Relaxation oscillation in LC resonance/ Toko testers

[regenfreak]

It may seem to be completely daft to try to measure this directly with a 50R VNA as the high Z secondary is hopelessly mismatched away from the intended 2450R when both primary and secondary are connected to the 50R ports of a VNA.

I suppose this is connected under unun configuration with each legs of the primary and secondary to ground. The frequency response should be fairly flat as its main function to do impedance matching with minimum insertion losses.

This is a genuine full 2 port s11 s21 s12 s22 data file taken with a nanovna.[I've had to change the file extension to .txt to get it to upload but you can change it back to .s2p once downloaded.

Woow so i was wrong. The NanoVNA dumps out S21 and S22 data too I am still having trouble to view the .s1p and .s2p files using Window 7. S21 and S22 are kind of new animals to me. I have not thought about them much.

However, because the VNA has measured all 4 s-parameters s11 s21 s12 s22 then the mismatch loss in the S21 measurement is kind of captured (and stored) in the highly reflective s11 and s22 measurements.

This means that if you take the 2 port model of the transformer and load it into a simulator (eg LTSpice) and change the port impedance of the secondary to 2450R then the transformer should then show the correct (low) insertion loss. This probably sounds too good to be true but this is correct.

I demo'd it with a youtube video a while back (sadly no sound) and this compares the insertion loss against the classic method where a 2400R resistor is placed in series with the secondary and a basic s21 sweep is made with a lab VNA. This will add 16.9dB of error in the insertion loss but this is corrected in the simulator. This shows the correct insertion loss in the purple trace in the top RH corner of the video.

To cut a long story short, if the nanovna derived 49:1 transformer model is any good it should show the same insertion loss as the top purple trace when the secondary port is set to 2450R in the simulator. You can see that the nanovna model nails this test really well at about 53 seconds into the video at 0:53 where the port is tweaked up to 2450 ohms and the insertion loss at about 20MHz is about 1dB.

yes i get the idea. Literally you can correct the insertion loss error in the simulator later in the post-processing by tweaking the load to an equivalent resistive load of 2450 ohms

When I was a spotty student in the early 1980s I remember visiting a G8 radio ham as he had a spectrum analyser. At the time this was the equivalent of someone having an exotic supercar and it was the first time I'd ever seen a proper spectrum analyser. I was quite excited to even be in the same room as that analyser and I felt like a privileged guest!

It was also slightly amusing because he seemed reluctant to switch it on to demo it. I did wonder if it was broken. Either that or he only wanted to power it up when it was really needed. I can't remember what model it was but it was something clunky from the 1960s and probably worthless to anyone but a collector today.

I just wonder what will be available in another 5-10 years' time from the far east for £100 or so...

Ha ha like a an exotic supercar

Whoops! I just had a look at the s-parameter data again and I forgot that my nanovna exports the data as # Hz S RI R 50 rather than the classic # HZ S MA R 50.Therefore, the exported data in the s1p file isn't the classic mag and angle reflection coefficient data. However, the sign still changes when the 1 port measurement goes through resonance at 10.54MHz.

I can see the sign change for angle is located near the zero crossing point. I am trying to relate Img S11 to zero angle in my head. When Img S11 changes sign, the S11 vector flips from above the positive angle to negative angle in the polar diagram? I have to keep whacking my head to remind myself that S11 is vector and not a scalar!

I'll demo this tomorrow and hopefully you will be impressed with the results the nanovna can deliver when it produces a transformer model.

I've just removed a 10.7MHz IF coil from an old CB radio and I'll do a 2 port measurement using both the lab VNA and the nanovna tomorrow. This is one place where the lab VNA will be a bit better because I can apply full 10 or 12 term error correction during the calibration process. However, the nanovna should still produce a highly useable transformer model even if the full s2p file shows some obvious (but small) errors.

I am looking forward seeing it Thanks for doing it

I ve attached the nanovna derived s2p data file from that 49:1 transformer video below and you could have a play with it in LTSPice or QUCS or maybe RFSIM99 if you can run it on your computer.]

I have no experience with LTSpice but i have used Elsie to model bandpass filters which is not as flexible as LTSpice.

[G0HZU_JMR]

Woow so i was wrong. The NanoVNA dumps out S21 and S22 data too

Note that to do this correctly the external software has to allow the user to swap the test fixture around. i.e. the software should allow the user to dump out s11 and s21 and then it should prompt the user to flip the device backwards so port 1 is now port 2. Then it should dump out what is effectively s12 and s22 and then absorb it all into one file of s11 s21 s12 s22.

