EEVblog #1104 - Omicron Labs Bode 100 Teardown

[EEVblog]

Teardown on the Omicron Labs Bode 100 Frequency Response Analyser / Vector Network Analyser

[jbb]

I guess some of the DG4xx switches are being used as precision mixers (and others used to switch filter banks etc.).

It could be very interesting to (carefully!) probe the outputs of the two DDS units. I suspect that they run at different frequencies to do the heterodyne mixing. Also keep an eye out for high speed comparators to turn the sine wave frequency reference into square waves to drive switches.

On the directional coupler front, I think your previous video showed a 50 Ohm series resistor with a differential amplifier across it. Maybe they’re doing electronic measurement because it would be very hard to make an RF coupler work at the lower frequency range.

[JS]

The reflection is easy to measure as they have access to the generators output previous to the 50Ω resistor, software is clear enough they are using that.

The second dds is likely to generate the IF for the mixer, what I haven't seen is the mixer, as said the only mixy thing there are the switches but filtering that is kind of a pain, as switching mixer generates lot of side bands, should do the math for that. They could run the second dds at the same freq of the first, 90° out of phase and take down to dc in phase and cuadrature to meassure the signal, but that would take too long to settle at each step. ADC is too slow to make wide bands conversions but they are likely to still be using a fixed freq and just switch the bw of the filter. Missing a 3rd dds to make for the cuadrature reference.

We have all we need to reverse engineer that transformer! Construction method, number of turns, impedance analysis, freq response... It's just matter of finding the right core material, with the impedance and number of turns permeability could be estimated... Would someone sell cores out of that material?

JS

[Wolfgang]

I tried to make a modest homebrew version of this and the best solution I found for injection transformers were current transformers for AC current measurement where I added some primary windings:

https://electronicprojectsforfun.wordpress.com/injection-transformers/

You have to be cautious with frequency ranges advertised; these are -3dB values, and phase shift at the band corners can be too much for meaningful measurements.
Another issue is the capacitance between primary and secondary. This could be way above 100pF due to the bifilar winding technique, maybe too much for some sensible circuits.

[Bud]

. Missing a 3rd dds to make for the cuadrature reference.

You can generate quadrature LO sequentially with a single DDS as it is done in the N2PK VNA.

[Bud]

what I haven't seen is the mixer, as said the only mixy thing there are the switches but filtering that is kind of a pain, as switching mixer generates lot of side bands, should do the math for that.

Filtering may not be needed, it depends on the architecture. VNAs are less demanding to filtering. Some, like the VNWA, dont have filtering at all. Moreover, it makes use of the aliases as  the operating principle per se and to extend the bandwidth.

[station240]

Although it doesn't answer your question on what the wideband transformer winding style is, this page does explain why it's done that way.
http://www.richieburnett.co.uk/temp/gdt/gdt1.html
(unfinished page)

I think the real trick is in the brown ferrite core itself, the colours denote frequency/saturation properties, I have heaps of rings but not any brown ones which I assume are RF ones.

[Bud]

Do not seem to be a RF core to me but rather a low freq material. The trick is in the bifilar coupled winding and the ransformer will work with no core at upper frequencies. The core kicks in at lower frequencies to increase the windings impedance. This is widely used in RF baluns to achieve wide frequency range.

[JS]

Hi, Ive done my homework and I'm back, but first.

Filtering may not be needed, it depends on the architecture. VNAs are less demanding to filtering. Some, like the VNWA, dont have filtering at all. Moreover, it makes use of the aliases as  the operating principle per se and to extend the bandwidth.

  Right, if the only thing you are interested is amplitude that's true, non linearities would be harder to pick this was as high freq harmonics will fall well beyond the ADC's bandwidth and it won't be able to determine a lot of factors. Anyway, even if filtering is done it will only catch LF harmonics and might get trickier as freq increases. I did used oversampling building something in the past, as long as you are only interested in the amplitude and missing a few narrow bands isn't a problem the only HF limiting factor is the sampling time of the sample and hold circuit. I mostly have an audio background and for HF stuff sometimes it seems more like a luggage than a background, it bites me every time.

