Injection transformer for bode plot - my DIY project

[nikifena]

Hi guys. I read a couple of topics here and over the net about how to make a DIY injection transformer that covers a lot of frequencies - usually, it is used to test power supplies for stability. As far as I know, the best is the  B-WIT 100 from Omicron Lab. So I found a photo of the internals - Dave did a teardown too, and I started to search for a suitable core. The core is "special" - not a ferrite core like most inductors. As I think the best solution is a core from Nanocrystalline. It is similar to the traditional toroidal transformer winded with sheet of metal, but the material is very thin and fragile. I measured it and it is 0.001mm. A piece of it just fell out.

So, I did some research and found this inductor from Wurth: 744839208072. Since I've been ordering a lot of Wurth, I asked for a free sample. I needed to remove the original wire, then to dremel some of the existing plastic material until I got a nice round ring core. I measured the inductance and with 41 turns (filling the entire core) I got about 250mH. The B-WIT 100 contains 40 turns. I used a twisted wire from an old lan cable.

There is a screenshot of the bode plot I did with my scope and I believe the final result is pretty good for a transformer for free   Just wondering why there is a peak at the high frequencies? The output of the transformer is not loaded. Can someone measure the inductance of the B-WIT 100 transformer?
What do you think guys about the graphs?

PS: I will do another capture but I will zoom the frequency response. Now it looks very flat and probably it isn't.

[nctnico]

The peak at high frequency is self resonance due to inductance and capacitance between the windings.

[TopQuark]

https://www.eevblog.com/forum/testgear/diy-transformer-for-use-with-bode-plots/
Some good discussions here. IMO the output should be 50ohm terminated for your measurements. A 1:1 transformer means 1:1 impedance ratio, so if you are using a 50 ohms signal source, the output of the transformer should be terminated with 50 ohms, same idea as how you should connect a 50 ohm source to a 50 ohm termination when using a direct coax cable.

Recently played with various cores to see if I can eek out a bit of performance with the bifilar transformer arrangement. The consensus has always been nanocrystalline cores are critical for the lower end of the frequency response, but the higher end of the frequency is usually limited to 10MHz. In my experiment, I used a large-ish cheap Chinese nanocrystalline core, and placed a couple of small Chinese knock-off type-8 micrometals core in the center in an attempt to extend the usable frequency pass 10MHz.

Results are not great, not terrible. I'd say the transformer is very usable from 10Hz to 10MHz, with the HF -3dB point around 13MHz. Don't think the micrometals core did much.

Note that as you approach 10MHz, the measurement setup has substantial influence to the measurement quality. For scope bode plots at low frequencies, I used a 10x probe connected in parallel to the transformer input with a T-connector as my reference input, and the output is connected to the scope's 50-ohm input. Pass 1MHz, I think a network analyser is a better tool for the job as it maintains a consistent 50 ohm transmission throughout the measurement path. I used a spectrum analyser with TG as a simple scalar network analyser, calibrated with a thru normalisation cal. The measured HF -3dB point is at 13MHz with this setup.

[nikifena]

Thanks!
The low frequencies from the scope generator are distorted due to the low-resistance coil. Your design has even fewer turns. What about that?

I will post more screenshots tomorrow.

[Weston]

The Wurth 744839208072 is $49(!) in digikey in quantity one?
 
It seems like the core is indeed nanocrystalline, but unless you have weird sourcing issues or a massive discount you would probably be better off buying the cores themselves, which are stocked on digikey. VACUUMSCHMELZE makes them.

Some people on the forum were using this core, and I have built a few injection transformers using it, but it seems to be out of stock right now https://www.digikey.com/en/products/detail/vacuumschmelze/T60006L2040W424/12530400

This core is pretty similar, but most of them should work. They all have really high Al values https://www.digikey.com/en/products/detail/vacuumschmelze/T60006L2045V102/12531748


Typically you are measuring the injection signal on the secondary side, so the response does not need to be very flat, just not too attenuating. Do you have any bandwidth goals? Driving with a 50 ohm source the -3dB low frequency point in Hz is 50/(2*pi*Lpri). With that second core I linked you would only need 30 turns for a -3dB point of 100Hz.

