Chasing ghosts??? #internals #calibration
|
In all honesty I was not sure if the approach I was following to determine imperfections in a calibration kit made sense. Various wise people on this group have given me hints which, due to lack of knowledge on my side, I probably failed to understand.
So I decided to to some more investigations. I used THREE different VNA's VNA 1: NanoVNA (we all love) VNA 2: Home build nanoVNA equivalent. VNA 3: Home build 3 GHz VNA (uses same type of resistive bridge and measurement principles as nanoVNA but it does not use overtone mode) and ONE calibration kit and ONE set of cables. I measured on all VNA's the same calibration kit directly on the connector for calibration, the same 1 meter RG58 cable with OSL and the same 30cm semi rigid cable with OSL All measurements where done till 900MHz. The calibration kit is assumed to be perfect. No compensation factors where used. Any difference seen between these measurements should come from the difference in VNA's After calibration all three VNA's show for the calibration kit a perfect calibration, no difference is seen. In attached pictures you will see the calibrated plots of the measurements on the three different VNA's of the logmag S11 of the long cable (RG58) and the medium cable (semi rigid) with open and short at the end. The only difference between the three pictures is the VNA. Cables and calibration kit are the same. It is safe to conclude there are substantial differences between the three VNA's First picture: VNA 1 (nanoVNA) seems at low frequencies to have a Z0 that best matches the impedance of the cables but shows a substantial deviation above 600MHz Second picture: VNA 2 (home build VNA) has a bit worse Z0 match and seems to have a S11 open of both cables that goes off track above 800MHz Third picture: VNA 3 (3GHzVNA) becomes very noisy above 550MHz (need to investigate why as this is new) but the oscillations of the logmag S11 show no sign of substantial deviation apart from gradual increase in amplitude and the cable loss in dB stays nicely linear with frequency This comparison underpins my initial assumption that the nanoVNA needs a bridge error model to correct the differences (in particular above 300MHz) and until these threeVNA's (with their individual bridge error models) give the same measurement results I better not try to calculate calibration kit error parameters Feedback is welcome -- Erik, PD0EK |
|
Hi Erik
Interesting test indeed. I understand that when you did not see any difference between the 3 VNA's and calibrating directly at the 3 VNA's it meant at a SMA female adaptor using the SMA male kit supplied with the NanoVNA. hat is quite OK and understandable they show same resulting S11 reflection. When you use the 1 m cable and semirigid cable you do not explain if it is a SMA male male cable. If so you probably use a female female adaptor to do the short/open test, and if that is the supplied SMA "barrel" you have a loss to consider, which is "mudding" the test, as its reflection impact is different which cable you use. Anyway you are right the 3 bridges are not compensated by any bridge compensation model, which explains differences It is quite a task to implement the 6/12 terms error model like done for the VNWA (in ted software) and the FA-VA5 (in the hardware and software when suing VNWA software), but maybe Rune will pick it up at some point in time for the NanoVNA-saver ? Kind regards Kurt -----Oprindelig meddelelse----- Fra: [email protected] <[email protected]> P? vegne af erik@... Sendt: 3. november 2019 14:34 Til: [email protected] Emne: [nanovna-users] Chasing ghosts??? #internals #calibration In all honesty I was not sure if the approach I was following to determine imperfections in a calibration kit made sense. Various wise people on this group have given me hints which, due to lack of knowledge on my side, I probably failed to understand. So I decided to to some more investigations. I used THREE different VNA's VNA 1: NanoVNA (we all love) VNA 2: Home build nanoVNA equivalent. VNA 3: Home build 3 GHz VNA (uses same type of resistive bridge and measurement principles as nanoVNA but it does not use overtone mode) and ONE calibration kit and ONE set of cables. I measured on all VNA's the same calibration kit directly on the connector for calibration, the same 1 meter RG58 cable with OSL and the same 30cm semi rigid cable with OSL All measurements where done till 900MHz. The calibration kit is assumed to be perfect. No compensation factors where used. Any difference seen between these measurements should come from the difference in VNA's After calibration all three VNA's show for the calibration kit a perfect calibration, no difference is seen. In attached pictures you will see the calibrated plots of the measurements on the three different VNA's of the logmag S11 of the long cable (RG58) and the medium cable (semi rigid) with open and short at the end. The only difference between the three pictures is the VNA. Cables and calibration kit are the same. It is safe to conclude there are substantial differences between the three VNA's First picture: VNA 1 (nanoVNA) seems at low frequencies to have a Z0 that best matches the impedance of the cables but shows a substantial deviation above 600MHz Second picture: VNA 2 (home build VNA) has a bit worse Z0 match and seems to have a S11 open of both cables that goes off track above 800MHz Third picture: VNA 3 (3GHzVNA) becomes very noisy above 550MHz (need to investigate why as this is new) but the oscillations of the logmag S11 show no sign of substantial deviation apart from gradual increase in amplitude and the cable loss in dB stays nicely linear with frequency This comparison underpins my initial assumption that the nanoVNA needs a bridge error model to correct the differences (in particular above 300MHz) and until these threeVNA's (with their individual bridge error models) give the same measurement results I better not try to calculate calibration kit error parameters Feedback is welcome -- Erik, PD0EK |
