AD7689: PSRR much worse than in the datasheet

Basti,

Can you confirm that you are using an internal reference option (REF = 3'b000)?   In fact could you please confirm your entire configuration and which model you are using?

As a comment I believe you are trying to infer PSRR of the data converter from the peak-to-peak ripple on what I have inferred is the REF pin, which would not match the test method used in the factory exactly.   In our factory, the input would be fixed at a known DC voltage and the energy in the bin containing the ripple frequency would be reported in volts and then the ratio computed from that output value with respect to the input ripple as you have computed in your message above.  Now depending on the selected input voltage a certain amount of attenuation of that ripple could be expected which would impact the output result so I'll need to go back and look at this to confirm what the original conditions were.   

My expectation is that (and I will try to reproduce it) our observations would NOT have seen that significant of a ripple injected at the REF pin, even if internally generated, which could indicate a layout sensitivity in your physical design.   Could you please share a bit more information about the capacitors you are using and the layout.   If this is sensitive you can always forward the information through your ADI sales representative.

Sean

Hello Sean,

thanks for the quick reply. Sharing part of the circuitry is no problem. Had to remove some critical stuff, though.

Yes, we are using the internal reference of the AD7689 in the 2.5 V configuration.

I know that normally the PSRR is measured via the ADC counts values, but without proper software support this will be tricky. I have to run the ADC at least twice my noise frequency of 20 kHz. We can do that, but it will take a lot of time and effort. My simple approach was, that the ADC count value can never be better than the reference used. What I can do is get single measurements by clicking a read button. Won't give me any frequency information but I should still see the peak to peak value. Under normal clean VCC conditions I see 5 LSB of noise, which is in line with my expectations. Have to redo that when our VCC becomes noisy.

I am currently setting up the ADI evalboard to see if I can reproduce my measurements.

Best regards

Basti

Looks like the if one uses the count values to measure PSRR it is much closer to the datasheet:

I can run some 30 kHz sampling rate with our test software. And I can change the "noise frequency" in our system (normally at 20 Khz). I can reduce the noise frequency to 5 Khz to get at least some oversampling

Using the counts for the ADC itself and calculating the peak-peak values give me some 3 mV. I know the noise on the supply rails increases when reducing the frequency. So from 80 mV probably to 320 mV (no measurement, allowed attendance in the lab is limited right now)

With these values and an external 1.25V source as reference I get

PSRR = 20* log10 (3 mVpp / 320 mVpp) = 40.5 dBe. This is much closer to the datasheet and won't become any better without spending a lot of time onthe setup.

I think this topic can be closed.

Basti,

We are still very much on restricted access here as well but if I can get into the lab to at least try to validate your observation of ringing at the REF pin on the evaluation setup I have for the LFCSP version of the device I will.    I would like to see if the test configuration used effectively missed a sensitivity of the reference buffer or if you simply have some sort of resonance in your circuit implementation that is a function of the components and layout.   If it is the former I would think it prudent to specify a PSR for the reference buffer in the datasheet.  

Can you please clarify which version of the product you are using?

Thanks 

Sean

Hello Sean,

we are using the WLCSP Version (AD7689BCBZ). I was using the barrel and tip method with my scope probe to minimize any ground loops.

Best regards

Basti