|
Re: Am I Fixing my BNC Calibration using Calibration Standards Adjustments in nanoVNA-Saver 0.1.2
On Tue, Oct 15, 2019 at 03:42 PM, Rune Broberg wrote:
Rune, Yes, your software can do a lot of things that I seem to need. Thanks, again, for continuing to work on this effort. You and your efforts are greatly appreciated. I will cover the changes in the open, short, and load separately below. To get the plots shown here, I averaged 3 (or more) copies of 5 scans from 50 kHz to 900 MHz. The total number of frequency samples was therefore 505 samples or about every 1.78 MHz. Averaging is very useful and important in nanoVNASaver, especially when looking at these kinds of measurements, especially above 300 MHz. First I'll talk about the open. On the Smith chart plot little change is observable after the tuning is applied because the changes being made represent small changes in the phase of S11. In fact, to see the changes one has to look at the phase of S11 on a scale of about +/- 2 degrees. The S11 phase of the open is shown in the file "S11-Open-Phase-NoTuning.png". The change is such a small change in the phase, it is almost impossible to see anything on the Smith chart. The data is all clustered near the right side of the Smith Chart as it should be after an effective calibration. However, it appears to me there is real data in this phase plot, even after calibration but before applying any tuning to the parameters in the "Calibration standards" form. When I changed C0 in the "Calibration standards" form of nanoVNA-Saver to 1200 and the offset delay to -55 ps. This change produced the phase plot in the file "S11-Open-Phase-with-Tuning.png". Notice two things about the plot after tuning. 1) There is no "quadratic looking" drop in the phase as we move from low frequencies to higher frequencies as there was before tuning. 2) Although I did not do this, a linear fit to the data in the phase plot after tuning will be a horizontal line near zero degrees. This is the definition of an ideal open. Thus, while you cannot readily see such a small change on a Smith chart, there is some significance to the changes because we removed frequency dependent artifacts of the open I used in the calibration process. Please also notice from the amplitude charts in the first post in this series show some improvement in the trend of S11 in dB when measuring the open RG-213 cable. This is not very compelling, but read on. Now let's talk about the short. The short I used produced a nice dot on the left side of the Smith Chart after the normal calibration; however, there was significant ripple remaining in the S11 amplitude as shown in the plot below "RG-213-WithOpenOnlyTuning.png" I suspected, but did not know that could be the fault of the inductance in the short that was used in the calibration. By modifying the L0 parameter in the "Calibrations standards" form we can remove the inductance that may be present in the short I used. The resulting S11 amplitude plot is attached to the first post above and called "S11-RG213-WithCalTuning.png". Also, please refer to my post to Kurt in this thread that shows the vastly improved Smith chart for the RG-213 cable. The circles are now circles and they are much more closely centered on 50 ohms. So, now to your question: What does the actual short used in the original calibration look like on the Smith chart once the L0 correction for the Short is inserted? The result is shown in the figure "Smith-OriginalShort-WithCorrections.png". This result should be exactly like a 1.2 nH inductor which is what I tried to remove using the L0 term. In fact, nanoVNA-Saver confirms this inductance after this correction is applied by indicating 1.2 nH is the "parallel X" when measuring the actual short. When the VNA is properly calibrated for a short, it shows the actual inductance of the physical device I used, equivalent to 1.2 nH. In the case of my BNC short, I don't think I have the ability to reliably measure a resistance that small. The ohmmeter I have won't go lower than 0.05 ohms even when shorting the test leads together. When measuring the short I used, it still measures 0.05 ohms. I doubt that many folks doing hobbyist work can measure the resistance of their short either. Such small resistance values have little impact in a 50 ohm system. Therefore, I don't think there is benefit to including a resistor as a part of the cal correction for a short. A discussion of the load follows. I won't go into the details about the tuning I did for the load. A similar procedure to what was used for the open was used for the load. Here, again, averaging a lot of sweeps is important in order to measure the trends in the phase of a signal with a return loss of some 40 to 50 dB. In order to flatten the S11 phase of the termination I used for calibration I did need to add a very small amount of inductance (50 e-12) in the calibration form. This is the equivalent of trying to remove 0.05 nH, a small amount indeed. At 900 MHz that is a reactance of about 0.28 ohms. Certainly not huge in a 50 ohm system. Of course, this small change is unobservable on a Smith chart using the nanoVNA but it is clearly visible on the S11 phase plot. Long papers have been written on all of the possible issues with trying to model various loads and what to do to compensate for various types of loads at various frequencies. I think there are too many topologies to deal with in this type of software. The capacitance can be in series with the 50 ohm resistor or it can be in parallel or there might be capacitance in series and in parallel.... The same can be said of the inductance. In my opinion, you could leave what you have in place and let folks work with it so long as you allow negative numbers everywhere. Then folks can think about things in whatever way they must as long as you are clear about what you are doing in the software with the non-ideal values entered in the various boxes of the "Calibration standards" box. By-the-way - When I reset some of the values in the Calibration standards box to zero after putting in and applying values, nanoVNA-Saver 0.1.2 immediately crashes. You might want to look into this. I am sure it is reproducible. Right now, I don't recall which ones do it. I hope this helps answer your questions. Again, thank you Rune for making the nanoVNA-Saver software available. I find it particularly valuable and useful. -- Bryan, WA5VAH |
|
Re: Nanovna-F is here
Where did I purchase? eBay and it took about 4 weeks to arrive.
