DIY Injection Transformer for Power Supply Control Loop Response Measurements

[diyaudio]

Its clear an Injection Transformer is required to perform signal isolation when performing loop response measurements,
however they cost big bucks, a Picotest J2101A costs $525.00 


https://www.picotest.com/products_J2101A.html





I want to start building a DIY Injection Transformer for my MSOX3104T, does anyone know what transformer material I can use to support a 1:1 10Hz to +/-40MHz signals.


   

[jahonen]

I have just used an ordinary mains transformer (230/5 volts or something like that, voltages are not that critical), I think it works sufficiently well up to at least few tens of kHz, but that is often enough for switch mode regulators. When both reference and output are measured, the transformer passband flatness is not so critical. I actually used a 192 kHz sound card and ARTA software for the measurements, good enough for most of my needs. This ordinary mains transformer is also used in this TI appnote: http://www.ti.com/lit/an/snva364a/snva364a.pdf

Regards,
Janne

[2N3055]

a bifilar winding on a toroidal ferrite core... Maybe a bit of experiment as to how many turns...

[MatteoX]

It is extremely difficult to acheive 10Hz - 45MHz transformer bandwidth (that is 6 1/2 decades or 22 octaves!).
Picotest mentiones use of specially annealed cores. If you look at Picotest J2101A Data Sheet, you will notice
resonance at 11MHz

https://www.picotest.com/downloads/INJECTORS/J2100A-J2101A_Spec_Sheet_FinalV2_Email.pdf

Many years ago I tried to make an isolation transformer for the same purpose. My goal was 10Hz to 1Mhz bandwidth
but I couldn't reach it so I give up after several experiments.

There were several discussions on the Internet that critiqued TI app note listed above and its use of mains
transformer.

I would be really curious if someone knows how to acheive such bandwidths.

Maxim App note 325  https://www.maximintegrated.com/en/app-notes/index.mvp/id/3245 has
some interesting details

[2N3055]

50-100Hz to few Mhz shouldn't be problem on common ferrite cores...
45 MHz not so easy, but I see no reason to that high on testing control loops on PSU...

22 nsec rise time in PSU control loop not so common...  Most PSU have bandwidth in 10s of kHz
testing to 100-200 kHz to make sure all is stable is good practice...
Testing to few MHz for fast PSU I get... 45 MHz i have no use for..

[2N3055]

I would be really curious if someone knows how to acheive such bandwidths.

Well probably it is Rogowski transformer or a compensated multi coil on a special core...

[T3sl4co1l]

Curious how much isolation capacitance it has.

That's a good price for such a fine transformer.  If you actually need that bandwidth, I suggest buying it!

If you don't, then reconsider how much bandwidth you actually need, and go from there.

Tim

[MagicSmoker]

It's the low end of the frequency range that is the hardest to achieve with a transformer. This is because operating down to a low frequency requires a very high permeability (mu) core and lots of turns, but a high mu core tends to have high losses vs. frequency, and lots of turns means lots of distributed capacitance (ie - both act to roll off the high frequency response). Note that you must design the transformer for the lowest expected frequency to avoid saturation.

If you want to test low bandwidth switchers like, e.g., boost PFC converters, then you really do need bandwidth to extend down below 1Hz, because the crossover frequency for the voltage loop is usually in the range of 5-15Hz.

Conversely, bandwidth exceeding a few hundred kHz will rarely be necessary unless you are designing state of the art GaN or Class-E/F switchers running at 1MHz+ - at a minimum you only need to test a little beyond the crossover frequency to ensure loop gain remains below unity without popping back up again or that the phase margin is adequate in and around the crossover region.

All that said, this would be an impossible task for a single transformer except for one thing: the signal level is usually in the range of 10mVpp (and never more than 100mVpp; at least not in my experience). This means you can get away with relatively few turns without running afoul of Faraday and Lenz.

Alternatively, you can use an isolation amplifier to inject the signal; an option that is arguably more popular these days.

