EEVblog #1104 - Omicron Labs Bode 100 Teardown

[precaud]

This thread has become more about transformers than the Bode 100, so I'll continue in that vein.

I've been meaning to make a 2:1 stepdown version of this, optimized not for control loop response but for making low-impedance measurements. The major difference is that the xfmr is working into a load defined as (1 Ohm + Zdut). For a 1:1 xfmr this results in a very restricted bandwidth, easy to saturate at low freqs, and other problems. I finally got around to making it Monday.

Prior to this, I've only put together 1:1 xfmrs so I wasn't sure how best to do different winding ratios. I was certain that I didn't want the windings in separate layers; reasonably certain that I wanted to use twisted-pair conductors for extended HF bandwidth. But beyond that, how best to arrange the remaining windings was not clear. What I ended up doing is first winding the 40-ish turns with the conductors in twisted-pair config, and then adding the additional primary turns as an outer layer. (Maybe they should have been the inner layer?) I have a spool of 22 ga. magnet wire on hand so I used that. The result was 88 turns for the primary, and 42 for the secondary, wound on a VAC T60006-L2030-W423 core:
https://www.mouser.com/ProductDetail/Vacuumschmelze/T60006-L2030-W423?qs=sGAEpiMZZMs2JV%252bnT%2fvX8PvC43ppqs%252bkNLW11zNOYmc%3d

The point of the stepdown is to maintain wider bandwidth with less loss while driving low-impedance loads. My goal is to get a -6dB bandwidth of 1MHz into 1 Ohm. We'll see how that worked out...

The plot shows the freq response of the xfmr terminated in 50, 10, and 1 Ohms. (The messiness above 10MHz is due to the long xfmr leads and unshielded clips used in the test setup. This will clean up in the final install. I'm not focused on the response up there... yet.)

For comparison, I've included the same plot for one of the better 1:1 50 Ohm xfmrs I've used in this application, a North Hills 0016PA 50 Ohm isolation xfmr.

Comparing the traces, you can see the homemade xfmr does indeed do a better job of driving the lower loads with less loss and wider bandwidth, at the cost of about -7dB stepdown loss. Compared to the 0016PA, -3dB bandwidth is about doubled into 10 and 1 Ohm loads; 240kHz into 1 Ohm, -6dB at just shy of 500kHz. This is headed in the right direction, but still short of the 1MHz goal.

On the low end, response is excellent, less than 1dB down at 1 Hz, and doesn't change appreciably with load.

So, two questions. It appears I'll need to use a 3:1 winding ratio to get the desired bandwidth into 1 Ohm, yes?
What's the best way to interleave/layer the windings for maximum bandwidth?

Any input would be appreciated.

[capt bullshot]

I'd try to wind it tri- or quadfilar and then connect two or three of the resulting windings in series for the primary.

Some time ago, I made a silly attempt to make a multi-core "coaxial" transformer - put five wires into a shield, wind a transformer from that and connect the wires in series. Don't have pictures and no spare time to provide some at the moment, but afair the result was impressive in terms of stray inductance but _not_ too impressive in terms of frequency response into higher impedances (I tested it as a step-up transformer, not a step-down). Would be interesting to test that thing into low impedance again.

Edit _not_

[precaud]

I'd try to wind it tri- or quadfilar and then connect two or three of the resulting windings in series for the primary.

I had thought of doing that, but nixed it and can't remember why... maybe cuz the twisted configs maintain more consistent impedance over a wider BW. I'd also considered a braid but that's tedious...  40 years ago my GF braided two 10-foot 3-cond cables using 16 ga... no complaints, bless her heart...  she said it was smaller than the hair bundles she normally worked with 

[T3sl4co1l]

To get infinite bandwidth on a ratio, you need a Guanella TLT.

The principle is using TLs in series-parallel combinations at each port, so that the signal sent down each TL is exactly the right fraction of the total, in exactly the right impedance, so that the waves are put back together exactly in phase, no delay shift.  Ultimate bandwidth is limited by the cross section of the TLs, the routing in the port area, and crosstalk between TLs (because, of course, no TL is ideal with infinite common mode impedance).

To get maximum (note, it will always be finite) bandwidth on an isolating transformer, you basically want 1:1.  Other ratios are effective, but you will have some compromise between a Guanella style construction, a Ruthroff type (where fewer TLs are used and the delays are unmatched), and a conventional (not-TL-driven) design.