Although it doesn't answer your question on what the wideband transformer winding style is, this page does explain why it's done that way.
http://www.richieburnett.co.uk/temp/gdt/gdt1.html
(unfinished page)

I think the real trick is in the brown ferrite core itself, the colours denote frequency/saturation properties, I have heaps of rings but not any brown ones which I assume are RF ones.

  Here is my homework, the thing seems to have 39 turns 40 turns, (I knew it couldn't be 39, my numbers are still with 39 but they should be pretty close anyway) if I know how to count, in the previous video we can see the impedance, 400Ω @ 300Hz means 210mH, that's AL = 140µH/N^2
  I estimated the size of the core, roughly a T184 core, height 18mm, outer diameter 47mm, inner diameter 25mm means a permeability of about 60000, seems like a lot but I was expecting something like that, small signal audio transformer use that kind of core permeability, low turn count low freq was asking for that, now let's look for a brown core with tens of thousands for µr... there are not that many materials in that range...
  I would bet it's nanoperm https://www.magnetec.de/en/nanopermr-products/, it's the closer one I found.
  Here are some products, http://www.feryster.com/polski/nanoperm.php?lang=en
  There's one with 40mm outside diameter, 25mm inside diameter and 15mm height, permeability of 75000, I gotchya!
  Oh s***t, it's blue! http://allegro.pl/rdzen-magnetec-m-083-rtn-40x25x15-nanoperm-i7430998721.html I don't know, I can't find the actual one, this looks like the closest, but still could work. I couldn't find a way to source the cores, always hard here in Argentina, if someone can source them would be cool to see some tests.

  Hf response would only depend on the parasitic inductance and capacitance, to keep inductance low they have twisted the wires, to get low capacitance low turns count and thick insulators, the tricky part is the low frequency. I do have some audio transformers I could use, they go up to tens of kHz but LF response is better than seen in the first video, having a few Henry inductance, after that a different transformer could jump in and make up for the rest of the way.

  @EEVblog pleeeeeeease grab some calippers and measure that core!!! 

JS

Edited...

[T3sl4co1l]

That's not ferrite and it's definitely not powdered iron, that's got to be nanocrystalline and nothing else.   Most likely Vacuumschmeltze, probably https://www.mouser.com/ProductDetail/Vacuumschmelze/T60006-L2040-W453?qs=sGAEpiMZZMs2JV%252bnT%2fvX8PvC43ppqs%252bksq4V5kp6Ay4%3d or similar.

  Oh s***t, it's blue! http://allegro.pl/rdzen-magnetec-m-083-rtn-40x25x15-nanoperm-i7430998721.html I don't know, I can't find the actual one, this looks like the closest, but still could work. I couldn't find a way to source the cores, always hard here in Argentina, if someone can source them would be cool to see some tests.

Never seen any in blue, I wonder if they got them as special order, or if there's another mfg I don't know about.  IIRC, it was just VAC and HMG (former Metglas) doing rapid-quench materials, but maybe there's new ones from China I don't know about?

Checking, I see very little on Ali Express, so probably not.

Hf response would only depend on the parasitic inductance and capacitance, to keep inductance low they have twisted the wires, to get low capacitance low turns count and thick insulators, the tricky part is the low frequency. I do have some audio transformers I could use, they go up to tens of kHz but LF response is better than seen in the first video, having a few Henry inductance, after that a different transformer could jump in and make up for the rest of the way.

HF response is easily calculated from the transmission line length.  It's a transmission line transformer, simple as that.  The twisted pair will have Zo ~ 100 ohms, so that for a step input, each port of the transmission line looks like 100 ohms.  That is, the equivalent circuit for short transients is:
pri start -- 100 ohms -- sec start
|                                      |
~open circuit       ~open circuit
|                                      |
pri end -- 100 ohms -- sec start

The open circuit is because the core gives the transmission line a very large common mode impedance, i.e., the two ports act as ideal ports, with no common mode connection (again, for short transients, but as it turns out, also for rather low frequencies, down to ~Hz).

After one transmission line delay, the start and end waves interfere with each other, and normal transformer action is had.