[Floyo]

I dabbled a bit with this after being unsatisfied with the mains transformer technique, or using a common mode choke as Jay_Diddy_B did in one of his threads. I then found some inductors I have had lying around for ages, I think they came out of some telecom gear. They have a stupid amount of primary turns, and a stupid high permeability. the primary inductance is around18 Henry(!). I plotted the response with my analog discovery, see attachments.

The secondary is terminated into 50Ohms, and the primary is being fed in "low impedance" output mode of the AD signal gen.  Resonance happens at around 5.5Mhz, and the low cutoff is around 5 Hz, though you could get a bit more out of it by virtue of the loop measurements being relative, and by changing the drive impedance.  If there is interest I could give some more details, but I'm at work right now, so that will have to wait a bit.

[MagicSmoker]

What instrument are you using to pick up such a small amplitude signal in a switching enviroment that at single digit frequencies would give a ratio in the 80-100dB range?...

Not sure where you are going with this... the isolation transformer we are talking about here is for injecting a very small signal into the feedback network which results in a much larger variation in the switcher's output voltage (or current, if that is what the control loop regulates).

[Kleinstein]

If you don't use the transformer very heavily, you could get away in splitting the measurement and use 2 separate transformers, one for the low frequency (e.g. 1 Hz -1 kHz) and one for the higher frequencies (e.g. 100 Hz - 10 MHz). This very much eases on the design. Some overlap can also help to check for limitations.

For the low frequency range one could even use some kind of balances current injection or other electronic circuit instead of a transformer - so one could measure down to essentially DC.

[2N3055]

 AcHmed99,

I think you didn't quite understand how it's done..

You don't inject current into PSU output.. You take PSU, you break feedback loop and insert injection transformer in series with feedback loop.. So it gets amplified by error amplifier, not suppressed......

[diyaudio]

I dabbled a bit with this after being unsatisfied with the mains transformer technique, or using a common mode choke as Jay_Diddy_B did in one of his threads. I then found some inductors I have had lying around for ages, I think they came out of some telecom gear. They have a stupid amount of primary turns, and a stupid high permeability. the primary inductance is around18 Henry(!). I plotted the response with my analog discovery, see attachments.

The secondary is terminated into 50Ohms, and the primary is being fed in "low impedance" output mode of the AD signal gen.  Resonance happens at around 5.5Mhz, and the low cutoff is around 5 Hz, though you could get a bit more out of it by virtue of the loop measurements being relative, and by changing the drive impedance.  If there is interest I could give some more details, but I'm at work right now, so that will have to wait a bit.

This looks very interesting, I too saw some possible isolation transformers in modem equipment I may possibly use one, can you upload some more details on your transformer?

As a side note: I don't think I need that high frequency response for the transformer is required...up to 1MHz signal will do to test compensation loops.


   
   

[Jay_Diddy_B]

Hi,

In this thread:


https://www.eevblog.com/forum/projects/dynamic-load-bode-plot-using-hp-35665a-dsa/msg309192/#msg309192

I showed how I used a common mode choke for injecting the signals.

In this message I showed how to estimate the useful bandwidth of the transformer based on magnetizing inductance and leakage inductance.

https://www.eevblog.com/forum/projects/dynamic-load-bode-plot-using-hp-35665a-dsa/msg310991/#msg310991


I prefer to use my HP3577A VNA analyzer for doing this measurement. It has tracking filters, which can be set as narrow as 10Hz, these reject all the frequencies except the one that you are injecting. The narrow bandwidth helps lower the noise floor.

The injection transformer does not need to be flat. The transfer function of the transformer is not part of the measurement. You are measuring on either side of the transformer with respect to ground.

The injection transformer is connected between the output and the top of the divider resistor. The control loop will try and hold the top of the divider constant so all the injected signal should appear on the output. At low frequencies the voltage on the divider should be very small. I normally inject a signal that is 0.5 to 1% of the output voltage. so for a 5V supply I will use 25mV rms. If the loop gain is 40dB only 25/100 =0.25mV is at the top of the divider. As the injection frequency is increased, the loop gain is reduced and the ratio of the signals changes. You do not want to make the injection amplitude too big. If you do you may not be measuring the small signal behaviour.

Since this I have developed a new injection transformer that uses a common choke with 6.5mH of inductance.