Why three, now 4, vnas? I wanted one of each of the identified versions to test whether one was better. Whether shields made a difference in performance. Those tests are done and now I am paring down to two...... one mounted in my lab and one for my toolbox for the field. WA8TOD |
|
Fairly good raw uncorrected source and load match - based on one sample
Dr. David Kirkby from Kirkby Microwave Ltd
One of the parameters specified for a VNA is the
* raw uncorrected source match and * raw uncorrected load match. The better the raw performance, the better the corrected performance. I measured these best I could using my 8753ES VNA. Had these been poor, I would have tried to improve them. But based on my sample, it is not worthwhile trying to improve them. My procedure was 1) Set 8753 to its minimum output power which was -22 dBm, as its default output (I think +5 dBm), would have overloaded the NanoVNA. 2) Use a cable that came with the NanoVNA it is it quite flexible. 3) Added a 2 dB attenuator on the cable, for no other reason than to protect my calibration standards. 4) Set the 8753ES VNA to use the 85033D or 85033E calibration kit. This has the same parameters as my 85052B 5) Calibrated up to 3 GHz, which is the limit of my 8753ES - I do have a 20 GHz VNA, but below about 860 MHz, the 8753ES has better performance than the 20 GHz VNA. 5) Measured return loss of port 2 with the NanoVNA powered on and off. There was some difference, but only a few dB. As you can see from the photograph, the worst return loss on the load is 17.7 dB. 6) I measuring the source port with the NanoVNA powered on. I could find no way to stop the source completely messing up the measurements. Setting the frequency down below 100 kHz still screwed up measurements at 900 MHz. There were huge spikes. Note that the 8753ES VNA source power would have been around 10 dB less than the NanoVNA output power. Ignoring the very significant spikes, I do believe that the return loss of the source wasn¡¯t changing much whether the NanoVNA was powered up or not. Anyway, the worst case return loss at the source, whilst powered off, was 28 dB. In comparison, the guaranteed performance of my 8720D VNA over the range of 50 MHz to 2 GHz is *HP 8720D (guaranteed)* *Source match: 12 dB* *Load match: 22 dB* *NanoVNA (sample of 1)* *Source match: 28.0 dB* *Load match: 17.7 dB* So my one sample, the NanoVNA had a much better source match than the 8720D, but a little worse load match. Since I know that the NanoVNA was getting worse as the frequency was swept up to 3 GHz, if I did add the firmware that went to 1500 MHz, I might consider adding a small attenuator or resistor on the load to improve the load match, but to be honest I am only likely to do reflection measurements up to 1296 MHz on the NanoVNA, so the match at port 2 would be irrelevant. I didn¡¯t think until after I had performed a 1-port calibration on the 8753ES, that it might have been worth measuring the isolation between the two ports on the NanoVNA. The 8753ES has a system dynamic range of 110 dB, which would have dropped to 90 dB at a much reduced output power. Probably good enough, but if not reducing the IF bandwidth, increasing the output power and averaging would have all helped. Anyway, I thought it interesting, and based on my sample at least, any attempt to improve the match at the ports would be best done on the receiver port rather than the source port, but it¡¯s pretty good. -- Dr. David Kirkby, Kirkby Microwave Ltd, drkirkby@... Telephone 01621-680100./ +44 1621 680100 Registered in England & Wales, company number 08914892. Registered office: Stokes Hall Lodge, Burnham Rd, Althorne, Chelmsford, Essex, CM3 6DT, United Kingdom |
|
Re: errors of "error" models
#49 :
first details of our final report 1 : nominal values comparison : our [NanoVNA] ~ our [VNA] - #30 : our final report 1 - 6 October 2019 : /g/nanovna-users/message/4179 - Hello, Allow us, please, to present the first details of our final report 1 regarding the Nominal Values of Rho for our [ref2007box], as they resulted by using our [VNA] and our [VNA]: Rho Magnitude: [VNA] : CYAN - pre-calibrated [VNA] : BLUE - Nominal Values [NanoVNA] : MAGENTA - pre-calibrated [NanoVNA] : RED - Nominal Values: Rho Argument: [VNA] : CYAN - pre-calibrated [VNA] : BLUE - Nominal Values [NanoVNA] : MAGENTA - pre-calibrated [NanoVNA] : RED - Nominal Values: Sincerely, gin&pez@arg 49# |
|
Re: Nanovna-F is here
I would be interested in finding that out also.