For example, consider the 1:N transmission line transformer, where a signal line is routed along a reference plane multiple times.  The plane forms a slotted loop (it might be a cylindrical shell made of copper sheet or foil, or a planar circle on a PCB, say), while the signal forms a spiral or helix (or equivalent).  The slot in the plane, in turn, connects to a parallel plate transmission line.

A cylindrical example looks like thus:



The "signal" line here is 1/8" (3.1mm) copper tubing (with sleeving), bifilar, inside a sheet metal surround.  The transmission line impedance I would guess is on the order of 50 ohms (per signal line).  The output line should be around 9 ohms.

The delay through the signal line is N turn lengths, while the delay through the plane is just one turn length.  If a step change in voltage is applied to the plane, that voltage is induced on the signal line, simultaneously, at each point the line crosses the plane split.  When this propagates out to the terminals, the result is a series of N steps (of complementary polarity from each end of the line, with respect to the reference plane).

The signal is always with respect to a reference plane, so it's balanced, and each end of the winding has that same reference.  This is good when you need a balanced signal, but when you need isolation, you need imbalance -- the delay of the signal line corresponds to the equivalent isolation capacitance and leakage inductance in the transformer.

Any construction that is similar to this, will get you pretty good bandwidth.

You can apply the same theory to conventional (multilayer) windings, as well: in that case, each layer of turns is the reference plane for the layers nearby, and so on, so the characteristic impedance of a multilayer (per segment) winding can become quite high indeed (to first approximation, the impedance is the impedance of how many turns tall+wide it is).  Predicting what happens in cutoff, in windings like this, who knows -- but estimating where the first cutoff is, isn't so bad!

Incidentally, the characteristic impedance of a common mode choke like this,
https://www.digikey.com/product-detail/en/taiyo-yuden/TLF9UA102W0R8K1/587-2788-ND/2573875
is around 600 ohms.  There's a few layers in there, and the two windings are rather poorly coupled to each other!  The resulting bandwidth is even lower than you'd expect given the wire length (which probably isn't much on a small one like this, though I haven't counted).  The isolation capacitance is quite low, though.

Tim

[precaud]

Thanks Tim. Much of what you wrote is over my head, but what I did grok gave me some ideas. Unfortunately work is interfering with my play... I'll be back in a day or two.

[precaud]

I made a 3:1 version, using a quadfilar twisted bundle. I didn't count turns, but it was a couple less due to the larger diameter of the bundle. It was actually easier to wind, but required two layers. As it should be, stepdown loss was about -10dB.

Unfortunately, freq response into 1 and 10 Ohms was little changed compared to the 2:1 version. This is not as it should be.

Thinking that perhaps the source impedance needed to be lower to drive it, I tried a buffer amp with 25 Ohm output Z, but no difference.

I then looked at the input impedance with the HP 4275A and didn't see anything that would explain the rolloffs. Output Z was about 5 Ohms as it should be.

Interwinding capacitance measured 220pF on one set and 330pF on the other.

None of this explains the HF response rolloff into 1 Ohm.

Is it a property of the core material?

[T3sl4co1l]

You have about the right impedance for the primary (one strand surrounded by three others is about 50 ohms), but you don't have the right impedance for the secondary (which is 50 ohms again, not 5.5 ohms).

Try three twisted pairs, primaries wired in series, secondaries wired in parallel.  That gets you closer to having a reference plane's worth of secondary.  Probably ideal would be three star quad cables wired up (which is a bit low for the primary (~25 ohms), but about the same amount high for the secondary), but that's more fiddly...

Core has essentially nothing to do with the HF cutoff.  Incidentally, you don't need to put on so much wire to see what's going on here, as long it's long enough to still capture the HF cutoff in the range of your equipment.

Tim

[precaud]

You have about the right impedance for the primary (one strand surrounded by three others is about 50 ohms), but you don't have the right impedance for the secondary (which is 50 ohms again, not 5.5 ohms).

It is my understanding that insertion loss scales with the turns ratio (i.e. 3:1 = 9.5dB ), and impedance scales with the turns ratio squared. Yes? So the output Z, with 50 Ohm on the primary, is about right at 5.6 Ohms. And the input Z is pretty close to 9 X (2ndary load R) in the passband.

