• WARNING: Tube/Valve amplifiers use potentially LETHAL HIGH VOLTAGES.
    Building, troubleshooting and testing of these amplifiers should only be
    performed by someone who is thoroughly familiar with
    the safety precautions around high voltages.

DIY Extreme performance Output Transformers - AMSOL

This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I mentioned this idea half jokingly in another thread (powdered iron xfmr thread), but after thinking about it some more I've realized its virtually certain to give amazing bandwidth output transformers. With some assistance from a Litz wire manufacturer, DIY OTs would be a 20 minute effort to build, if that.
This would revolutionize DIY tube amplifiers. (probably can get audio to Mhz bandwidth too)

What I'm talking about is an Amorphous Metal Serve On Litz (AMSOL) transformer. A what?? OK, here goes:

Here's a technique that a home Diy-er could use to make an unconventional output xfmr with likely VERY good performance results and NO complicated design problems (like interleaving or sectioning, in fact, not even any winding required!). Start with a spool of Litz wire with maybe 20 strands of enamel insulated wire in it. (I see Litz wire on Ebay regularly) Unserved Litz is fine. (the serving on Litz is the fiber wrapping around the wire strands to hold them together and provide extra insulation)

Now you need a spool of approx. say 1/4 to 1/2 inch wide amorphous magnetic alloy tape. Most likely you will have to buy this new from the manufacturer.(Allied) (See below about avoiding all this hassle) This stuff is very thin. Be careful when handling this stuff, its got razor sharp edges on it. Now you need a rig that can spin the tape spool around the wire as the wire gets pulled thru. Essentially, one would be putting an amorphous film serve on the Litz wire. Can experiment with how dense a spiral layer gets put on by altering wire pull rate versus tape spool spin rate. But unlikely to need a layer thicker than half the radius of the Litz wire.

This process should be a piece of cake for any commercial Litz wire manufacturer, so hopefully one can eventually just buy this stuff off the shelf, ready made on a spool. (I'm going to call a Lits wire manufacturer Monday and discuss this. I think this is clearly in their interest to offer this kind of product.)

The final output xfmr. would be configured by connecting half of the Litz wire strands (the little wires in the Litz) in series for the primary and half the strands in parallel for the secondary. Should probably use triple or higher enameled wire strands to reduce capacitance between "turns", but at least leakage inductance will be near zilch. (its the product of leakage inductance and distributed capacitance that determins xfmr bandwidth) Another approach to lower distributed capacitance further would be to use a large central wire for the secondary, surrounded by individually insulated small wires in a twisted shell. Can optimize distributed capacitance that way.

The whole final length of the Litz assembly could be coiled up for compactness, but this won't have any effect on magnetic coupling unless wrapped around an additional magnetic core. The equivalent of this Litz technique is often used to make broadband RF transformers using ferrite beads strung on multiple wires.

You have to get enough inductance in the primary to handle the lowest frequency. While this technique does not get to take full advantage of a large number of turns to get inductance, it does take advantage of minimal magnetic path length around the wire circumference, which will provide a large boost to inductance.
Inductance = u*N*N*Area/(magn. path length)

My first quick analysis led me to believe their would still be an overall disadvantage for getting the same primary inductance as a conventional E core design achieves using the same amount of magnetic material (volume or weight). However, on closer inspection, I have found the two are equivalent.

Here's a quick calc. to illustrate. Lets say we are talking about a 1000 turn primary transformer on E core with a 10 to 1 turns ratio. The Litz version gets 10 wires for primary, and 10 for secondary. Assuming the E core secondary uses about the same winding space as the primary, this would be equivalent to 2000 turns on the E core. So, magnetic path lengths in each case are roughly scaled as 2000 for E and 20 for Litz. Now looking at turns, the inductance is proportional to turns squared, N*N. So 1,000,000 factor for E (primary turns) and 100 factor for Litz. We seemingly come up a factor of 100 short for the Litz inductance:

(20/2000)*(1,000,000/100)* A/A'= 100

Whats missing here, however, is the area factors "A" and "A' " are not the same in these two cases for the same volume of material. With magnetic path length reduced by a factor of 100 we are free to increase one dimension of the area factor by 100 to get the same volume of material. This means the Litz wire length gets increased from the length of one turn on the E core winding to 100 times that length.