Also, just a heads up if you have not noticed, but with the bode plot functionality on the RTB scope you can not directly measure the low frequency -3dB point, as the reference signal is measured on the output of the function generator (after the internal 50 ohm output impedance), so you need to add an external 50 ohm resistor to emulate the driving impedance which determines the low frequency -3dB point.

[julian1]

Are the commercial units for power-supply loop testing - with potential HV present, also wound bifilar?
 

[TopQuark]

The low frequencies from the scope generator are distorted due to the low-resistance coil.

Not really, the low frequency distortion is mostly from core saturation. Magnetic flux is the integral of the voltage, lower frequency means higher magnetic flux build up per cycle, pushing the core towards saturation. You can see this effect in action if you lower the input waveform amplitude. See the screenshot of the same 10Hz signal, same transformer, but 10x different in amplitude. Any half decent function gen (Your scope's function gen certainly counts as one) can drive a low resistance or even short circuit just fine.

Your design has even fewer turns. What about that?

More turns introduces more parasitics and drive the core flux density higher. More turns is not always better. You can see a demonstration of the same core, but with as many windings as I can fit into the center of the core. The HF limit is reduced to less than 8MHz, down from over 13MHz.

[TopQuark]

Are the commercial units for power-supply loop testing - with potential HV present, also wound bifilar?

Not aware of any injection transformer with explicit HV rating, but I haven't looked very hard either. The B-WIT 100 is isolation rated to 600 V CAT II.

Some of the FRA machines out there straight up has isolated input/output, so no injection transformer is needed at all. Examples include:

https://www.venableinstruments.com/products/frequency-response-analyzers/6300-series
https://cleverscope.com/products/CS548 (Mainly oscilloscope with isolated input/output, but has a decent FRA function it seems)


If you are into HV or offline mains converter, chances are you will have access to a high pot tester for compliance testing, so you can just test your injection transformer like any old smps transformer/coupled inductor. (With that said, I don't have a hi pot tester, it is on my to-buy list though  )

[Hydron]

Looks rather familiar...

https://www.eevblog.com/forum/blog/eevblog-1104-omicron-labs-bode-100-teardown/msg1670324/#msg1670324

Mine was also a from a free sample, it's not an economical way to get a core otherwise.

[nikifena]

Thanks, guys!
I reduced the number of turns to get a better high-frequency response. Here is the result

[Weston]

One subtle thing to keep into account when testing: When wound with twisted pair these are transmission line transformers.

It looks like you are using a pair from CAT-5 cable, which is 100 ohms impedance, and you are terminating in to 50 ohms, but even so, there should be some transmission line effects.

Depending on which wire of the output is grounded you can get transmission line behavior. If the transformer is inverting and you have a common ground in your measurement setup it will act as a transmission line. If it is non-inverting its a normal transformer that likely has lower bandwidth.

A normal transformer is limited by the stray L and C, but in a transmission line transformer that gets folded into the transmission line and the bandwidth is really only limited by loss in the transmission line and impedance mismatch causing reflections.

Discussion to be had on if this actually impacts anything in application, but for loop injection one side should be low impedance, which is basically a ground. So you can use it like a transmission line transformer in some situations.


Also, for your measurements you should twist the input and output wires together to lower inductance, it might impact your measurements.

[TopQuark]

So I have had a bit of a breakthrough in the injection transformer design.

Usually in injection transformer builds, there's a resonance valley in the transmission path (S21) around 10MHz. I am thinking it has to do with a LC resonance between the interwinding capacitance and the leakage inductance of the transformer. The measured LC values match up with the measured resonance frequency.

Somehow the fix I found is by wrapping either the input pair or output pair around an EMC suppression ferrite. Or better yet, route the input pair against the output pair through the ferrite. See the attached photos on what I mean. The ferrites I used are TDK ZCAT clamp on type.

The way I am currently thinking about it is imagining the transformer as a perfect 1:1 transformer, existing as an invisible blip in an otherwise perfect coax path. Then the ferrite is just acting as a common mode choke, with the coax wrapped around it.

Anyways, now I am able to supress the S21 resonant valley around 10MHz, and have a smooth roll-off past 10MHz. The -3dB is now 20MHz, and very usable as an injection transformer to 50MHz IMO.