Try three twisted pairs, primaries wired in series, secondaries wired in parallel.  That gets you closer to having a reference plane's worth of secondary.

That would give 9:1 turns ratio, and 81x Z multiplier!

If I'm reading the "reference plane" bit correctly, it suggests that a single layer of interleaved windings is desirable to maintain HF BW. Yes? I could drop down to a smaller wire gauge and achieve that.

  Probably ideal would be three star quad cables wired up (which is a bit low for the primary (~25 ohms), but about the same amount high for the secondary), but that's more fiddly...

Yeah, that sort of layup is for machines to do...

Core has essentially nothing to do with the HF cutoff.  Incidentally, you don't need to put on so much wire to see what's going on here, as long it's long enough to still capture the HF cutoff in the range of your equipment.

Thanks for confirming that. I see your point re: # of turns may be excessive. Maybe I'll try removing the second layer and rewiring before junking this and doing one with smaller wire.

Thanks fot your input!   

[precaud]

BTW, for comparative purposes, would someone who has or has built one of the NVT's show plots of (or measure) the -3dB bandwidth with 50 Ohm source into 1 Ohm and 5 Ohms loads?

Picotest specs their J2100A and J2101A injection xfmrs (both 1:1 designs) into 5 Ohms. I inquired of them yesterday about their BW into 1 Ohm, they replied that insertion loss would be greater, low-freq -3dB point would be lower by 1/5th, and high-freq -3dB point would be "substantially the same" as it is into 5 Ohms. Well I don't believe that latter for a second. No 1:1 xfmr I've measured behaved that way. They also offered to measure it if I needed it, which I'll ask them to do, and will post the results.

[Jay_Diddy_B]

BTW, for comparative purposes, would someone who has or has built one of the NVT's show plots of (or measure) the -3dB bandwidth with 50 Ohm source into 1 Ohm and 5 Ohms loads?

Picotest specs their J2100A and J2101A injection xfmrs (both 1:1 designs) into 5 Ohms. I inquired of them yesterday about their BW into 1 Ohm, they replied that insertion loss would be greater, low-freq -3dB point would be lower by 1/5th, and high-freq -3dB point would be "substantially the same" as it is into 5 Ohms. Well I don't believe that latter for a second. No 1:1 xfmr I've measured behaved that way. They also offered to measure it if I needed it, which I'll ask them to do, and will post the results.

Precaud,

We can look at the effect of the load resistor on the low frequency -3dB by modelling.

I have set the magnetizing inductance of the transformer to be 100mH in this example. The NVT transformers may have higher magnetizing inductance.

Model

The Load is stepped in a 10, 5, 3 sequence from 100 Ohms to 3 Ohms.
The amplitude is lower with a lower load resistance, because of the 50 Ohm source impedance.
The -3dB point can be found easily when the phase shift is 45 degrees.



Modelling Results



You can see how the -3dB point is at a lower frequency for lower values of load resistor. But also the amplitude in the flat portion is also reduced.

Simplified Model

Since the transformer is 1:1 and the leakage inductance has little effect on the LF response the model can be simplified to:



This has the same transfer function as the model above, so the results are not shown.

Analytical Result


The BW can be shown to be:




Similar analysis can be done for other turns ratios.

Regards,

Jay_Diddy_B

[precaud]

Thanks Jay. My modeling software is old (pre-Spice) and on a computer that is down at the moment. I'm more concerned about the HF cutoff than the low-freq one.
In that model, L3 is the primary determinant impacting the high-freq cutoff. Where does that come from? How is it determined?

[Jay_Diddy_B]

Thanks Jay. My modeling software is old (pre-Spice) and on a computer that is down at the moment. I'm more concerned about the HF cutoff than the low-freq one.
In that model, L3 is the primary determinant impacting the high-freq cutoff. Where does that come from? How is it determined?

The leakage inductance, shown as L3 above, and the winding capacitance has an impact on the high frequency response.

The leakage inductance is measured by shorting one of the windings and measuring the inductance of the other winding.

Model for Leakage Inductance



The magnetizing branch is ignored, because it should not impact the high frequency response.

