Post Go back to editing


I am using AD7685 to read a CMOS linear camera sensors. The sensor has 512 pixels. It reads 512 consecutively at 125Khz, wait some time, then read it again. 

The first few AD readings, for some reasons, are always off a little bit. The first reading has the most error in it.  I wonder if it is issue with the AD7685, where there is sampling issues with the first few pixels due to the sudden signal level change. So checked the track time, it has more than 1.8us that the spec says. 

I looked the sensor output, I don't see the discrepancies for the first few pixels, they are all pretty smooth. 

I wonder if there is sleep low power mode before you start the first conversion.  Takes long time for the chip to get out of the sleep mode. 


Added a tag for internal tracking purposes.
[edited by: tschmitt at 7:11 PM (GMT -4) on 23 Sep 2020]
  • Hi Andrew,

    Thanks for posting your query.

    My first guess is that you may be witnessing a reference voltage drive issue, meaning the voltage on the AD7685 REF pin is deviating across the first few samples (i.e. readings) before eventually settling out to its steady state value. The reason I think this is because the AD7685 REF input is a dynamic load, where it draws very little current while the AD7685 is idle (i.e. not converting) and then draws short bursts of peak current while it is converting. This behavior is common for all precision SAR ADCs, since the bit trials in the converter core require pulling charge from the REF input to generate an output code.

    Basically if the circuitry providing VREF to the REF input is not designed properly, i.e. with low output impedance/load regulation, this can cause the output voltage of the reference circuitry to ring for a few samples before it fixes itself. The response is similar to looking at the step response of an underdamped system.

    Can you share what components you are using in the circuit driving the AD7685 REF input? Did you include a REF decoupling capacitor placed very closely to the AD7685 REF input as recommended in the Voltage Reference Input section of the AD7685 data sheet?



  • See the first few pixels. I get a large swing of more than 100 counts. If the ADR421 Vref is too slow to respond and cap too far, I suspect few counts, not in the magnitude of 100. I don't have the cap right next to REF pin on the ADC. I put several caps under the ADR421. ADR421 and AD7685 are next to each other, trace width is 0.19mil, half inch trace length roughly. 

  • See the first few pixels. I get a large swing of more than 100 counts. If the ADR421 Vref is too slow to respond and cap too far, I suspect few counts, not in the magnitude of 100. I don't have the cap right next to REF pin on the ADC. I put several caps under the ADR421. ADR421 and AD7685 are next to each other, trace width is 0.19mil, half inch trace length roughly. 

  • Hi Andrew,

    A few things here:

    First off, I see the ADR421 is used to provide the REF pin voltage for the AD7685, but what is the ADR435 you mentioned used for? Is that what you're using to drive the AD7685 IN+ input? And how are you conditioning the ADR435 output to make it 2.5V (the ADR435 is a 5V output device)?

    Second, if the ADR435 is not providing the AD7685 IN+ voltage, then what is exactly? You mentioned you're measuring the output of multiple pixels, are you doing that for this measurement here? How do you get the output of multiple pixels into the input of a single ADC? What type of sensor is it?

    I ask all these questions because the type of error you're seeing could be from the reference or the analog inputs also not settling properly. If you are willing to describe the analog signal chain components/configurations between the pixel/sensor and the AD7685 IN+ input, that would be helpful in assessing if it's an analog input settling issue. But if you can't share that then I would direct your attention to the following article for more information about analog input settling requirements for SAR ADCs:

    Lastly, on the reference circuit design that we were originally discussing - when you say the REF decoupling cap is "under" the ADR421, do you mean they are on the opposite side of the board and are routed to the AD7685 REF input through multiple board layers through vias? This can be problematic due to the extra impedance that these vias introduce. Even tiny amounts of trace resistance and inductance can result in noticeable errors in the VREF voltage on the AD7685 REF pin.

    I would direct your attention to article below as well. The REF decoupling cap serves the purpose of providing quick bursts of charge to the AD7685 ADC core during the ADC bit trials. As you can see in the article, the REF input of a SAR ADC can have large current spikes, and if the decoupling cap isn't placed properly then that charge can't be sourced quickly enough to maintain a proper VREF, resulting in conversion errors. I suspect that the REF decoupling cap placement is an issue contributing to the errors you're seeing.




  • Sorry, for the project it is AD421. I mixed it up with another project that uses AD435. So for this discussion, it is AD421. 

    It is linear sensor so pixels are scanned one after another at 125KHZ

    Settling into the AD7685 input is not issue because I checked with scope. There is an OPAMP before the AD7685 input. 

    I will test the cap under the REF pin. My take is that it is an issue, but should not cause that big a voltage swing. 

  • I put 2 10uf 0805 cap on top of the REF pin, I don't see any difference. 

    So it must be some other reason. I will put a new AD7685 chip on board. 


  • Hi Andrew,

    Thanks for giving that a try.

    I think the next step would be to double check the analog front-end design. The signature you're seeing in the pixel measurement vs. time plot you provided lines up with what we'd expect if the analog input voltage at the AD7685 isn't settling properly.

    First off, what voltages do you expect from the sensor output across each of the pixels? Are they all more or less the same voltage?

    You mentioned there's an op amp between the sensor and the AD7685 input. Can you clarify which op amp model this is? Or at least what it's GBW specification is, and whether it's in a unity gain buffer configuration or some other configuration?

    Also, did you implement an RC filter between the op amp and the AD7685 input, as mentioned in the article I referenced last week?



  • Yes, pixel output should be more of less the same. continuous. 

     I check the opamp output on the scope, they all look okay, no settling issues.  I am very sure the problem is the from the AD7685.  It is running very slow as well, only 125Khz. 

    Yes, I have the RC in front the ADC. The opamp is OP4350. 

  • Hi Andrew,

    Looking at the op amp + RC circuit provided, I am inclined to agree with you that this circuit should achieve proper settling. But for a sanity check, would you be able to do one or both of the following:

    1. Slow down the sample rate even further, even down to 50kSPS or something in that ballpark. I suggest this because if the accuracy does improve as the sample rate gets slower, then it would point to some sort of settling issue, and I assume this would be a relatively quick test to try.

    2. Instead of driving the op amp circuit with the pixel sensor, can you try driving it with some fixed DC source with relatively low output impedance to isolate any performance impacts to just the ADC + op amp circuitry? 

    Apologies for my ignorance here but if you could provide a data sheet/model number for the sensor itself, that would be appreciated as well. The reason I ask about this is I want to get an idea if the sensor output is interacting with the op amp circuit in a way that could be causing a settling issue, hence my request in (2) to drive the circuit with some other test voltage.

    Thanks again for your patience,