Thanks, Dick, W1KSZ Sent from Outlook<> ________________________________ From: [email protected] <[email protected]> on behalf of David Hartley <dehartley@...> Sent: Tuesday, October 15, 2019 4:11 PM To: [email protected] <[email protected]> Subject: Re: [nanovna-users] Nanovna-F is here Warren, where did you purchase your NanoVNA-F? Dave |
|
Re: Am I Fixing my BNC Calibration using Calibration Standards Adjustments in nanoVNA-Saver 0.1.2
On Tue, Oct 15, 2019 at 05:26 PM, Kurt Poulsen wrote:
Kurt, Please find attached below the Smith Chart after tuning the parameters. The S11 plot after tuning is the last chart in the first post #4963 above. I think there is a misspelling in so far you used F for the inductance and H for the capacitance. I was just reporting what is in the Calibration standards form on nanoVNA-Saver. I am not sure how these are implemented. The image of that form after tuning is attached as (LoadTuning.png) below as well. You may notice that I also did some minor tuning on the load and the Through offset delay. -- Bryan, WA5VAH |
|
Re: Am I Fixing my BNC Calibration using Calibration Standards Adjustments in nanoVNA-Saver 0.1.2
Hi Bryan
Congratulation. Tweeking parameters in NanoVNA-saver is great fun. You missed to show us a sweep of the coax cable after tuning to celebrate the victory ? I think there is a misspelling in so far you used F for the inductance and H for the capacitance, anyway it does not matter it is the numbers and the sign in front which provide the results. Kind regards Kurt -----Oprindelig meddelelse----- Fra: [email protected] <[email protected]> P? vegne af bryburns via Groups.Io Sendt: 15. oktober 2019 23:28 Til: [email protected] Emne: [nanovna-users] Am I Fixing my BNC Calibration using Calibration Standards Adjustments in nanoVNA-Saver 0.1.2 Folks, Because of convenience and several other factors I usually use my nanoVNA through SMA to female BNC connectors. As a result, I calibrate both the nanoVNA and nanoVNA-Saver through the SMA to BNC connectors. One reason I do this is to avoid changing the connector mating to the SMA connectors on the nanoVNA. I realize that BNC calibration kits are notoriously poor. Mine is no exception. The open I use is the female BNC to SMA adapter that is connected directly to the nanoVNA. Also, the BNC short that I use is likely to have some inductance because there is no insulation around the shorting pin. And, the shorting plane cannot be very close to the center pin of the female BNC even when it is properly connected to the mating connector. The 50-ohm load I am using is 50.0 ohms at DC based on a precision ohmmeter measurement. It does compare quite favorably with other terminations that are rated up to 3 GHz when measured on wider bandwidth VNAs. With this cal kit, and doing both the nanoVNA calibration and the nanoVNA-Saver calibration, I get a nice clean dot in all of the right places. The short does register on the left side of the Smith chart, the load in the center and the open at the far right side as it should be after the OSLT calibration. However, we know that issues remain. The big question is: How to fix them or determine what errors may exist? Some of the issues can be most easily observed by connecting a fairly short length of good, low-loss 50-ohm coax on the CH0 port and observing S11 on a Smith chart or looking at the return loss in dB over the frequency range from 50 kHz to 900 MHz. The cable used in the measurements shown in the attachments is a 20" (~0.5m) length of RG213 which has SMA connectors on both ends. It was connected to the BNC through a male-male BNC adapter and a female-BNC to female SMA connector. Yes, this is maximum adapter usage to connect the SMA cable to the SMA connector on the nanoVNA. But, the task here is to determine what is going on with the BNC calibration. My experience says that at frequencies below 900 MHz all of the connectors I am using should be pretty good. The first 2 plots I have attached show the result. I also used a commercially manufactured (surplus) piece of RG-233 with BNC male connectors with very similar results. So, this is definitely ans issue and most likely with the calibration. Taking a careful look at the figure below contained in "Smith-RG213-WithoutCalTuning.png" I observe at least a few problems. 