Modelling Results



When I measured HF bandwidth of a common mode choke in this message:

https://www.eevblog.com/forum/blog/eevblog-1104-omicron-labs-bode-100-teardown/msg2196759/#msg2196759

It did not match the bandwidth predicted by the leakage inductance model, probably because of capacitance.

Leakage inductance is minimized by bifilar winding.

What is your application that requires you to be concerned about the HF response?


Regards,

Jay_Diddy_B

[precaud]

OK, I thought it might be the leakage inductance. I'll have a look at that.

Somewhere I have a more comprehensive transformer model, one developed and used by Deane Jensen before he passed on. I haven't been able to find it, though.

What is your application that requires you to be concerned about the HF response?

It's part of a setup to measure active vreg output impedance using a VNA/FRA. Right now, the setup is good up to 100kHz. At higher freqs, it's all about the distributed bulk capacitance and pcb trace series R and L, which is easier to measure passively. But there's a grey zone, typically in the 100kHz-500kHz region, where sometimes it needs to be powered for accurate results, and sometimes not. I'm trying to extend the bw to enable accurate measurement up into that region, hopefully to 1MHz. The xfmr is the limiting factor.

[capt bullshot]

Yes, stray inductance matters a lot at low load impedance and MHz frequencies.

This is the transformer I mentioned above, made of five wires pulled through a braid shield - the wires in series as the primary and the braid as the secondary. Load impedance is 2 Ohm (five paralleled 10 Ohm resistors).

The core is ferrite and doesn't have have as high an AL value as these nanocrystalline toroids, so the low -3dB point is rather high at 200Hz, but the high -3dB is at 4.8MHz

The primary magnetizing inductance is 24mH (@ 10kHz), stray inductance (secondary shorted) is 54uH, and the other way round it's 963uH and 1.28uH.

Edit: added the impedance plot (with 2 Ohm load impedance)

[precaud]

Interesting xfmr, capt. Could be very interesting with a different core material.

Leakage inductance on my 3:1 measures 680nH. X_L is 1 Ohm at 240kHz. Which is the -3dB point on the 1 Ohm load curve. So that is indeed the culprit. In order to get -6dB @ 1MHz, L_leakage needs to be 330nH.

Time to start stripping off turns...

[precaud]

I peeled off the second layer of turns, leaving 78 primary / 26 secondary. Pri inductance is still a generous 460mH. Leakage inductance is down to 470nH.

This resulted in a definite improvement into 1 Ohm: -3dB @ 315kHz, -6dB @ 630kHz. More to come (off) tomorrow.

[precaud]

After reading a refresher on the basics of transformer design (maybe should have done that first, eh...), I decided to be cautious and remove 5 more winds. I might have done more, but was concerned that the gaps created between windings would negatively impact core utilization, causing L_Pri and L_Sec to plummet.

This turned out to be a valid concern. The measured data for the now 63:21 xfmr are:
L_Pri  280mH
L_Sec 33mH
L_Leakage 395nH
C_Interwinding 132pF

So to get a 15% reduction in leakage inductance, pri and sec inductance went down by 40%. Wow. I can't continue this trend by removing more winds.

The resulting transfer response (output / input) is attached. Compared to the 78:26 version, insertion loss was (surprisingly) lower in the passband, and slightly more energy in the rolloff region, resulting in -3dB @ 265kHz, -6dB @ 540kHz. I have no explanation for the better efficiency at low freqs; the impedance profiles look the same as before. I checked all windings with a milliohmmeter and they are all identical.

So, while this version drives 1 Ohms with less loss than the last one, the -3dB bandwidth is in fact worse, and I'm no closer to the goal. L_Leak will have to be lower than previously calculated to get the useable bandwidth to 1MHz. Interwinding capacitance at 132pF appears not to be an issue (at 1MHz). Further lowering L_Leak is needed. So what's the next step?

One idea: by using a larger diameter wire, I could remove a few more winds and at the same time cover the entire core, restoring L_Pri and L_Sec. This should lower the C_Winding as well.

Any other ideas?

[capt bullshot]

Wire resistance would matter for insertion losses, your wire is shorter now - so less losses.
Otherwise, while experimenting with transformers, I found the upper frequency being a function of wire length (for twisted pair 1:1 construction).