I'm not sure the nanovna saver software allows this feature yet although it really should have been there right from the start.

Probably the first thing to do is check how good your cal kit and test fixture is organised for a simple 1 port measurement of s11. If you are using the bundled nanovna SMA cal kit then it will be very fiddly to set this up such that you can measure a resonator correctly on a test PCB. It can be done but it will involve some nerdy cal kit corrections.

This is where the vast majority of nanovna users seem to slip up. They use the bundled SMA cal kit and end up with an incorrect test fixture calibration and they get duff results when they measure a real test network. They then condemn the nanovna hardware when really they haven't given it a chance.

I can show you a basic cal kit and test fixture system but each connector costs about £4 and you will need three for the cal kit and then at least two more to use in a test fixture. Realistically you would be looking at about £25 in parts to make up a decent cal kit and fixture that would work extremely well up to 50MHz. Above 50MHz the nanovna starts to limit the performance although it can still work quite well up to about 300MHz.

[regenfreak]

Note that to do this correctly the external software has to allow the user to swap the test fixture around. i.e. the software should allow the user to dump out s11 and s21 and then it should prompt the user to flip the device backwards so port 1 is now port 2. Then it should dump out what is effectively s12 and s22 and then absorb it all into one file of s11 s21 s12 s22.

It makes sense to me. This essentially replaces an expensive VNA that has b-directional RX and TX in both two ports. The question is:under what scenario do we need S22 and S21 in a real world application?

Probably the first thing to do is check how good your cal kit and test fixture is organised for a simple 1 port measurement of s11. If you are using the bundled nanovna SMA cal kit then it will be very fiddly to set this up such that you can measure a resonator correctly on a test PCB. It can be done but it will involve some nerdy cal kit corrections.

For Toko small transformer, I made a calibration rig board (see photo) using two 100 ohms smd resistor in parallel, and the short is a wire. I used test sockets that would fit toko transformers or crystals. I just made sure the calibration planes for open, short and load are roughly the same.
For bulky valve type IF transformers, i have to restore to croc clips(see photo) . In calibration, i used miniature non-inductive 50 ohms ceramic resistor as load. Once i tried this on 455khz and 10.7MHz IFs installed in the chassis of vintage radios, the results could be affected by movement of the cables. Once i built a test board using micro (expensive) silver component tester clips, they were very fiddly to use!!!

I can show you a basic cal kit and test fixture system but each connector costs about £4 and you will need three for the cal kit and then at least two more to use in a test fixture. Realistically you would be looking at about £25 in parts to make up a decent cal kit and fixture that would work extremely well up to 50MHz. Above 50MHz the nanovna starts to limit the performance although it can still work quite well up to about 300MHz.
Today 11:43 am

Yes please. I would love to see how you do it it I always thought the calibration kit comes with the NanoVNA are very good for 10.7mHz.


I managed to view the .sp1 file on my andoid phone using Touchstone Viewer. It only plots the graph automatically but does not show the raw data. What software do you use to view the .sp1 data and convert them to txt?

[regenfreak]

I have downloaded QucsStudio, it can open the HZ S RI R50 file, sweet!!!

[G0HZU_JMR]

The question is:under what scenario do we need S22 and S21 in a real world application?

It's useful to also measure s22 and s12 when measuring/modelling filters that aren't designed for 50 ohms and also for modelling active devices. It can also be useful for modelling transformers.

For example, I think that an ideal 1 to 4 (impedance) transformer would have the following s-parameters (note this is the conventional mag angle MA format)

# MHZ S MA R 50
freq s11 s21 s12 s22
10.7 0.6 -180 0.8 0 0.8 0 0.6 0

You can see that s21 and s12 are the same but the reflection coefficient for S11 and s22 are not the same. S11 has an angle of 180 degrees and s22 has an angle of 0 degrees. This is because this network is a transformer.

Here's the nearest 10.8MHz s2p data line from a really good 4 to 1 transmission line transformer I designed a few years ago.

10.8 0.59759 174.07643 0.79433 -2.73648 0.79396 -2.67945 0.59604 0.45719

You can see it is pretty close to ideal, the 174 degrees angle of s11 is very close to the -180 of the ideal and the s22 angle is very close to 0 degrees.

If I didn't measure s12 and s22 and just reused the same data as s11 and S21 I think the transformer model would misbehave quite badly in a simulator. It would probably mimic a fairly strange active device rather than a passive device.

[G0HZU_JMR]

Originally Posted by regenfreak

I have downloaded QucsStudio, it can open the HZ S RI R50 file, sweet!!!