Interestingly, this ends up with the same primary wire length and resistance as the E core version, since 100 lengths of 10 wires in the Litz is the same as 1000 turns on the E core. So both the Litz and the E core are identical in inductance and resistance for the same weight of core. But the Litz version has far far better control of leakage inductance.

Since one is likely to coil the Litz version up for convenience sake anyway, a further improvement is possible yet. By coiling the Litz on a large toroid core, say an old 20 amp variac core, the magnetic material area "A" can be doubled, tripled or more. One can either go for extended low frequency response this way, or one can shorten the Litz wire to get the same design inductance, for a cost saving on the Litz/amorphous metal purchase.

So, how much work does the DIYer have to do to make an output transformer with this idea? If the pre-assembled Litz wire/amorphous metal serve is available off the shelf on a spool, you just have to figure the length needed to get your required design primary inductance and order a spool of that amount. You will want the correct number of wire strands in the Litz for the design turns ratio plus any other windings. With extra strands, one can make a flexible, variable impedance with taps xfmr., or go for additional windings like partial cathode feedback or isolated screen grid UL. Actual construction consists of soldering 10 wires (or so) together for a 10 to 1 turns ratio xfmr. and optionally coiling up the Litz (before soldering!) on a variac core. 20 minutes, my guess.

I doubt I will ever build another conventional E core or toroidal output xfmr., this simply blows them all away!

Hi Dhaen,

The primary voltages are definitely a concern. However, to minimise distributed capacitance, the primary turns would be triple or higher enamel coated, which should solve this problem.
I recall the McIntosh bifilar xfmrs had double or something enamel coated wires, and they faced the same problem. One could use a nylon or teflon sub serve layer around the primary wires if needed. The extra insulation causes the magnetic path length to increase however, so one will be trading off low leakage inductance for low distributed capacitance. Bandwidth probably staying the same with optimised construction. This is actually the same issue faced by RF broadband xfmrs. for their transmission line impedance. The length of Litz wire assembly required for low frequency primary inductance would increase however with the longer magnetic path length, so would increase cost obviously.

Another note, when optionally adding a variac core or whatever to increase magnetic material area, I should have mentioned that this does not directly add to the magnetic area figure for the amorphous layer, since the core has a much longer magnetic path length than the amorphous layer. Probably can about double the total effective magnetic area with a variac core added, which makes sense since we are just adding back the equivalent E core model.

Also, I have found an error in my calculation of the magnetic path length based on number of turns, so I have to do some more figuring on that yet. My gut feeling is that the two, E core or Lits/amorphous serve, will still come out comparable for material volume required. Working on it.

Super wide bandwidth Output trans

If you succed in producing one of your Litz super transformers, the world will be your oyster and you will become very rich.

I have spent 40 years on the holy grail of transformer designing in the real world of commercialism. And still have not succeded in DC to light!
I actually built a couple versions of this many years ago, I took a 25ft coil of copper tubing and ran a 25 conductor cable through it, hooked the wires up into one loop and hooked the end of the coper tubing up to the speaker.

I actually got sound out of it! The inductance was way too low so the LF response was terrible.

I then tried adding some iron wires, this worked better, it was getting into the midrange. I have no idea what the composition of that iron, it was just some soft iron wire I found cheap at the hardware store.

It did look kind of interesting though, the amp was sitting inside the OPT, very "futurustic" looking.

Thats about where I left it, the things are still on the shelf in the garrage.

The magnetic tape version has some intersesting possibilities. Just one problem, COST. Last time I checked a spool of that tape was very expensive, have you found an inexpensive sourece?

John S.
hey don,

i have been told by a friend that the small cobalt amorphous toroids are just what you need, they can be easily un-encapsulated and you are left with a roll of 1 mil amorphous tape.

the problem i see is the amorphous i have played with is amazingly brittle and there is no way that wil will serve around wire unless the bundle is quite thick to keep the bend radius above where it "snaps" Plus once served i would also expect "rolling" it up would cause the serving to break due to its inability to stretch.

maybe amorphous wire?

you could try to get the material annealed to soften it, but then how would you re-anneal it to get back the desired characteristics.