[T3sl4co1l]

After reading a refresher on the basics of transformer design (maybe should have done that first, eh...), I decided to be cautious and remove 5 more winds. I might have done more, but was concerned that the gaps created between windings would negatively impact core utilization, causing L_Pri and L_Sec to plummet.

For a given core, winding area doesn't matter -- only number of turns.

Magnetizing inductance goes as N^2, while leakage goes as N (for a helical/cylindrical winding shape -- constant wire length per turn), so you get a better deal with more turns.  Assuming you can tolerate the higher leakage in the first place, of course.

Magnetizing inductance also goes as Ae, so a larger core (cross sectional diameter) also gives proportionally more length/turn, but squared more area, so you get a better deal with more area.  Basically, they're equivalent, down to a constant (which will depend on geometry, so your choice of core matters).

If you have a hard limit on leakage and therefore turn length, and Zo is already correct, then the only remaining option is to improve the core.  If nanocrystalline still isn't adequate.... have you considered some really nice op-amps instead?

Note that stacking cores (when applicable -- toroids basically) isn't as good a deal.  The turn length goes up faster (compared to getting a bigger core), because the cross section goes rectangular.  A low-aspect (square or round) cross section is best.

One idea: by using a larger diameter wire, I could remove a few more winds and at the same time cover the entire core, restoring L_Pri and L_Sec. This should lower the C_Winding as well.

Amazingly, wire size doesn't matter, it's the relative geometry that matters -- this determines the transmission line impedance.

A pair of 0.25mm dia. wires, separated by 0.5mm, has identical impedance to a pair of 2.5m pipes separated by 5m (heh, assuming they're far enough above ground not to worry about that).

You might be using too much length to get away with, say, fine magnet wire, or wirewrap wire, in terms of insertion loss -- resistance is independent of TL impedance.  But as long as that's adequate, you don't need any larger wire than that.

Speaking of magnet wire -- the enamel is quite thin, so if you can tolerate the lower voltage rating, you'll get a much lower impedance, the wires are closer together.  Typical twisted pair / bifilar wire is in the 50 ohm range, quite a lot lower than, say, CAT5 pair.  That's half the leakage inductance!

Any other ideas?

Have you tried this yet?:

Try three twisted pairs, primaries wired in series, secondaries wired in parallel.

You noted earlier,

That would give 9:1 turns ratio, and 81x Z multiplier!

No (there aren't even nine wires in that construction!).  Just three secondary turns/segments in series, and three primaries in parallel.  3:1 current and 1:3 voltage. 

Tim

[precaud]

For a given core, winding area doesn't matter -- only number of turns.

Hmmm... I didn't see such a large L loss when removing turns from the 2nd layer. Now that it's down to 1 layer, I assumed it had to do with core coverage. But maybe it was just proportional turns.

The other reference says, to lower L_leakage:
"Minimize leakage inductance by using a window shape that maximizes winding breadth, and/or by interleaving the windings. And reduce physical separation between windings."

The toroid maximizes the shape. But after removing turns and spreading them symmetrically over the core, I've created gaps between the windings.

Amazingly, wire size doesn't matter, it's the relative geometry that matters -- this determines the transmission line impedance.

At low freqs, true. But not at high freqs, where things like skin effect and proximity matter, yes?

You might be using too much length to get away with, say, fine magnet wire, or wirewrap wire, in terms of insertion loss -- resistance is independent of TL impedance.  But as long as that's adequate, you don't need any larger wire than that.

At this point, winding length isn't that long, 3-4 ft. The 1:1 xfmrs I've been using have much higher Rdc in the windings, so I think I could go with a smaller wire. I have some 26 ga I could try for the next iteration. It would conform to the core closer than the 22 ga bundle does.

Speaking of magnet wire -- the enamel is quite thin, so if you can tolerate the lower voltage rating, you'll get a much lower impedance, the wires are closer together.  Typical twisted pair / bifilar wire is in the 50 ohm range, quite a lot lower than, say, CAT5 pair.  That's half the leakage inductance!

Yes, I'm already using magnet wire for that reason. Plus, less capacitance and more winds per layer.

Have you tried this yet?
Try three twisted pairs, primaries wired in series, secondaries wired in parallel.

