Issue with nanoVNA-Saver in at least the last several versions.
|
diyer,
You make a good point about the harmonic content of the square wave. They do extend out to much higher frequencies than 50 kHz. I am not sure how these harmonics would impact the 50 kHz result in a case where both inputs are terminated in 50 ohm loads. And specifically, how they would contribute to an answer at 50 kHz of 0.0 + j0.0. I do see zero S21 value at frequencies above 50 kHz. I see intermittent zero values for S21 up to about 10 MHz but not usually above that. An example is shown the attachment "S21Isolation-singlescan.png." I am not sure why. I have a nanoVNA and a nanoVNA-F and the -F model shows this behavior more frequently. The nanoVNA-F model seems to have smaller S21 values in the isolation measurement below 50 MHz than the nanoVNA that I have as can be seen in the S21Isolation files attached. These were created by averaging 20 scans. When averaging 20 scans, they don't show 0.0 + j0.0 or 0.0 dB for S21 at any frequency. I attribute most of this issue with zero S21 values to the likelihood that the S21 signal with both inputs terminated in 50-ohm loads at low frequencies is not really noise from the input but some sort of very low-level, repeatable internal leakage signal. I am wondering if the S21 isolation calibration process causes the S21 calibrated answer to be identical to zero for repeatable internal leakage which is 100 dB or more below the reference signal. None of these actual indications in the device are a concern for me in the actual measurement. I would just like nanoVNA-Saver to deal with the 0.0 + j0.0 value received from the VNA in a little more appropriate way than to indicate that S21 is 0.0 dB in this case. Apparently in the nanoVNA device itself the 0.0 + j0.0 answer is dealt with properly because I never see an S21 indication of 0.0 dB in the logmag plot when terminating both inputs with 50-ohm loads. -- Bryan, WA5VAH |
I would guess that there are some significant strong harmonicsI have no reason to suppose that harmonics at low frequencies are relatively stronger than at higher and should be problematic mostly when measuring nonlinear networks. My neighbor and I were tinkering with a NVA and when we measured a 10mh chokeThat resistor's value was presumably fairly comparable to choke reactance... I use a worse clone nanoVNA e.g. to measure capacitive loading of phono cables and preamp inputs, which involves calibrating and sweeping e.g. from 10kHz - 1mHz or higher, because capacitive reactances at lower frequencies are too high for accurate measurements, keeping in mind that nanoVNA bridge does poorly above 500 Ohms. That also applies for very low inductive reactances. |
|
Alan,
I have done some more investigation of the issue I am seeing to further isolate the cause. Here are the results. For the nanoVNA (2.8" display) I thought sure I had seen the problem with this device; however, at this time I cannot reproduce the symptom of getting 0.0 + j0.0 for an S21 measurement when both input ports (CH0 and CH1) are terminated. I did terminate both ports when performing the isolation calibration. With both ports terminated I see single-scan (50 kHz to 50.05 MHz) S21 values from a peak of about -78 dB to a minimum of about -100 dB. As a result I don't see any issues with indications in nanoVNA-Saver with the S21 values. This is true whether I terminate port 0 or port 1 or both. I see very little change when terminating the ports or leaving them open. In a post above, I attached a file (S21Isolation-nanoVNA.png) showing the result of averaging 20 scans which lowers the noise floor substantially, as it should, when compared to a single scan. The maximum value when averaging 20 scans in nanoVNA-Saver from 50 kHz to 50 MHz in that plot is about -89 dB which is quite impressive. And, the result seems to have a random variation with frequency which is consistent with it being a noise floor. The result above caused me to go back to the nanoVNA-F (4.3" display) to try to verify the exact conditions which cause the issue with it that I showed in an earlier post (the file S21Isolation-Singlescan.png.) Again, I did perform the isolation calibration with both PORT1 and PORT2 terminated in 50 ohms. I just went back and looked at S21 values with no load on either port of the nanoVNA-F. I have attached a plot from nanoVNA-Saver of an example (S21Isolation-nanoVNA-F_noterm.png.) Note that this plot has a maximum value in the single scan of about -82 dB and a minimum value of about -106 dB. In this plot the data does look like random noise which is approximately uniform over the 50 kHz to 50.05 MHz range being tested. In this example, there are no 0.0 + j 0.0 results for S21. And, I very rarely see what appears to be a 0.0 + j0.0 result on the S21 display of the nanoVNA-F when there are no terminations on the input. However, when they are there the device plots the point at the most negative value on the display. It is so rare that I have been unable to capture a 0.0 + j0.0 with no terminations on the device. In order to observe the data in the isolation case on the display I had to set the scale/division to -20 dB/division. I then added a single termination to PORT2. An example plot is attached in the file S21Isolation-nanoVNA-F_termOnPort2.png. In this file I just happened to catch an S21 value of 0.0 + j0.0 at 4.05 MHz. So I also saved the s2p file which is attached as S21Isolation-nanoVNA-F_termOnPORT2.s2p. Looking at