AD7793 noise and accuracy


I have a couple of questions abou the AD7793. On page 3 of CN-0206, the resolution of the system  is calculated to be 0.014C. The p-p noise is stated as 6.6*0.088uV=0.581uV. But isn't the 0.088uV an output noise spec (per table 8 of the datasheet)? With a gain of 128, the sensitivity of the thermocouple is now 128*40uV/C=5.12mV/C.
Shouldn't the output noise of 0.581uV be compared to a sensitivity of 5.12mV/C?   I'm saying this because Table 8 doesn't say this noise is "referred to the input" of the device.

Also, I'm trying to figure out what the accuracy of a thermocouple system would be using the AD7793. For now, I'm trying to calculate the error caused by only the 0.01% reference. Please check my math and methodology: For 40uV/C and a gain of 16, the ADC input signal is now 640uV/C. For a K-type thermocouple, the signal at 1370C is 54.819mV. The gain of 16 brings this to 0.877104V (I've chosen a gain of 16 because if I increase the gain to 32, then the ADC input will be above 1.17V, and I won't be able to use the internal reference). With a max signal of 0.877104V and a 0.01% reference, the error due to the reference is 87.7uV max. 87.7uV/640uV/C = 0.137C.  I'm thinking this would be the error just due to the internal reference for a K-type thermocouple at 1370C. Have I done this correctly? I'd appreciate any feedback you can send my way.

Thank you very much,

John Araujo

  • 0
    •  Analog Employees 
    on Feb 27, 2012 4:55 PM

    Hello John,

    the rms noise numbers listed in the datasheet are input referred. This is not stated clearly in the datasheet.

    I agree with your error calculations regarding the internal reference.



  • Thank you very much, Mary.

    May I please run another question by you?  The AD7785/AD779x have two differential inputs (not counting the reference inputs). This made me think that it wouldn't be hard to develop a two-channel thermocouple system, as long as we had another way to do the cold-junction compensation. However, there is a AD779x Frequently Asked Questions Doc that warns that using the internal bias circuit creates differential noise on the input when filtering components are present. The magnitude of this noise isn't dicussed, but since it's brought up in the document, I'm assuming it's a problem and should be avoided. Would you agree?

    One suggestion to get around this is "if the ADC has an unused channel, the bias voltage can be made available on this unused channel, then applied to both AIN+ and AIN-". That's fine, but now I won't be able to implement a two-channel thermouple.

    So I had this idea instead. What if I used one of the Iout channels to turn on a low noise 1.2V reference. Then, this 1.2V reference could be used as the bias voltage for each channel. Do you think using the same reference for both channels might cause a problem? I don't think I'd have to pay for an accurate reference, since it's just being used as a bias, and it's the differential thermocouple voltage I'm looking for. I'm thinking it's much more important to use a low noise, low tempco reference. Is this a good idead? Can you recommend a reference?

    Thank you very much!


  • I've asked another question but I've thus far  received no response. I have more questions concerning the AD7793/AD7785.  I'd have to conduct a tolerance analysis at 25C for a thermocouple system. Here are the errors I think are related to the AD7785,

    • reference error
    • INL
    • offset error
    • full scale error

    Shouldn't there also be another term for the gain error? For example, if you choose a gain of 16, what is the tolerance on that. I don't see that mentioned in the datasheet at all.

    Are there any other terms I forgot?

    Thanks you,


  • 0
    •  Analog Employees 
    on Mar 2, 2012 5:42 PM

    Hello John,

    a R-C filter is required on the analog inputs to the ADC for anti-aliasing purposes. A typical filter would be a 1K resistor in series with each analog input, a 0.1 uF cap from AIN+ to AIN- and a 0.01uF cap from each input to GND. With this network, it is possible to see a signal at the ADC input due to a differential error introduced by the bias voltage.

    You could use a part such as the AD7794, which  is a 6-channel version of the AD7793. You would have 2 channels for the thermocouples and a third channel could be used to bring out the bias voltage for the thermocouples. You would also have an extra channel which could be used for cold junction compensation.

    If you prefer to stay with the AD7793, then I suggest an external bias voltage as you suggested. I do not understand your suggestion regarding IOUT. IOUT is an excitation current. You could set up the bias voltage by using 2 resistors to generate a voltage of AVDD/2 which could then be used a the bias voltage. Another option is to use a reference such as the ADR380. This is a 2.048V reference. The 1K price is $0.78.

    Your error analysis is correct. The FS error includes the error due to the offset (gain error is the error with the offset removed). At a gain of 16, the AD7793 has a full scale error of +/-250 ppm max after performing an internal full scale calibration. This spec is over the complete temp range from -40 to +105 degC.