Audio Transformer Inductance

[CJ]

<...whether it's Bessel, Butterworth, peaked or overdamping is all dependent on the damping....>

don't forget that winding geometry can have an  influence on the curve,

for example, the LA2a reissue used a different winding geometry for the ha 100 x input copy,

instead of 2 pies per coil, they wound one pie.

much easier to wind this way, UTC had the dual winding thing down,

i do not know of anybody doing the side by side sections anymore,

so the old UTC HA100x has less leakage, which gives it a different curve,

 

[ruffrecords]

Just had an interesting response to a request for a special transformer from Edcor. I asked for:



Can you please give me a quote for a custom transformer to the following specification:

Turns ratio 12:1
Primary inductance 120 Henries
Power rating 1 watt.
Primary voltage rating 50V rms
Frequency response 10Hz to 40KHz +- 1dB
Transformer to be ungapped (no dc current)
Application is for headphones amplifier ac coupled from SRPP drive stage


And the reply I got today to my message (sent on 18th January) was:

Ian,

We don’t make this type of unit,  It takes a special core.

Phyllis Weston

 

[ruffrecords]

I have updated the document in the original post to include some corrections and additional data on Mouser transformers by NY Dave.

http://www.groupdiy.com/index.php?topic=47515.msg597880#msg597880

Cheers

Ian

 

[lassoharp]

Just had an interesting response to a request for a special transformer from Edcor. I asked for:



Can you please give me a quote for a custom transformer to the following specification:

Turns ratio 12:1
Primary inductance 120 Henries
Power rating 1 watt.
Primary voltage rating 50V rms
Frequency response 10Hz to 40KHz +- 1dB
Transformer to be ungapped (no dc current)
Application is for headphones amplifier ac coupled from SRPP drive stage


And the reply I got today to my message (sent on 18th January) was:

Ian,

We don’t make this type of unit,  It takes a special core.

Phyllis Weston

I'm not sure if I already asked you this - did you try approaching Sowter with same request?

 

[ruffrecords]

Just had an interesting response to a request for a special transformer from Edcor. I asked for:



Can you please give me a quote for a custom transformer to the following specification:

Turns ratio 12:1
Primary inductance 120 Henries
Power rating 1 watt.
Primary voltage rating 50V rms
Frequency response 10Hz to 40KHz +- 1dB
Transformer to be ungapped (no dc current)
Application is for headphones amplifier ac coupled from SRPP drive stage


And the reply I got today to my message (sent on 18th January) was:

Ian,

We don’t make this type of unit,  It takes a special core.

Phyllis Weston

I'm not sure if I already asked you this - did you try approaching Sowter with same request?

Sowter already make a transformer to this specification which was the one I used in the original headphones amp design but it costs about 5 times as much as an average Edcor,hence my interest in an alternative source.

Cheers

Ian

 

[ricardo]

Here's Great Guru Baxanadall's take on HF transformer equivalents to go with my Reply #87 stuff from RDH4 & Morgan Jones.  It's from
http://www.douglas-self.com/ampins/wwarchive/wwarchive.htm#baxihqa HIGH-QUALITY AMPLIFIER DESIGN Wireless World jan1948 when he was a young whippersnapper, fresh from the war & hobnobbing with Alan Blumlein.

It's a pentode power amp so the transformer characteristics would be critical.  This would have quite different behaviour with the RDH equivalent circuit.

I don't think I've ever considered a transformer might behave like this.  Any comments?  Did GG B change his mind later?

There's several articles with transformer details on this page which show why Ted Ashley, who designed the last of the great LEAK valve amps, told me Radfords were good cos they wound their own trannies and LEAK, cos of their good relations with Partridge over several decades.  That's not belittling both companies making significant contributions to valve amp circuit design.

 

[CJ]

that is pretty simple, no dcr or anything,

looks like he lumped the pri and sec models together,

seen about 8 different models for transformers, not much you can dow with them,

you can see how leakage C and L form a resonant circuit, which is the most important thing to get from a xfmr model,

thanks for the cool link, 

 

[abbey road d enfer]

This would have quite different behaviour with the RDH equivalent circuit.

I don't think I've ever considered a transformer might behave like this.  Any comments?

There's so many ways of lumping elements, depending what you want to analyse. I think he chose that topology because it demonstrated well enough his point.
I would probably put the cap after the leakage inductance most of the time, but here, considering the source is simplified to a pure current gen, it doesn't make much difference.

 

[ricardo]

There's so many ways of lumping elements, depending what you want to analyse. I think he chose that topology because it demonstrated well enough his point.
I would probably put the cap after the leakage inductance most of the time, but here, considering the source is simplified to a pure current gen, it doesn't make much difference.

Actually it DOES make a big difference in this case.

His circuit takes feedback from a tertiary winding which he says is arranged as if feedback is take from the anodes.  If so, the loop gain at the anode will drop at HF cos the current source feeds the winding capacitance.  If the capacitance is after the stray inductance, the loop gain at the anode will rise at HF cos the current source feeds the inductance.

But what about triode or ultralinear working where the anode resistance is 'near' the load resistance?  Assuming capacitances before & after like Jones clashes with the 12dB/8ve roll-off I think I see after SRF.  (better confirmed by phase measurements rather than plotting response).

 

[abbey road d enfer]

Actually it DOES make a big difference in this case.

His circuit takes feedback from a tertiary winding which he says is arranged as if feedback is take from the anodes.  If so, the loop gain at the anode will drop at HF cos the current source feeds the winding capacitance.  If the capacitance is after the stray inductance, the loop gain at the anode will rise at HF cos the current source feeds the inductance.