Note that this does NOT magically have extreme CMRR -- there's as little as 100 ohms, directly from primary to secondary (again, for short transients).  At frequencies well below the electrical length, it approximates as a capacitance from each end of primary, to the respective ends of the secondary.  (The exact capacitance can be calculated from line length and impedance.)

Likewise, leakage inductance is the LF equivalent of transmission line inductance, and can be calculated from length and impedance.

If the line is, say, 5m long (to take a ballpark guess), and vf ~ 0.8, then Cp = 208pF (total, so, say, 104pF where the 100 ohmses are indicated above), and LL = 1.67uH.

Tim

[EEVblog]

We have all we need to reverse engineer that transformer! Construction method, number of turns, impedance analysis, freq response... It's just matter of finding the right core material, with the impedance and number of turns permeability could be estimated... Would someone sell cores out of that material?

If you could sell a $100 unit with the same performance, you'd sell a truck load of them.
Would be interesting to try and duplicate their design and what performance it has with just some random core.

[EEVblog]

Although it doesn't answer your question on what the wideband transformer winding style is, this page does explain why it's done that way.
http://www.richieburnett.co.uk/temp/gdt/gdt1.html
(unfinished page)

Many are saying it's a bifilar winding, and I actually said that in the video as my first guess, but I edited it out because I don't think it is.
I thought bifilar implies same coil opposite direction and (usually?) non twisted?
This one is opposite pair and twisted, totally different thing to a bifilar coil winding IMO 

[T3sl4co1l]

I thought bifilar implies same coil opposite direction and (usually?) non twisted?
This one is opposite pair and twisted, totally different thing to a bifilar coil winding IMO 

I use "bifilar" and "twisted pair" interchangeably for winding purposes.

Hmm, I don't actually know if "bifilar" is, say, a brand name or something, and implies parallel wires glued together and laid flat.  In any case, the characteristics will be very much the same for either method (with the twisted pair being slightly worse just due to the twist taking up some velocity factor).

Tim

[twistedresistor]

Bifilar just means the wires are wound next to each other, regardless of the direction. The opposite would be sectional winding where the two windings are on opposite sides of the core, like seen in chokes for common mode line filters.

What intrigues me the most: When i spoke to the guys at Omicron they've told me that one challenge for the transformer was that it had to be low in capacitance. But the coupling capacitance in a bifilar winding is usually higher than a sectional wound inductor.

(EDIT: T3sl4co1l beat me to it. Bifilar is not a brand thing)

[tautech]

Bifilar just the wires are wound next to each other, regardless of the direction.

That's as I understand it too.
Twisted pair is different.

[twistedresistor]

A twisted pair has nothing to do with winding an inductor. You can wind a twisted pair around a core - that's what they did - then you got a bifilar wound inductor.

That's as I understand it too.
Twisted pair is different.

[T3sl4co1l]

What intrigues me the most: When i spoke to the guys at Omicron they've told me that one challenge for the transformer was that it had to be low in capacitance. But the coupling capacitance in a bifilar winding is usually higher than a sectional wound inductor.

Inseparable from the nature of the transformer -- you can make bank windings, but you royally screw the leakage inductance.  Basically making a very high impedance transmission line.  But with a lot of self-capacitance balled up in the sections, rather than shared, so the bandwidth is awful.

A typical bank wound CMC has Zo ~ 300 ohms and BW < 10MHz.  Zo isn't really that much higher, because of the self-capacitance.

This TLT will have BW out until TL effects cause peaks and notches; these are dampened when Zin and Zout are matched, i.e., 100 ohm source and load.  At 50 ohms, the peaking will be noticeable, and the cutoff will be a bit worse (about 1/2 bandwidth).

Incidentally, the rolloff will NOT be that silky smooth curve they showed.  You can see this in CMC datasheets, the ones that are bifilar wound (data chokes), there are multiple peaks and valleys in the diff mode response, because the measurement is made as a shorted transmission line.  That it has peaks, means your data will propagate safely all the way up there. (For common mode, that is.  For isolation mode, there are peaks.)

Tim

[Kleinstein]

. Missing a 3rd dds to make for the cuadrature reference.

You can generate quadrature LO sequentially with a single DDS as it is done in the N2PK VNA.