Have a play with the s2p files below.

One of them is the really nice (LF to 50MHz) 4 to 1 impedance transformer I designed a few years ago. It is designed for 50R at port 1 and 200R at port 2. The insertion loss is really low and the port match is excellent across a few MHz to about 50MHz.

There are also a couple of files for the Toko 10.7MHz transformer from the CB radio. This transformer appears to work best (lowest insertion loss) when port 1 is about 50 ohms and port 2 is 370 ohms.

Note that the data format is different as two of these files use the classic MA format but QUCS should cope OK I hope.

[G0HZU_JMR]

I'll set up the nanovna to have a go at producing a valid s2p model of the CB transformer. To do this I have to swap the fixture around part way through the process.

I think that within a few years (at most?) there will be a full 2 port nanovna that can do this without having to do the annoying swap around.

My first VNA was an old 3GHz HP8714B and like the nanovna this is classed as a T/R VNA because it can only measure S11 and S21. However, it did come with an external port switch box made by HP that permits full 2 port measurements. However, this is really fiddly to set up. I haven't used this old VNA for many years now.

[regenfreak]

I'll set up the nanovna to have a go at producing a valid s2p model of the CB transformer. To do this I have to swap the fixture around part way through the process.

Thanks. I like to see how a through measurement is done and your DIY calibration kit.

I have done a reflection measurement of a Toko FM using the test board i made a long time ago. I yet figure out a way of increasing the number of sweep points.... The Nano F can display Img S11(2nd photo) but not phase angle.But the spike of reflection S11 logmag is very sharp anyway. For a long time, i thought the frequency of spike was adversely shifted by the large impedance mismatch. The resonance frequency is around 10.826MHz. I really try to make sense of all the numbers at this stage.

[G0HZU_JMR]

I think that in the case of a conventional transformer with a tuned secondary the resonance frequency will also be affected by how you terminate the low impedance (untuned) winding.

You should be able to show this for the 10.7MHz CB transformer in QUCS if you mess with the port impedances. It looks like the resonant frequency with the input port unterminated is about 10.49MHz but if the input is left at a 50 ohm termination the resonant frequency goes up to about 11.00MHz. However, I don't think the transformer is used like that in the CB circuit. It is fed from a much higher impedance at port 1.

Port 1 only has to be a few hundred ohms for the resonant frequency to stick close to 10.50MHz. I'm not sure how well adjusted this transformer is because it came from a 40 year old scrap CB board. It's L6 from an old mk1 Cobra 148GTL-DX CB and it may have been tweaked by whoever trashed the CB before I was given the salvaged PCB as a parts donor board way back in the 1980s.

[regenfreak]

sorry the previous test was configured incorrectly in the test board. I forgot that the sockets were configured for transmission measurement and not reflection. So the results are invalid

[G0HZU_JMR]

My cal kit for the nanovna is made usng these 18GHz SMA end launch connectors from Johnson. Sadly, the prices have gone up a bit recently.

I usually buy them in bags of 25 and they cost about £4 each (£100 for a bag of 25). However, Farnell seem to want about £7 for just one.

I'm about to watch the football but later I'll show you how to make a cal kit using these connectors.

It does require some careful work to make a decent open and short. Ideally, the difference in delay between the open and short should be less than 2ps if you want to make fairly critical measurements and this requires some care. Otherwise, the delay error can be corrected electronically using something like nanovna saver.

Once you realise how critical the delay accuracy is you will realise how poor your current cal kit system is. The short looks like you have used a link wire for example. When full s2p measurements are made the through cal needs to be corrected for delay offsets/errors as well.

[regenfreak]

I think that in the case of a conventional transformer with a tuned secondary the resonance frequency will also be affected by how you terminate the low impedance (untuned) winding.

You should be able to show this for the 10.7MHz CB transformer in QUCS if you mess with the port impedances. It looks like the resonant frequency with the input port unterminated is about 10.49MHz but if the input is left at a 50 ohm termination the resonant frequency goes up to about 11.00MHz. However, I don't think the transformer is used like that in the CB circuit. It is fed from a much higher impedance at port 1.

This goes back to my original doubt-how valid the the VNA measurement if there is a significant input or output impedance mismatch? It is like if you use a cheap old style analogue voltmeter with internal impedance of 100k to measure the resistance of a 10M resistor.

[G0HZU_JMR]

Note that you don't 'have' to use these expensive connectors for doing stuff at 10.7MHz. There are cheaper SMA end launch connectors that can be used for making a decent cal kit. However, the delay issue is still relevant.