I believe that the Picotest injector, must use special high frequency techniques, like a Pearson current transformer.

I would love to see a teardown of Picotest transformer or any of the other picotest units.

There is another way of doing this, using op-amps. This would work very well at low frequencies, but I have forgotten the details.

Regards,

Jay_Diddy_B

[T3sl4co1l]

As a side note: I don't think I need that high frequency response for the transformer is required...up to 1MHz signal will do to test compensation loops.

Beautiful!  We have all the numbers we need to work with!

Suppose the transformer is 1:1 at 50 ohms.  It should be made of 50 ohm transmission line, in maximum electrical length 1/8 wave at 1MHz.  If velocity factor is 0.67, then 25 meters is the allowable length.

Which is a pretty damned good bit of wire!

It should be something like #34 AWG or finer magnet wire (cough*), twisted together.  This won't be quite as low as 50 ohms, but it'll be close enough.

Good cores:
Hi-mu toroids, especially nanocrystalline (obtain from Mouser or Elna Magnetics?)
Silicon steel, with fine laminations (under 10 thou), GOSS or better
Pot cores (hi-mu ferrite, e.g. 3E27 or better)

Making a core size choice isn't obvious.  You do have a known quantity of wire to wrap around the thing.  You can probably get on the order of 500 turns around a 5uH/t^2 core.  Which is 1.25H, which gives a -3dB cutoff at 6.4Hz.  Not too bad.  Actually half that, because you'll have a 50 ohm signal generator, acting in parallel with the 50 ohm secondary termination resistor.  So 3.2Hz.

You should be able to obtain higher inductivity (>20uH/t^2) in the fancier steels (NiFe, permalloy) and amorphous and nanocrystalline materials.  This would push the cutoff below 1Hz.

The winding should be bifilar, wound in layers (with tape between each), to prevent too much crosstalk from adjacent layers.  That will help avoid weird dips and peaks in the response.

*Checking... you may actually want much heavier wire, since 25m of #34 is 21 ohms.  And that's just for one winding!  #28 is about the smallest you can get away with (< 3dB insertion loss), and #25 would be ideal...

Needless to say, larger wire will invite much larger cores and windings.  A UR style core (the kind they make flyback transformers with) might be attractive, because of the wide cylindrical winding area available.

Needless to say, any excuse you can make to reduce the winding length will save on both resistance and HF response.

Tim

[2N3055]

AcHmed99,

I think you didn't quite understand how it's done..

You don't inject current into PSU output.. You take PSU, you break feedback loop and insert injection transformer in series with feedback loop.. So it gets amplified by error amplifier, not suppressed......

The injection is done at the output top of the Rdiv string through what is commonly refered to as an injection resistor 22 oHms or so.

Huh? The opamp has high rejection at low frequency so it will attenuate (by a factor of 40-100dB) the disturbance which is what is shown in Basso book and in the literature with FRAs. If the opamp is actually amplifying the disturbance at low frequency (DC) then your supply has problems.

I thank Jay_Diddy_B for explaining it, my English is failing me sometimes (most of the time  ).. I wanted to say that injected signal gets superimposed on top of normal feedback voltage (vector added to be more precise).. As Jay_Diddy_B explained, result is modulation of output, using feedback amp and pass transistors (let's presume linear PSU for a moment) as amp with unity gain..
 For audio enthusiasts A class amp 

[Floyo]

   

I think the specific core I used is a TDK RM-8 in H5C2 material, the data can be found in this pdf https://product.tdk.com/info/en/catalog/datasheets/ferrite_mz_tl_rm_en.pdf
the AL is pretty high. I don't know exactly how many turns it has, but going from the 18H primary inductance and the specced AL that comes to around 1000 turns. The secondary being about 26db down in the plot gives a pri/sec ratio of around 1/19.5 so that makes the secondary around 50 turns. Wire size is "really thin" and airgap is "0". I hope that helps a bit, I didn't arrive at this transformer by any maths, just experimentation, and luckily the 1000 turn primary was already done when I found the thing in my junk bin .

[2N3055]

The attenuation part..