There is no need for a quadrature LO if the ADC is not seeing zero IF, but more like an IF in the audio band. The 2 nd DDS would than be the LO, slightly offset from the output frequency. The quadrature part is than at the second virtual IF done in software.  So to a good part this is very similar to the simple SDR receivers. It is kind of doing the quadrature at a different time though more continuous phase shift instead of switching between 2 cases.

As the LO is used to drive CMOS switches as mixers, there would be digital LO signal and thus no problem generating quadrature signals with 2 D flip-flops and running the DDS at twice the LO. Still the ADC seems to be single channel only - so this way is likely not used, though it might offer slightly lower noise.

I would guess the FPGA might have enough processing power to do the IF processing and thus much less data transferred via USB. Using an FTDI USB chip also points to a rather low data rate.

[JS]

We have all we need to reverse engineer that transformer! Construction method, number of turns, impedance analysis, freq response... It's just matter of finding the right core material, with the impedance and number of turns permeability could be estimated... Would someone sell cores out of that material?

If you could sell a $100 unit with the same performance, you'd sell a truck load of them.
Would be interesting to try and duplicate their design and what performance it has with just some random core.

I could certainly try, at least that would justify buying a decent freq gen for the lab. Could you measure the core size for me? First out would certainly be a mailbag for the EEV!

That's not ferrite and it's definitely not powdered iron, that's got to be nanocrystalline and nothing else.   Most likely Vacuumschmeltze, probably

That's for sure, the banner picture on the ones I posted had blue and brown cores, the ones I found for sale with the part numbers I was looking shown blue pictures but that's the only reference I got and I can't even read the web in that language!

...
HF response is easily calculated from the transmission line length.  It's a transmission line transformer, simple as that.  The twisted pair will have Zo ~ 100 ohms, so that for a step input, each port of the transmission line looks like 100 ohms.  That is, the equivalent circuit for short transients is:
pri start -- 100 ohms -- sec start
|                                      |
~open circuit       ~open circuit
|                                      |
pri end -- 100 ohms -- sec start

The open circuit is because the core gives the transmission line a very large common mode impedance, i.e., the two ports act as ideal ports, with no common mode connection (again, for short transients, but as it turns out, also for rather low frequencies, down to ~Hz).

After one transmission line delay, the start and end waves interfere with each other, and normal transformer action is had.

Note that this does NOT magically have extreme CMRR -- there's as little as 100 ohms, directly from primary to secondary (again, for short transients).  At frequencies well below the electrical length, it approximates as a capacitance from each end of primary, to the respective ends of the secondary.  (The exact capacitance can be calculated from line length and impedance.)

Likewise, leakage inductance is the LF equivalent of transmission line inductance, and can be calculated from length and impedance.

If the line is, say, 5m long (to take a ballpark guess), and vf ~ 0.8, then Cp = 208pF (total, so, say, 104pF where the 100 ohmses are indicated above), and LL = 1.67uH.

Tim

I would guess under 3m, 10MHz wavelength is 30m, about 20m inside the wire, so transmission line is not quite needed in this case, I think concentrated parameters should get to a reasonable guesstimate. Also, in the first video, HF impedance looks capacitive, which makes sense for me I guess. I didn't took the time to look for looking at HF as I was trying to get the core parameters.

Here is the plot
https://youtu.be/66s9easZKxU?t=35m42s
The plot shows 200Ω at 1MHz, that's about 800pF. That's in the ball park from 100kMHz to 7MHz. It looks too high to me but still being a lot of things around, which will be used with this transformer anyway, still looking pretty high... Phase is over 70º and parallel resistance looks like 2k so I don't know why such a high capactance, I could bet it's not from the winding. I've wound a few transformers and with a few hundred turns of enamel coated wire around a chunk of iron, trifilar wound, got about 5nF. This has shorter 40 turns of polymer insulated wire, not teflon, the solder melted a little of it...

Then a bit over 10MHz there is a resonation, that might be inductance of wires going to the transformer, after that it start to behaves as you see and there might be due to the wavelength getting closer to the winding length.

Bifilar just means the wires are wound next to each other, regardless of the direction. The opposite would be sectional winding where the two windings are on opposite sides of the core, like seen in chokes for common mode line filters.

