What is real noise performance of nanodac/nanodac+?

There are many mismatches in noise characteristics found in datasheets of nanodac and nanodac+ families.

Let's examine AD5689R one first.

DAC output NSD (DAC code = midscale, 10 kHz; gain = 2) is 300 nV/√Hz and LF noise 6 µV p-p. With gain 2 and midscale code noise at DAC output should not be less than reference noise (240 nV/√Hz at 10 kHz and LF noise 12 µV p-p). So for density data looks adequate but for LF noise is smaller than expected. Noise density in figure 11 matches numeric value of 240 nV/√Hz at 10 kHz. LF noise in figure 12 is 60 µV p-p and absolutely does not correspond to 12 µV p-p data in reference characteristics section. Figures 36 and 37 show DAC output LF noise for external and internal reference usage cases. LF noise is estimated as 9 and 12 µV p-p respectively. Figure 38 contain NSD for various code setting. Zero scale data should be determined by buffer noise only. So we may estimate buffer NSD 50 nV/√Hz at 1 kHz and 16 nV/√Hz at 10 kHz - far below reference NSD. Full scale NSD should be determined by reference and buffer noise and still buffer noise is far less reference noise than by reference noise only. NSD for full scale code in figure 38 is twice reference NSD in figure 11 so we assume that gain=2 setting is used. NSD at midscale code should be determined by all 3 sources of noise i.e. reference, resistor string thermal noise and buffer noise. NSD at midscale code at DAC output in figure 38 coincide with internal reference NSD trace in figure 11 in frequency range from 10 Hz to 100 kHz. So for one more time we see that buffer noise is far less than reference noise and resistor string noise is far less too. Assuming that resistor string is same as in AD5689 its value is 32 kOhm at gain=2 setting so thermal noise produced by resistor string at midscale should be 24 nV/√Hz (8kOhms equivalent resistance at buffer input gained by 2). So that noise is far less than reference noise too.

Then look AD5689 datasheet data i.e. same DAC but without internal reference. As we examined before buffer noise is estimated to be approximately 16 nV/√Hz at 10 kHz, resistor string noise at midscale and gain=2 should give additional 24 nV/√Hz resulting in total DAC output noise 29 nV/√Hz. However NSD in AC characteristics section is same as for AD5689R i.e. 300 nV/√Hz. Also LF noise is same.

There is no NSD figure in typical characteristics section of datasheet and LF noise plot in figure 32 is just copy of  figure 36 from AD5689R documentation.

So my primary question is what is noise performance of AD5689 chip when low noise external reference is used? Should it be 30 or 300 nV/√Hz - it is of great difference.

Second question is what data is reliable and what is mistaken in AD5689R documentation cause there are mismatches between data from different sections of datasheet.

Same situation (equal noise performance in datasheets for DACs with and without internal reference) is also found in other nanodac(+) family members datasheets.

  • 0
    •  Analog Employees 
    on Aug 9, 2016 11:56 AM

    Hi There,

    The AD5689 output spectral noise should be 100nV/rtHz, not 300nV/rtHz. This is an edit that we are currently working on as part of an update to the data sheet for that product.

    In the AD5689R, figure 12, the vertical scale on the plot should be 2uV/div, not the 10uV/div that rev A currently shows, an update will be forthcoming here showing the correct value in the near future, apologies for the confusion this has caused.  Does that clarify the discrepancies between the different sections for you?  If there are other issues let me know and I'll address them if I can.



  • Thank You Michael.

    Does output noise of 100nV/rtHz mean that most part of noise is produced by buffer itself and not by resistor string?

    If so and according to "real" DAC architecture explained in next thread then NSD would not change significantly when gain is changed. And this will lead that SNR with gain=2 setting should be better than with gain=1 cause signal is "gained" twice but noise introduced by buffer not. It is obvious that quantization noise is still little bit higher but there are applications where "analog" noise should be taken into account too.

    Best regards.