This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

ADC3683EVM: Auto Zero shows no effect on spectrum data

Part Number: ADC3683EVM

Dear All,

 I am trying to sample with 65 MSPS. I tried various input frequency  1 MHz 6.5 MHz and 13 MHz. I tried to enable/disable the Auto-Zero by writing to address 0x11 a 0x1 (Enable) or a 0x0 (Disable). By reading the register back I see the values are being written. 

The spectrum in the 1/f region (100 Hz - 10 kHz) looks the same for both cases (AZ enabled or disabled). I am wondering whether the functionality is really enabled / disabled ? I guess I am doing something wrong...any idea? 

Best,

Uros 

  • Hi Uros,

    Interesting, we can certainly take a look at this. Can you post some images of what the 1/f noise looks like?

    I would also request for you to include the FFT data as CSV of the close in upper side band for both cases (up to 100kHz offset please)? To save the FFT data as CSV, in HSDC Pro single tone view, right click on the FFT display and select Save FFT Data and Measurements. 

    Regarding input frequency, I would suggest to use whichever input frequency closest to 5MHz available which is properly filtered at a level to achieve -3dBFS fundamental power.

    A few other details, please increase the capture depth to 524288 samples (choose Data Capture Options → Capture Option) and change the Analysis Window size to 524288 as well. This will increase the RBW to it's max at roughly 124Hz per FFT bin.

    Regards, Chase

  • Dear Chase,

    Thank you for your fast response.

    the comparison is shown below. 

    1. Data is taken with the HSDC. 524288 samples per channel.

    2. I took 100 shots and averaged them - simple video averaging

    3. The FFT is done offline. I do demodulation to I/Q and calculate phase. This is what is shown. So it is a close-in to carrier phase noise spectrum. 

    4. The fundamentals were all att . app -2 dBFS

    5. the dashed lines are auto zero disabled. Full traces are auto-zero enabled. For individual frequencies you see no difference. 

    6. signals is generated by the SMA100B. The 1/f of the generator is much lower from what shown in the FFTs. 

     I would expect a stronger influence of the feature at lower IF frequencies but I see no influence? Is it enabled at all?  

    btw. are the strong spikes visible expected? - I guess the switching capacitor or? 

    Best,

    Uros

  • Hi Uros,

    Thanks for explaining in such detail - it will help us out a lot. We will check internally to see why the described auto-zero function is not affecting the close-in performance. Kindly allow us a few days into next week to discuss.

    Regards, Chase

  • Hi Uros,

    I just wanted to let you know that we are still checking on this and we will provide more info soon. Thank you for being patient with us.

    Regards, Chase

  • Uros,

    Can you attach a zip of the averaged data?

    Thanks

  • Dear Chase,

    Try if this works....

    https://syncandshare.desy.de/index.php/s/ycmyM6Ct4MzGRQf

    please let me know when you downloaded the data.

  • Can you upload the data to this link instead? It will only be active for 7 days. https://tidrive.ext.ti.com/u/29wymeNcMrXJ3DKe/3c21ab33-4e9f-4696-b690-ac18deed1666?l 

  • Hi Uros,

    I can't see it. Can you try to upload again please?

    Thanks

  • should be there now...

  • , can you look to see if you can download the data? I am unable to see anything at the ti drive link above. 

  • This is what I see in the folder...the zip is there. It is however very large (0.7 GB). Maybe give it some time....

  • Hello,

    I downloaded the uploaded data and plotted it in our software tool. I am checking with our design team on the issue with the autozero register write, thank you again for your patience. 

    Regards, Amy

  • Hi Gruffalo,

    The comparison chart posted above shows a very low noise floor. Can you clarify the y-axis scale? 

    Thanks, Amy 

  • Hi Amy,

    For details I recommend this book.

    https://www.amazon.de/Frequency-Stability-Oscillators-Cambridge-Engineering/dp/052115328X

    It is a normalization to /Hz. What any SSA (Keysight) or FSWP(R/S) would show...The so called L-script. 