I made a mistake in my earlier calculation of self inductance of the Lits version for the same amount of magnetic material volume. The magnetic path length only reduces as the square root of # of turns, not linearly. So going from 1000 turns to 10 turns only drops path length by 10. There is a small compensating factor of around 2 due to geometry change from a square winding window to a tightly wrapped circular serve, but this still leaves one with only 1/50 of the inductance of the equivalent E core design and 1/10 the voltage rating at the low freq. end.

The inductance is generally the more limiting factor in OTs. Thus, pretty much ruling out the practicality of this idea, unfortunately. If the wire diameter could be reduced by 1/sqrt(7) then things would fly. Looks like one needs super conducting wire to implement this design, silver wire just won't do it. A higher permeability amorphous material would also help. One could also contemplate putting more turns in the Litz wire, but this just ends up heading back to the original E core design wound with a 1000 wire Lits of one turn. Oh well. Back to the drawing board I guess.

The idea still could be used in a sort of trade-off design to still get high performance at possibly quite moderate cost. The low leakage inductance is only important at the high frequency end of the audio band, and one doesn't need the large self inductance figure there (the low freq. end of the band needs it).

So instead of trying to put enough amorphous tape on to satisfy the low end, one would just put enough amorphous tape on the Lits to satisfy the high frequency end of the audio band. (probably just need a single wrap of amorphous tape, should be cheap too!)

Then wind the Litz on an E core (10 turns in the Litz and 100 turns of the Lits on the E core) to satisfy the self inductance requirement for the low frequency end of the audio band. The result would provide the original high performance across the whole band. So amorphous metal serve on Litz should still be a useful product.

(I like the transforming speaker cable idea, too bad it needs superconducting wire to do it. But, maybe one of these days ......)

Don :)
Hi Dave,

You may have a show stopper there for using amorphous metal. I just tried bending a piece out of one of those security tags and it cracked. Oh well.

But out of the flames a new Phoenix arises! Magnetics Inc. uses 1/2 mil or thinner permalloy tape for cores. Some of the special alloys like Supermendur have really high permeability, but are mechancically sensitive so may not work for bending around wire. I happen to have a small spool of either 1/2 mil or 1 mil permalloy, 1/2 inch wide tape around. Have to go find it. So will give it a try to see how it works on Litz. Probably will have to go to 1/4 mil by 1/4 inch tape though for practical application. The permalloy stuff is expensive too, but if only a thin layer is needed on the Litz for the hybrid design, it may be within reason.

Light at the end of the tunnel is not an oncoming train!

Hi Peter,
The permalloy wire might work fine for a SE xfmr where one needs a little air gap. The problem for a normal xfmr is that the magnetic flux needs to make a loop around the wire so has to jump the gap from turn to turn along the length of wire. So would lower effective permeability of permalloy covering.

AV8R, "spent 40 years on the holy grail of transformer designing"
"And still have not succeded in DC to light!"

The solution was published in Wireless World a while back, and I've seen it publihed elsewhere too. Well, near DC to microwave and no isolation. Here it is:

For transformation from Zi to Zo=n*n*Zi one uses n equal length transmission lines of n*Zi characteristic impedance. The Zi end of the lines all get connected in parallel and the Zo ends of the lines all get connected in series.
Common mode inductors are required on the outside of the lines to handle the signal swing at the high impedance ends. This can be done using a single large E core with the trans. lines all wound on it with the outer lines having more turns with linearly decreasing turns as one approaches the center lines. (ie. the common mode voltages must all show up as same volts per turn on the core) Ferrite beads get put on the lines also as HF common mode inductors too. Equal length lines are required so all signals traverse in the same time. No round the loop transmission is required as in some simpler broadband xfmrs.

Hooray!!!!!!!!!!!!! Party Time!!!!!!!!!

I think I have solved the original Litz problem! But don't quote me on this till I have a chance to sober up and re-check my calculations again. The problem was due to not being able to get the Litz wire diameter down enough (for shorter magnetic path length) to get the required self inductance of the primary. But I believe there IS a way. We start with a Litz wire containing only two wires. For the 10 to 1 turns ratio example, we cut one of the wires into 10 equal length sections and bring the ends out for access. Now coil the whole thing up as 10 turns so that the access wires all come out conveniently at the same position around the loop. The 10 sections get wired in parallel. (Of course, practically speaking, we would use 10 sections of Lits to do this)

The reduced Litz wire diameter now makes it possible to get the same self inductance on the primary winding, as in the E core model, using the same amount of magnetic material volume. Only problem is that the Litz is so small in diameter now that any wrapping tape will have to be ultra thin and flexible. This leaves out amorphous. But permalloy is still possible I think. Should be possible to reduce distributed capacitance a little better this way two. Another approach might be to use ferrite beads strung on the Litz instead of permalloy tape, but it will take an awful lot of them.