Yeah, my previous understanding of it was in error, I was equating it with parallel impedances. I haven't done it yet, it's more difficult to wind than a single quadfilar bundle. Is there a benefit at high freqs to it? The other source advises against parallel windings: "Paralleling windings or wires within windings succeeds only if the expected division of high frequency current results in the smallest energy transfer."

[T3sl4co1l]

The other reference says, to lower L_leakage:
"Minimize leakage inductance by using a window shape that maximizes winding breadth, and/or by interleaving the windings. And reduce physical separation between windings."

That's just for conventional full layer windups.  You've already solved that problem by literally bringing the secondary along with.

Try it -- make another twisted pair(s), in free air, connect it up as you would for a 1:3, and measure the high frequency response.  You will find the same cutoff; although in this case (no core) you may find the low frequency cutoff is so high it's interfering with this measurement.

This is another potentially surprising fact: leakage doesn't depend on core.  That should be kind of obvious, actually, since it's leaking somewhere -- well, that's what it means, it's not in the windings, and if the windings are on top of each other, it must not be in the core either.   It's in the space between windings.  (A bank-wound or opposite-limbs winding of course will depend on core, but those also have a lot of leakage to begin with.  To put it another way: the windings are so far apart that, the space between them, of course includes some core.)

Amazingly, wire size doesn't matter, it's the relative geometry that matters -- this determines the transmission line impedance.

At low freqs, true. But not at high freqs, where things like skin effect and proximity matter, yes?

But those are loss factors, not reactance factors.  Hence, they factor into insertion loss, not characteristic impedance.

You're also maximizing performance with a TLT design, in that proximity effect is minimized because the windings are always together.

Yeah, my previous understanding of it was in error, I was equating it with parallel impedances. I haven't done it yet, it's more difficult to wind than a single quadfilar bundle. Is there a benefit at high freqs to it? The other source advises against parallel windings: "Paralleling windings or wires within windings succeeds only if the expected division of high frequency current results in the smallest energy transfer."

Again, that's for whole-layer or multi-layer-per-section conventional builds.  The wires are already parallel, and by that I mean the primary and secondary are.  Putting more TL segments in parallel only provides more area for image currents to flow, and therefore lower impedance, and therefore lower leakage.

I guess there are only so many ways I can describe this.  My reasoning is based on EM theory, and proven time and time again in the lab.

Tim

[precaud]

Please don't be offended, Tim. I'm just bouncing off of you my limited EM knowledge, and things I've read or seen online. Some of it is apparently BS, or not relevant. For instance, the bit about "core coverage" and "symmetrical placement of the windings" came from a guy on Youtube who was making RF baluns using mulit-filar windings, and said they were important.

I'll try your 3x twisted pair and report back. Thanks again for your help!

[precaud]

I put together a 3-strand twisted pair version as per Tim's recommendation. I used 26 ga. magnet wire, which is a lot easier to work with than the 22 ga. was. Twisted each pair to about 5 turns/inch and wound one full layer on the core. Here's how it measured:

L_Pri  330mH
L_Sec  37mH
L_Leakage 235nH
C_Interwinding 290pF

Compared to the last windup using 22 ga., L_Leakage is nearly halved, and C_Interwinding is more than doubled. See the attached plot to see how that plays out in the measured response.

Into 1 Ohm, transfer response is -3dB @ 360kHz, and -6dB @ 700kHz. This is the best result yet, by a little bit.

Also interesting is that the response into 10 Ohms is better behaved than it is into 50 Ohms. Once it is installed in a proper box with BNC connectors, it appears the HF response should extend cleanly up to 30MHz and beyond with those loads.

Tomorrow I'll evaluate the low freq extension and see if I can remove some turns, which would lower L_Leakage and C_Interwinding.

Thanks to Tim for your recommendation. You were spot on!   

[T3sl4co1l]

Bingo. You've got a secondary with characteristic impedance near 10 ohms, hence it performs best matched into that load.  Probably it's slightly better still at... 5-7 ohms maybe?

Tim

[precaud]

Yes, it would be better at 5.6-ish Ohms, that's the nominal output impedance with 50 Ohms on the primary.

The low freq response is flat to 10Hz and -3dB at 1Hz. Without modeling it, I'm leary to start removing turns to reduce L_Leak and C_Winding. If you go too far, you can't put turns back... So this may be as good as it gets.

So how do we reduce L_Leak and C_Winding further? Use a smaller core?