What you said is perfect if you take low impedance signal generator and just modulate output.. Regulator "fights the change" and will negate your disturbance, and will suppress it as a part of regulation.

If you inject into the loop, you simply modulate  feedback signal , and by measuring frequency and phase response to that signal (minus DC of PSU) you are basically measuring it for PSU loop..

I apologize if there is misunderstanding...

[Jay_Diddy_B]

Hi group,

I have put together an LTspice simulation to illustrate how this works. By studying an op-amp circuit we easily determine if we have the correct answer.



In the model I have used an op-amp to amplify a 10V reference to 20V a gain of two. I have included a network analyzer in the model, connected between nodes A and B. The LTspice will measure the loop gain, in the time domain, the same way as if I used my HP3577A in the lab. There is no need for an isolation transformer, because in SPICE the voltage source is floating.

If we start at node B and go clockwise around the loop. R1 and R2 are a divider, divide by 2 or -6dB. It has an output impedance of R1 in parallel with R2, R1//R2, = 10k.

The divider resistance combined with R3, 40k, has a gain of 4x or +12db. The total gain in resulting from the resistors is -6dB + 12dB = 6dB.

There is a low pass filter caused by C2 and R3

F= 1/2 x Pi x 1600pF x 40,000 = 2.5kHz (-3dB)


There is a zero at the frequency caused by R3 and C1

F= 1/2 x Pi x 0.16uF x 40,000 = 25 Hz (+3dB)

The results from LTspice are:



If we look at the relative size of the signals at A and B we see:



The units are dBV, that is dB relative to 1V rms.

And if we look at them in the time domain, at one frequency, 316 Hz we see:



The vector sum of the signals A and B is always equal to the injected signal.

If this works for this example, it will work for more complex circuits too, where it is difficult to know the right answer.


I have attached the LTspice model.

The model could be expanded by adding the transformer model between the source Vac1 and points A and B.

Regards,


Jay_Diddy_B

[MagicSmoker]

AcHmed99,

I think you didn't quite understand how it's done..

You don't inject current into PSU output.. You take PSU, you break feedback loop and insert injection transformer in series with feedback loop.. So it gets amplified by error amplifier, not suppressed......

Yes! Thanks to you and Jay_Diddy_B for explaining this while I was off doing chargeable work!

[MatteoX]

Few references for electronic signal injection:

Galinski  "Use Op-Amp Injection for Bode Analysis"
http://www.edn.com/design/analog/4369185/Use-op-amp-injection-for-Bode-analysis


Keeney "Simple Signal Injection Aids Controol Loop Analysis"   
http://powerelectronics.com/power-electronics-systems/simple-signal-injector-aids-control-loop-analysis

[EmmanuelFaure]

What about using a linear optocoupler instead of a transformer?

[LeonV]

Any one know what the frequency response from an air core transformer would be?
Or what if you wind a cooper loop into another loop? Like copper wire wound into a toroidal shape, then a winding around that as if it was a toroidal core.

[T3sl4co1l]

Any one know what the frequency response from an air core transformer would be?

Awful.  I'd be impressed if you got bandwidth more than 10 times the center frequency.

Tim

[rx8pilot]

I have been using a Jensen VB-1BB isolation transformer used for analog video. Not perfect, but not bad assuming that 10hz is not a necessity. It does ok down to rather low frequencies though. I paid $20 on eBay. BNC 75 Ohms in/out

http://www.jensen-transformers.com/product/vb-1bb/

I tried to use it with the Keysight power application on my MSOX6000 but was quickly stopped by it's inability to deal with the noise in SMPS. The newer software on the 3k's and 4k's seems to be a lot better at filtering and allowing a LUT for the injection voltage to get a better SNR on a range of frequencies. I was able to get measurements, but manually stepping through the frequencies, injection levels, averaging, and math filtering. A bit slow, but I think I was able to get believable measurements. After getting some pricing on dedicated FRA's, I am determined to figure out how to get what I need on the scope. I will design about 4-5 SMPS's per year so I need this capability for a grand total of maybe a week per year as I work out the supply design.



Having a real VNA would be nice, but certainly not a must.