non-microphonic (aka non-piezoelectric) which the X5R are not! The dielectric has been changed from the Barium compound previously used.

    Sorry to be a couple of days late, and best of luck on your project. Is this for a commercial application, or is this a scholastic project?



  • Hi

    thanks for your detailed answers. Now I know why you have received an out of office reply because at the moment I'm using the log in of a colleague who is on holiday.

    I'm very sorry if this was a little confusing for you and delayed my response.

    Now again th the facts and to the questions and Ideas you have mentioned above:

    1. Input range: Yes you're right we have an input range of 0 -250 VRMS divided into to input stages  (0-50V RMS) and (51- 250VRMS). An input voltage divider breaks down  voltage to max. 5V peak- peak voltage wich is amplified by an opamp  after that. Depending on the input volatge we switch gain from 4.99  to 1.01 so that output range of the OP will not reach unallowd values. Mainly we're talking about an sine- input signal which was used for testing in that case.
    2. Behaviour of the output signal of AD8436: Output of AD8436 was always to low. For better understanding I have attached a pdf showoing the issue. In comparsion to a calibrated multimeter the noise on the signals was not evalutated for TRMS- measuring by AD8436 so the value was to low. But after adding the input filter at input OP now the output signal of AD8436 is somewhere in the middle of the value of the calibrated TRMS- meter and the former value of AD8436 without input filter. Now this value is comparable to another TRMS meter which does not evalute noise as well. The difference to the calibrated devices now is minimized but still there. If you have an idea to increase accuray more in direction to the values of the calibrated TRMS- meter please let us know.
    3. CAVG-/CLPF- Caps: We also tried out several Capacitor combinations. First of all we thought of using a 47uF cap to have the ability of measuring values <30 Hz as well but after checking the values we could not find any difference between unsing a 10uF and a 47uF Cap so we finally decided to use the recommended 10uF. We also tried to use different comibinations and values but there was no difference using a film or ceramic capacitor. Without input filtering values stayed to low. Because of the fact that we're designing a hand held device in as small housing huge sizes for component are a no go that means we have to stick to ceramic capacitors anyhow. When I remove CLPF of 3.3uF I could see that ther wa a high output ripple in range of 20mV -80mV as a function of the amplitude of the input signal ripple frequency was twice the input signal frequency. The peak value then was the expected TRMS- Value the calibrated TRMS- meter delivered. In my eyes that must be the noise of the input signal wich is not evaluated and averaged when adding CLPF. Or do you have another explanaition for that issue?
    4. Project type: We're desiging for a commercial product.

    I hope above information will help to understand our issue. Any idea or advice for improvment is very welcomed.

    Best Regards


  • 0
    •  Analog Employees 
    on Sep 20, 2012 3:34 PM

    Goof morning Christian,

    Our field applications engineer, Bernd Kraetzig, will contact you shortly. If acceptable to you, and at your convenience, I’d like to discuss this with you on the phone.

    Thanks very much,