1) The plot should stay within the Smith chart but does not. 2) The circles which should proceed around the Smith chart in a counterclockwise direction are a) not circles and b) not centered on 50 ohms. Similar issues are observed when looking at the plot in the file "S11-RG213-WithoutCalTuning.png". Clearly their should be no positive values in the plot but there are. Furthermore, there is significant ripple in the plot caused by issues that are not likely the short cable. So, I asked myself the question: "Can I manipulate the "Calibration standards" parameters in the nanoVNA-Saver application to fix what is wrong with the calibration?" I hasten to point out that I do not have access to high-quality measurement equipment and therefore I am left to figure out what is lacking in my BNC calibration kit. Here is what I did. I began to try first of all to fix the open calibration by continuing to observe the Smith chart as I changed the "open" parameters C0 and the offset delay. Also by blowing up the scale on S11 phase I could see that using the BNC female as an open there remained a negative going and somewhat quadratically increasing phase ramp at frequencies above 500 MHz which could be due to the capacitance in the BNC female I was using as an open. I found that I could get rid of this negative going and quadratically increasing phase ramp by setting the "Open C0" value to 1100 e-15 H. With this setting the "hook" was removed and a linear phase trend remained. However, with that correction in place, the entire phase had a linear phase ramp. Adjusting the time-delay offset to -55 ps I found that I could then level the phase and get less than 0.4 degrees of phase deviation from zero degrees all the way from 50 kHz to 900 MHz. This seemed to me to be a big improvement. Clearly this is an empirical result but it could be a real description of the difference between a female BNC connector and a real open circuit. After these two adjustments the curve shown in "S11-RG213-WithOpenOnlyTuning.png" was collected. This is an improvement in that the trend in the S11 data is now approximately correct; however, large ripples remain, During the adjustments of the open parameters, I continually looked at S11 in dB and S11 phase to make sure that what I was adjusting was making the phase more constant over frequency and S11 amplitude move in a correct direction (downward at the high end of the sweep.) I then began to adjust the tuning parameter "L0" for a short. After a few guesses, I arrived at an adjustment for the short parameter of 1200 e-12 F. Said another way, this is like 1.2 nH of inductance is in the short I am using when compared to a real short. This is plausible for the short which is clearly not at the reference point which I consider to be the end of the center conductor for the female BNC center pin of the adapter to SMA. After this additional adjustment I created the chart shown below as "S11-RG213-WithCalTuning.png". I observe the following 1) There are no longer positive values - a definite improvement. 2) The trend is fairly monotonic out to 900 MHz but not perfect. 3) I haven't included the data for this, but the return loss out to at least 300 MHz is now almost exactly (within <0.02 dB) twice the measured S21 loss of the RG213 cable. Before the adjustments listed here, this was clearly not the case. Because there seems to be a limitation of 4 attachments in this forum, I did not include the S11 Smith chart below; however, the plot is now completely within the Smith chart, the circles get smaller as the number of times around the Smith chart increases, and everything is much more nearly centered on the Smith chart. These are all what I expect for this fairly simple measurement. I can put the Smith chart in a separate post if there is interest in it. Is this really an improvement or am I kidding myself by tweeking parameters to improve the results of this simple measurement? My opinion is that it is a definite improvement and an approach that other may want to explore. I welcome your comments. -- Bryan, WA5VAH |
Starsekr
Warren Allgyer
gin&pez@arg
Dick
Kurt Poulsen