What intrigues me the most: When i spoke to the guys at Omicron they've told me that one challenge for the transformer was that it had to be low in capacitance. But the coupling capacitance in a bifilar winding is usually higher than a sectional wound inductor.

(EDIT: T3sl4co1l beat me to it. Bifilar is not a brand thing)

The capacitance they needed to make low is probably the equivalent parallel capacitance of the secondary (refered to the secondary as it will act as a 2nd order LPF there together with the leakage inductance. I'm considering concentrated parameters, not transmission lines, but we are in a middle ground here at 10MHz. As said twisted wires gives the lower leakage inductance which is a very good thing for HF transformers.

. Missing a 3rd dds to make for the cuadrature reference.

You can generate quadrature LO sequentially with a single DDS as it is done in the N2PK VNA.

There is no need for a quadrature LO if the ADC is not seeing zero IF, but more like an IF in the audio band. The 2 nd DDS would than be the LO, slightly offset from the output frequency. The quadrature part is than at the second virtual IF done in software.  So to a good part this is very similar to the simple SDR receivers. It is kind of doing the quadrature at a different time though more continuous phase shift instead of switching between 2 cases.

As the LO is used to drive CMOS switches as mixers, there would be digital LO signal and thus no problem generating quadrature signals with 2 D flip-flops and running the DDS at twice the LO. Still the ADC seems to be single channel only - so this way is likely not used, though it might offer slightly lower noise.

I would guess the FPGA might have enough processing power to do the IF processing and thus much less data transferred via USB. Using an FTDI USB chip also points to a rather low data rate.

This post is getting long, as I write I receive more answers... ADC could still be multiplexed, time difference between corrected in DSP, as long as IF is low enough. I was expecting a balanced mixer making filtering easier, but we are in a digital world so digital it is, there was a 595 somewhere in there, so...

For the times it takes to the software to load a different measurement mode and the real time data plot, I would guess you are right about processing in the FPGA, if it would be just pushing data via USB there's nothing to load other than a different algorithm to ram, which I'd expect to be quicker load times, slower plotting.

JS

[Wolfgang]

I would say the big difference between bifilar and twisted bifilar is better coupling and a better controllable transmission line impedance for the twisted bifilar case.
This thing is definitely a transmission line transformer using a very high mu core.

Useful information about transmission line transformers and their construction can be found in the "Transmission Line Transformers" Bible by Jerry Sevick.

[Bud]

It is not a transmission line transformer. You cant excite opposite ends of a transmission line and call it TLT just because you used a transmission line for it. 

[T3sl4co1l]

It is not a transmission line transformer. You cant excite opposite ends of a transmission line and call it TLT just because you used a transmission line for it.

So then what is it?

Transmission lines are a superset of whatever you think "transformers" are; it's well worth learning, both for understanding their use, and creating their design.

Tim

[Bud]

of course Tim it worth learning, so feel free to pick up ANY book on TL and start reading transmission line theory.

[Wolfgang]

Who on earth can stop me doing exactly that ?
Whats wrong calling a transformer made from transmission lines a transmission line transformer, independent of the the circuit that it is embedded into ?
Confusing ...

[Floyo]

On the diy front there is some more info to be found here: https://www.eevblog.com/forum/projects/diy-injection-transformer-for-power-supply-control-loop-response-measurements/
There I attempted one version of an injection transformer, but this teardown made me try something more akin to the Omicron setup.

I took a magnetec M-012 Nanoperm core and wound around 50 turns of thin 50 ohm coax onto it, the "shield" is the primary, and the inner conductor the secondary. Attached are the measurements with the Analog discovery. Test were done with a 47 and 1K2 ohm load resistor on the secondary side. Drive impedance from the analog discovery wavegen was switched from 50 to "0" ohms, the latter increases the low frequency amplitude. Primary inductance is 72mH and pri-sec capacitance is 400pF. 

Results look quite ok for just slapping some turns on a core. Keep in mind that the power supply injection measurements are relative to the secondary, so absolute amplitude and phase of the transformer doesn't matter too much, if its decently well behaved.