:) :) :) :) :) :) :) :) :) :) :) :)
Hi smoking-amp,

You got me thinking now:) Obviously i dont know as much as you do about magnetics...

I`m just throwing out some ideas :)... Maybe i learn something:)

Just a thought...

A sandwich of a very thin mylar insulated copperfoiltape and permalloy tape... the permalloy tape slightly wider than the copperfoiltape.

A sandwich of - permalloy-primary-secondary-permalloy- ten of this sandwich stacked on top of eachother(10:1) primarys in series ,secondarys in parrallell. Coil this stack to a flat round cake, and bridge the gap with a sheet of permalloy on the flat sides.

Will this work at all?
The permalloy will be verrrry close to the copper, and it will have the same distance to the windings over the whole length and width.
The capasitance might be a problem?

:) :) :)
Hi Peter,

Likely would need to do a little better completion of the magnetic path than just close spacing of the permalloy sheets. Might be able to continuos spot weld the edges of the permalloy.

The distributed capacitance is definately a major concern in the Litz or analogous designs due to the continuous proximity of primary/secondary and permalloy. It affects bandwidth and would sum up to a capacitance across the primary for the tubes to drive. For this reason I suspect a teflon or foam insulator between the wires may be required with some thickness. I'm thinking maybe two twisted wires of typical teflon insulated wire. But maybe just something like twisted kynar wire wrap wire would work. Have to calculate capacitance per foot or measure a sample. I'm also wondering if there might already be a product out there that fits this design, might be called shielded twisted pair, but has to use permalloy shielding.

The smaller the overall Litz wire diameter (due to thinner dielectric), the shorter the magnetic path length, so the shorter the length of Litz assembly required to reach a given primary inductance. Looks like the resultant capacitance of the primary may then be a constant no matter what insulation thickness is used, although thicker permalloy coating allows shorter length independently.

Another technique to consider is the possiblility to plate permalloy onto copper. This was done in the past to make plated wire memory for computers. In this case one would probably use a coax configuration for the primary/secondary with plating on the outer surface. But coax has a much higher capacitance per foot than twisted pair, so is at a disadvantage here.

Hi Don,

Continous spotwelding, i can`t see how that could be done with the insulated copperfoil inside, the insulation would melt. And how would that affect the permeability of the permalloy. i guess annealing would fix that, and melt the insulation completely.(maybe kapton insulation could withstand the high temp?)
What about making the coil perfectly flat to accept the sheet with no gap, and use a thicker permalloy tape?

Yes, that would be like continous interleaving, no (low) leakage inductance, but high capacitance.

Electrostatic shielding between primary and secondary? Hmm... Increases the leakage inductance:bawling:

What about foam just between primary an secondary(Cu foil coil)? Shouldn`t affect the primary inductance too much?

Pimary inductance, parasitic capacitance, leakage inductance, a lot of tradeoffs here.

I am seeing only zero turns with this technique, even it it were remotely physically possible. Without the wires wrapping around the core, no magnetic flux is induced. No flux induced = no magnet coupling. However, if you were able to anneal the permalloy after wrapping, the wire would be well shielded from external magnetic fields.

Hi Jlsem,

I don't follow your reasoning. The topology is really no different than an ordinary xfmr. A turn in this case is the length of wire thru the outer covering. The magnetic core is the cylindrical permalloy covering. It just looks strange since the core is stretched out dramatically in length. But can visualize this as a geometrically distorted toroid with one turn of wire thru it.

Or maybe the problem you are seeing is the lack of multiple turns to perform the N to 1 turns ratio? A simple way of seeing the impedance transform is to consider each of the 10 sections as a 1 to 1 xfmr between the two wires passing thru that core section. The impedance transform is done by connecting the secondaries in parallel and the primaries in series. This is a pretty well known technique.

Did I miss something? Edit: OH!, maybe you are talking about one of Peter's configurations.

This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.