this file you see that the S21 value at 4.05 MHz is 0.0 + j0.0. Further we see that nanoVNA-Saver shows S21 gain of 0.0 dB which is clearly not what we want for a 0.0 + j0.0 response from the VNA. With a single termination on PORT2 of the nanoVNA-F, it is very hard to capture such an measurement. Watching the nanoVNA-F display for several minutes, I did not see any more such indications even though a scan is repeated every 2 seconds or so. I was simply lucky to get the result I have attached. Also note that the maximum value of S21 in this plot is still about -83 dB and the minimum is about -120.0 dB at 18.55 MHz. Looking at the s2p file saved from nanoVNA-Saver I observe the reading was 1e-6 -j0.0. I can't be sure of what came out of the nanoVNA-F; however, I think it put out 0.000001 - j0.000000 Since the output from the nanoVNA-F is in ASCII decimal notation with 6 digits to the right of the decimal point, the value reported at 18.55 MHz is literally the smallest value that can be represented other than 0.0 + j0.0. So what may be happening here is a limit on the precision of the representation sent from the VNA to the PC. I then added a termination to PORT1 of the nanoVNA-F with no other changes. Remember, this is the exact same configuration used to calibrate the device for the isolation measurement. A plot of a single scan from the device is attached as S21Isolation-nanoVNA-F_termOnPORT1_PORT2.png. Here we see an S21 value of 0.0 dB at many frequencies at and below 10.05 MHz. I saved the s2p file (S21Isolation-nanoVNA-F_termOnPORT1_PORT2.s2p) for this scan as well. Looking at this file we see 0.0 + j0.0 values at all frequencies below 10.55 MHz except 2.05 MHz, 6.05 MHz, and 9.05 MHz. All frequencies where 0.0 + j0.0 is reported in the s2p file are plotted in the nanoVNA-Saver S21 Gain as 0.0 dB. Those frequencies listed here that are not 0.0 + j0.0 are reporting pretty small amplitudes that are not too far above 1e-6, the minimum value that can be reported in the ASCII stream of data sent to the PC from the nanoVNA-F. On my device with the calibration I created, this type of result is quite repeatable. Every scan with both inputs terminated in 50-ohms will have some values which report as 0.0 + j0.0. What this means, in practical terms, is that the isolation between PORT1 and PORT2 is really good or that the leakage between those ports is very repeatable below about 10 MHz and calibration is removing those values. The 20 scan average shown in the post above (S21Isolation-nanoVNA-F.png) shows a decided lowering of the S21 response with both ports terminated below about 25 MHz. I would not expect this to be a reduction in noise at lower frequencies. It could mean that the leakage signal is more repeatable at those frequencies and isolation calibration is removing repeatable errors. In either case, I really don't think this is a serious problem for practical measurements. This configuration of loads has many more results that are reported as 0.0 + j0.0 from the nanoVNA-F; however, it is also the configuration used in isolation calibration. Interestingly, the values sent to the PC have to be on the order of 0.000010 to 0.000030 for signals down by 90-100 dB. The quantization for these small values cannot have a great deal of precision given that 6 places to the right of the decimal all that is being sent to the PC in ASCII decimal notation. Quantization error can be significant on amplitude and/or phase. I would call this a limitation not a problem to be addressed. The fact that the nanoVNA-F seems to have lower isolation values than the nanoVNA is perhaps a good thing but is, in some ways, being limited by the precision of the data being sent to the PC. Again, I don't see it as an issue but it is a limitation. The creators of the software had to make some choices about the representations they send to the PC once they decided to send it in ASCII decimal format. Choosing 1e-6 as the minimum value to represent is reasonable; however, it does result in potential quantization errors when we are talking about signals that are smaller than the reference signal (the signal being sent out PORT1) by something like 90 to 100 dB. I don't consider this an issue but simply a limitation in the dynamic range that can reliably be measured by the nanoVNA-F device. The real issue is that nanoVNA-Saver is not working with this limitation well when the device sends back 0 + j0.0. It should plot and treat this indication from the nanoVNA differently for ALL S21 measurements and calibration. A relatively simple software fix should be available for Rune to implement. I have not but will try calibrating the isolation measurement using only a load on only PORT2 to see if that changes anything. I'll post the result here as soon as I have time to create the result and evaluate the information. -- Bryan, WA5VAH |
|
Here is a Collins voice multiplex mechanical filter centered at ~ 88 kHz. The filter is ~ 3 kHz wide, an USB filter. Its been kicking around since the mid 50-60's. Excuse the ripples for its age!. Anyway, it has the shape factor and the rejection that one would like to see from such a class of filter. Again, 1010 points now from 50 kHz to 200 kHz. About 50 Hz or so per step. I do not see any strange events occurring at this point. The cal was a full so called 2 port cal includes "isolation".
|
Did I miss anything?Yes, you posted this in the wrong group. Instead, try one of these: /g/nanovna-users/wiki/home#nanoVNA-F |
Oristo