    Regardless of the normalization please focus on why the doted and the solid lines do not differ.  

    Here are my speculations:

    1. I am doing something wrong with the interface bit toggling and the function is not being toggled. Please check the process described above. Copied below:

    " I am trying to sample with 65 MSPS. I tried various input frequency  1 MHz 6.5 MHz and 13 MHz. I tried to enable/disable the Auto-Zero by writing to address 0x11 a 0x1 (Enable) or a 0x0 (Disable). By reading the register back I see the values are being written."

    do you see any problem? 

    2. the auto-zero is always ON. The toggling doesn't do anything because the interface has a bug.

    3. the auto-zero is always OFF.   The toggling doesn't do anything because the interface has a bug. 

    If you repeat the measurements in the lab do you see any effect on the spectrum data when enabling/disabling the auto-zero? Does the setup work on your side? 

    Best,

    Uros 

  • Hi Uros,

    Figure 8-9 on page 32 of the datasheet shows a small improvement in performance. How much improvement were you budgeting from enabling the auto-zero function? I will set this up in the lab to compare with your results. To align with what is seen in the datasheet, could you scale your plot to 100 kHz (the current scale of the plot goes out to 1 MHz). With the holidays coming up, I may not have an answer for you for another week or longer. Thank you for your patience.

    Regards, Amy

  • Amy,

    Slight smile..well the improvement is 10 dB and this is what I would expect since this is what TI is advertising. 

    the middle plot at 6.5 MHz (the bunch that shows -142 dBc/Hz at 100 Hz) is the closest to figure 8-13. If you would average more the 8-13 you would see 10 dB improvement at 100 Hz. The measurements I posted above shows app. 0 dB improvement - no need to rescale the plot. so there is a clear problem here...

    if I may, I think it is very clear how to solve the problem:

    1. make a setup in your lab and repeat the measurement above

    2. if you get it working let me know the solution

    3. if you dont...well, the feature doesnt work

    4. enjoy your holidays

    Best,

    Uros

  • Hi Uros,

    Thank you, since the scale of your plot was out to 1 MHz I wanted to make sure that your expected improvement was in sync with what is provided in Figure 8-9 in the datasheet. At 0.01 kHz the plot shows a ~10 dB improvement which reduces to (at best) ~5 dB improvement at 10 kHz offset. 

    As mentioned above, I will work on reproducing this in our lab and update you within the next week or more. 

    Regards, Amy

  • Hi Uros,

    The 1/f flicker noise plot in the datasheet is referring to DC noise, and the benefits of auto-zero function are visible around DC, not the upper side band of the analog input. See below plot with data taken in our lab compared to the data used for the datasheet plot. For reference. The data taken in our lab is using two SMA100B's (both with B711 ultra low phase noise option) for each of the sample clock and the analog input.

    The difference is not quite 10dB, but I am seeing around 7dB improvement. 

    Using the same set of data, I've also plotted the close-in upper side as below to show that we see no improvement here either. To me, personally, 1/f noise is always indicative of the slope of the Fin's upper side band skirting, however as I've just learned today, it can also refer to the slope at DC as well. 

    Regards, Chase

  • Dear Chase,

    Thank you for following up on this. 

    If I understood it correctly the improvement (app. 7dB) is only expected at DC. The plots in Fig. 8-9, 8-11, 8-13 in the datasheet are not close-in spectra (DC in these plots is NOT the carrier) but are showing the near-DC spectra of the sampled Fin (which would be a line at the given Fin if the plot would show it). 

    In fact the Fig. 8-8, 8-10, 8-12 show the full spectra but it is not so obvious whether the zoom-in (Fig. 8-9, 8-11, 8-13) is then on DC or close-to-carrier. 

    Best,

    Uros

  • Uros,

    Your understanding is correct. Sorry for the confusion on this topic.

    Regards, Chase