I measured on all VNA's the same calibration kit directly on the connector forIs it practical to capture uncorrected data for OSL and DUTs from each VNA, then perform calibration corrections using some PC software, to better isolate hardware from firmware effects? |
Thanks for the links.!One problem is that scikit-rf wants S11 files for ideal SOL but I did not find how scikit-rf can generate S11 from e.g. delay, R, L, C values. Since I have SDR-Kits BNC cal kit, perhaps S11 for those can be generated using VNWA software, but scikit-rf also seemingly does no interpolation; raw measurement frequencies must match ideal S11, so that will be another challenge. |
|
Hi Oristo
When the cal kit calibration kit is loaded in the VNWA software you simply open a custom trace and in the Expression line write s_load or s_open or s_short or s_thrumatch or s_thru then the trace is shown according for the custom trace to be shown as you like in Smith, dB of what ever representation and then to save directly as s1p file by a right click on the trace and use the "Export to s1p file" function. You must decide on frequency span and number of points in advance Simpe as that ? Kind regards Kurt Fra: [email protected] <[email protected]> P? vegne af Oristo Sendt: 3. november 2019 20:42 Til: [email protected] Emne: Re: [nanovna-users] Chasing ghosts??? #internals #calibration Thanks for the links.! One problem is that scikit-rf wants S11 files for ideal SOL but I did not find how scikit-rf can generate S11 from e.g. delay, R, L, C values. Since I have SDR-Kits BNC cal kit, perhaps S11 for those can be generated using VNWA software, but scikit-rf also seemingly does no interpolation; raw measurement frequencies must match ideal S11, so that will be another challenge. |
|
Hi Kurt -
When the calibration kit is loaded in VNWA software.. and another piece of the puzzle falls into place.. Thanks ever so much! |
|
On Sun, Nov 3, 2019 at 08:34 AM, <erik@...> wrote:
VNA 1 (nanoVNA) seems at low frequencies to have a Z0 that best matches theI am new to VNAs, but have wondered during these past few weeks as I experimented with my NanoVNA-H if it is better to calibrate /within/ the 3 bands that use different harmonics. 0-299MHz 301-599MHz 601-900MHz I suggest this, because I don't know where the Nano switches modes, and staying within the harmonic bands should guarantee that each cal uses a single harmonic. I do all my cals this way, and don't see a penalty, because as a ham, I don't expect to use the same antenna for HF and (example) U/VHF. I think that I need to recal for each geometry anyway, so no point trying to cal to the whole bandwidth. I see all these traces, and it strikes me that we are looking at a sin(f)*e^(-f)...that is, some sort of sine wave response that decays exponentially. Just a guess...but still wonder what you results would be in each of the 3 Nano harmonic bands. -- On the banks of the Piscataqua Rich NE1EE |
I am wondering what mechanism causes S11 impact depending on a cable (theRotations are mostly by cable delay, and I guess that differences from 180 degrees between rotations for open and short measurements are provoked by non-ideal open and short calibration characterization. This can be simulated e.g. @ 5 MHz with 2m LMR-400 by substituting 3.5 wire turns for short and 390pf for open. I intend to experiment for reducing this using VNWA software.. |
|
Hi Gabriel -
From those graphs it's obvious your calibration is way off because S11 exceedsI guess you meant message 6456, and agree that was a poor choice of graphs. I do not calibrate for such a wide range. So, second attempt: Looking at logmag for CH0 and CH1 terminated by 50 Ohms (AKA Isolation), then you will notice that return loss gets poorer > 100MHz. |
|
Attached the plots you requested
All data from 3 VNA's calibrated with the same calibration kit measuring the same two cables 1: nanoVNA 2: ownVNA 3: 3GHzVNA No VNA specific corrections applied Top right is the smith chart plot of the calibration kit to prove the calibration was done correctly. I use "+" to plot the data otherwise it would be impossible to see the dot's as they are exactly on the same location and too small to see The smith charts of the cables (bottom two) include the plot where the reference plane is set to the end of the cable (all VNA's with exactly the same shift per cable, labeled with "rotated") so you can easily see the remaining errors And yes, some cables move outside the smith chart at higher frequencies and that is indeed the problem I try to solve with a VNA specific error correction approach -- Erik, PD0EK |
|
And here is a hint a first order cable correction (assuming the cable impedance is not exactly Z0) can improve the results
The model corrects for the long cable impedance mismatch for the 3GHz VNA Mismatch is identified to be frequency independent and around -0.015-0.769i ohm, impact is visible in the upper left chart yellow/orange logmag S11 of the long cable which becomes more of a straight line. Now I can use this approach to reduce the cable impact on the other VNA's and see what is left to correct in the bridge. -- Erik, PD0EK |
|
And here is the nanoVNA data with the impedance correction for the long cable ( -0.075 -0.56i ohm)
Substantial oscillations are still there and these do NOT come from the calibration kit or the cable so must come from the bridge or an interaction between the bridge and the cable. Next step in modelling to be taken -- Erik, PD0EK |
Kurt Poulsen
Oristo
Rich NE1EE