You're right; I had looked too quickly, thought FB was taken from the plate.
In this particular case, lumping the elements in one single place is not right. The capacitive load on the anode cannot be neglected, as well as the 2nd-order resonance with the leakage inductance. Then I would split the stray capacitance in two halves. In fact, that's what I do most of the time when I need to investigate the behaviour of high-Z transformers (> 10kohms). Pls have a look at the attached sim. It shows that lumping in one place does not give an accurate evaluation. It is either grossly optimistic or pessimistic; splitting in two elements is not enough. I reckon splitting in four is not too complcated and gives a pretty good estimate.

capacitances before & after like Jones clashes with the 12dB/8ve roll-off I think I see after SRF.  (better confirmed by phase measurements rather than plotting response).

Indeed measurements are better than simulation, but I don't see anything that would justify the idea that the 12dB/octave is tampered with.
I'll have a look at voltage-driven.

 

[abbey road d enfer]

Voltage-driven, even with a large-ish impedance shows the same hierarchy and the same predicted 12dB/octave response, but a much better HF extension, which is not a surprise.
It shows that Peter J. was not erring with his simple model; his conclusions were right, albeit inaccurate numerically.

 

[ricardo]

Thanks for these Abbey.  But are L11 L14 L15 L12 all inductively coupled to L13 and are those values 16H and 256H?

That's sorta winding capacitance (HF) with primary inductance (LF) so I'm not sure it represents the HF behaviour.  Your curves show 6dB/8ve so I think that's simply winding capacitance with 'load' resistance with no leakage inductance.

IIRC (I'm no SPICE guru), leakage inductance is simulated by making the 'primary' inductances not quite add up to the 'secondary' inductances.  They should be less than 10 mH for a good UL trannie.

Also I think your R2 = R4 = R6 = R8 = 30k is more like a output pentode than a 'voltage source'.  I make EL34s in 43% UL about Ra = 1k3 and even less for triode mode.

 

[abbey road d enfer]

Thanks for these Abbey.  But are L11 L14 L15 L12 all inductively coupled to L13 and are those values 16H and 256H?

yes.

That's sorta winding capacitance (HF) with primary inductance (LF) so I'm not sure it represents the HF behaviour.  Your curves show 6dB/8ve so I think that's simply winding capacitance with 'load' resistance with no leakage inductance.

You're right, I've completely missed this point. I shall introduce some leakage by altering the coupling factor.

Also I think your R2 = R4 = R6 = R8 = 30k is more like a output pentode than a 'voltage source'.  I make EL34s in 43% UL about Ra = 1k3 and even less for triode mode.

As you can see in the inductance values, this would be a transformer for let's say an ECC82. The primary inductance for an EL34 would be about 20H.

 

[abbey road d enfer]

That's the new graph with the leakage inductance (not explicit, but expressed by altering the coupling factor k to 0.99 - most audio xfmr's have 0.99<k<0.997).
It shows definitely, as you surmised, that Peter J. was not right with his simplified model. The leakage inductance is in series with the main, which just modifies slightly the resonant frequency and the 12dB/octave asymptote just doesn't happen! In a NFB analysis, that's seriously lacking.

 

[abbey road d enfer]

And the revised voltage-driven graph.
It shows problems that do not exist in reality; the discrete notches don't happen in real life. I think I'll investigate the possibility to model this with a transmission line. We'll see.

 

[PRR]

> discrete notches don't happen in real life

They do; but it is usually convenient to stop the sweep (or prune the graph) so they don't show. Honest byproduct of multi-winding construction.

 

[abbey road d enfer]

> discrete notches don't happen in real life

They do; but it is usually convenient to stop the sweep (or prune the graph) so they don't show. Honest byproduct of multi-winding construction.

I meant those notches that are a result of the arbitrary number in which the inductance is split for simulation.

 

[ricardo]

> discrete notches don't happen in real life

They do; but it is usually convenient to stop the sweep (or prune the graph) so they don't show. Honest byproduct of multi-winding construction.

Got any examples?  Abbey is trying to simulate distributed winding capacitances so this should hold even with simple 2 winding trannies.

Gotta admit its a lo...ong time since I looked at a valve O/P trannie.  But I expect to see 12dB/8ve after the SRF like on much smaller transformers.

 

[abbey road d enfer]

Abbey is trying to simulate distributed winding capacitances so this should hold even with simple 2 winding trannies. 

To make it clear, I'm trying to see how to improve on the single-lumped parameters approach, which finds its limits very quickly. I reckon a larger number of splits would be closer to reality. At the moment I don't make any assumption about the construction. If I was to make a sim of an existing xfmr, with a specific construction, I would alter the model, e.g. if the xfmr is sandwiched, I would make the split in accordance with the actual number of windings.
I've done specific analysis a long time ago (before Spice) on quadfilar construction and came to the conclusion that splitting the capacitances in two was enough to provide reliable results. These were low-inductance xfmrs, so this conclusion is to be taken with the necessary caution.

Gotta admit its a lo...ong time since I looked at a valve O/P trannie.  But I expect to see 12dB/8ve after the SRF like on much smaller transformers.

I agree. The main difference would be the high-level behaviour, where the main inductance decreases noticeably and the leakage inductance increases.

In addition, Peter J. 's assumption of a pure current-drive is another level of oversimplification.

 

[abbey road d enfer]

Done a sim with 8-splits. Increasing the number of splits indeed increases the number of notches and moves the asymptote about a 6th-octave higher. I don't think it justifies the added complexity.
As I mentioned I wanted to try the transmission-line approach, but I can't find info about it in LTSpice. I have sent a question to the LTSpice forum. We'll see...