AD9689 linearity - DNL Measurement


we made DNL measurements with our in-house developed 2.6GSPS 14-bit AD9689 based adc board.
Measurement method is to calculate DNL from a cumulative histogram.
Samples will be taken with a sine wave near full scale input signal.
Required sample size is around 1 MS for confidence Level 99%, 14-bit and expected typical DNL error +-0.4 LSB.
DNL result is almost like to find in adc datasheet.
If we take more samples as required the calculated DNL result gets better (smaller).
But there seems to be no end of this trend.

To check the calculation i tested with some histograms of another adc, in example 16-Bit 125MSPS ADC AD9268.
It shows a behaviour like expected.
With more and more samples the DNL gets less and less better and tends toward a typical value.

Complete different in histograms of both adc's is the recognizabiltity of significant high or low DNL values of some codes.
For it will be made some DNL measurements with same parameters and sample size.
I don't know, if it is o.k. to call it signature or fingerprint if a view at DNL errors plots of some measurements shows large and small values always at same position.
DNL plots of 125MSPS ADC showing a signatur but 2.5GSPS ADC doesn't.
That's why the DNL result is getting better and better with increasing of samples size.

How can this be explained?
Is it due to dither enabled, how can it be disabled?
How can be made a correct DNL measurement for AD9689?

Please help
Best regards


Top Replies