These connectors have something like 53ps delay (from memory) and they are only about 13mm long. So to get a delay error down to less than 5ps the open and short have to be within 13mm/10 = 1.3mm of each other. However, I would recommend trying much harder than this. It is possible to make the cal kit such that the residual delay error is much less than 2ps and the rest can be cal'd out electronically for the most critical measurements.

Many people will probably dismiss this as 'phoolery' at just 10.7MHz but it depends on how accurately you might want to measure the ESR of an inductor or capacitor at 10.7MHz. A 5ps error becomes a big deal here even at these frequencies.

[regenfreak]

My cal kit for the nanovna is made usng these 18GHz SMA end launch connectors from Johnson. Sadly, the prices have gone up a bit recently.

I usually buy them in bags of 25 and they cost about £4 each (£100 for a bag of 25). However, Farnell seem to want about £7 for just one.

I'm about to watch the football but later I'll show you how to make a cal kit using these connectors.

It does require some careful work to make a decent open and short. Ideally, the difference in delay between the open and short should be less than 2ps if you want to make fairly critical measurements and this requires some care. Otherwise, the delay error can be corrected electronically using something like nanovna saver.

Once you realise how critical the delay accuracy is you will realise how poor your current cal kit system is. The short looks like you have used a link wire for example. When full s2p measurements are made the through cal needs to be corrected for delay offsets/errors as well.

Thanks. The NanoVNA V 2 has the delay menu but i dont know how to use it.

[G0HZU_JMR]

This goes back to my original doubt-how valid the the VNA measurement if there is a significant input or output impedance mismatch? It is like if you use a cheap old style analogue voltmeter with internal impedance of 100k to measure the resistance of a 10M resistor.

But it surely isn't fair to the VNA to measure the transformer if you then want to change something about how the network is loaded. Ideally you should include that loading effect when making a 1 port measurement.

If you make a two port measurement of the transformer you can add whatever load you like in a simulator at a later time.

You could put the 2 port transformer model into LTSpice along with a model of the rest of the IF circuitry. eg resistors caps and transistors and model the whole thing. That is the primary function of a 2 port VNA. It exports a valid network model of whatever is between the two test ports. This can then be used as a component in a simulator.

[G0HZU_JMR]

[QUOTE=regenfreak;1385574]

Thanks. The NanoVNA V 2 has the delay menu but i dont know how to use it.

I think the edelay command for my nanovnaH has a different scaling factor to a regular lab VNA from HP/Agilent. I recall I had to scale it by two. I use the edelay command to improve the cal kit calibration by a few ps.

[regenfreak]

If you make a two port measurement of the transformer you can add whatever load you like in a simulator at a later time.

Sure. I am not up for LTSpice at this stage yet. I dont have much experience with the software.

Here i tested a valve double tuned FM 10.7MHz transformer using S21 with each legs to ground. The minimum of logmag S21 is at 13.8MHz. The use of crocodile clips are unavoidable in real applications involving bulky components in vintage radios. So the question is: can the measurements be trusted or it is meaningless on its own? There is always uncertainty putting aside the issue of calibration kit.

[regenfreak]

Here is the S21 sweep of a DIY 2nd-order Butterworth filter designed for 40m band. It uses transformer coupling for both 50 ohms input and output impedance matching. So I am confident this is is correct as I did hand calculations using the method described in Zverev's book and simulation using Elsie. The results are fairly close.

[G0HZU_JMR]

The double tuned transformer innards are unknown to me and I'm not sure how you have arranged the connections but I did try sweeping through a parallel resonator on a simulator with 50 ohm ports. In order to get the same response and selectivity as your plot I ended up with 11.2uH and 12pF.
This only needs an external 8pF to tune it down to 10.7MHz.

The plot below is just a single resonator being swept so probably nothing like what is inside your IF can.

[regenfreak]

My cal kit for the nanovna is made usng these 18GHz SMA end launch connectors from Johnson. Sadly, the prices have gone up a bit recently.

I usually buy them in bags of 25 and they cost about £4 each (£100 for a bag of 25). However, Farnell seem to want about £7 for just one.

Unfortunately i dont have a trade account as a hobbyist. I have to get by with the normal ones. I have found a spare board and replaced the short wire.

I have done a measurement of series and parallel resonance frequencies of a 10.7mHz crystal. Since the impedance of a crystal may be the order of several hundred ohms, the impedance mismatch is not bad, therefore the S21 measurements can be useful in the characterization. The full steps for the crystal parameter measurements here:
https://www.qsl.net/w2aew/youtube/cr...ts_nanovna.pdf