QUAD 63 (and later) Delay Line Inductors

There was some interest in another thread about the modern Quad inductors that form a portion of the delay line in the 63's and newer Quad electrostatic loudspeakers.

Strangely, the British electronics in these speakers is quite robust. But over the 24+ years I've been fixing quads, i've seen just about every failure you can have. I've only seen a couple damaged inductors and they were the results of massive overdriving of the speaker. Below is my knowledge of these inductors, although I haven't really dug into them very deeply (both metaphorically and physically) which you'd want to do if you were going to reproduce them.

The inductors are air core units with a second shorting winding that increases the inductance dramatically beyond what a single coil of this size and wire gauge is capable of. The inductance is about 2.7H each if my measurements are correct. The attachment below shows my method and results. (NOTE: The DCR is in Kohms not ohms as noted on the sheet. And I went back and re-measured the inductor resistance and got the same numbers as the first two inductors; I beleive all inductors are identical). I used a function generator, a resistor and a voltmeter to measure the inductor by forming an AC voltage divider. Using the fact that and inductor has an impedance of: Z = 2 * PI * f * L, I was able to calculate L because I know, Z, and f.

http://quadesl.com/diyaudio/inductorCalc.png


As seen in the pictures, the copper coil is sitting in a plastic "bobbin" with three leads which are soldered to the delay line circuit board. The entire bobbin, and actually the entire delay line circuit board is encapsulated in bee's wax which makes these things easy to work on.

Two of the leads are connected via a buss bar inside the bobbin, and the third (middle) lead runs from the center of the coil. The shorting coil is attached to both the left and right attachment points on the buss bar (forming a short through that bar. And the main inductor coil is attached to one side of the buss bar, at the outside of the coil and the other end is attached to the center lead coming from the middle of the coil.

I don't have a feel for how the shorting coil is arranged relative to the main coil. I'd think it would be bifilar wound, but I have never completely unpotted one of these to check it out.

The way I've seen these fail is that the very fine wire breaks near the solder connection at the buss bar. It appears that the wire overheads and melts, but on come specimens the wire looks thinned which doesn't make sense with a pure melting failure mechanism.

To fix them, I melt the wax around the buss bar and in the center of the coil and using a short piece of jumper wire, I re-solder the fine magnet wire to the jumper and then the jumper to the buss bar. I use the jumper wire because without un-potting the entire coil, the wire where it breaks is too short to reach the buss bar.

Here's a couple pictures of the inductor including one through my stereo-microscope eyepiece. Taking a picture with my cell phone through the eyepiece is a real kludge but works reasonably well to document things.


http://quadesl.com/diyaudio/IMG_4075.JPG

http://quadesl.com/diyaudio/IMG_4076.JPG


Sheldon
quadesl.com
 
  • Like
Reactions: 1 user
2.7 H air cored? Wow!

You have probably known this for decades, but according to Peter Baxandall (in chapter 3 of Loudspeaker and Headphone Handbook, J. Borwick, ed., Butterworths, London, 1988), the shorted windings are meant to damp the transmission line for high audio frequencies, so treble is only produced by the middle of the loudspeaker. That has something to do with preventing edge effects due to the finite size of the ESL-63.
 
  • Like
Reactions: 1 user
The placement of the delay induktors is not a coincidence either... well nothing that Peter Walker did was. Each Induktor is affecting the nearby ones, all is taking care off in the design, measured and calculated and solved! Peter Walker, soo ahead of it´s time.
 
…The way I've seen these fail is that the very fine wire breaks near the solder connection at the buss bar. It appears that the wire overheads and melts, but on come specimens the wire looks thinned which doesn't make sense with a pure melting failure mechanism.
Thanks muchly for posting these pics!
Your mentioning the wire looking thinned, makes me wonder if the wire was unintentionally stretched beyond yield tension before or after wires were soldered terminated, and “necked” down. The increased resistance at that point would certainly lead to local heating. Alternatively, it is not uncommon for thin wires to fail near terminals when the solder joint is poor(ie of high resistance).
 
Hi Sheldon,

I've tried to capture the ESL63 in a LTSpice simulation.
Maybe you're interested in seeing the results, resp. freq. range and step response for the 8 sections and the impedance as seen from the Power Amp.

Hans
 

Attachments

  • Graph1.jpg
    Graph1.jpg
    559.3 KB · Views: 996
  • Graph2.jpg
    Graph2.jpg
    267.7 KB · Views: 977
That is very interesting, it seems that in your simulation, there isn't significant rolloff in the first few sections. That doesn't seem to follow what my ears are telling me, but I've never measured each section individually. To do it right, I'd have to do more than disconnect the sections, you'd need to replace the removed sections with an equivalent capacitance.

I could do that experiment with my test jig and a second uncharged mid panel. I have a panel test fixture (as seen here: SDS Audio Labs - Quad Electrostatic Loudspeaker Measurements) that would work well for such a test.

Sheldon
 
That is very interesting, it seems that in your simulation, there isn't significant rolloff in the first few sections. That doesn't seem to follow what my ears are telling me, but I've never measured each section individually. To do it right, I'd have to do more than disconnect the sections, you'd need to replace the removed sections with an equivalent capacitance.

I could do that experiment with my test jig and a second uncharged mid panel. I have a panel test fixture (as seen here: SDS Audio Labs - Quad Electrostatic Loudspeaker Measurements) that would work well for such a test.

Sheldon
Thanks for replying.
However there are still a few points to clear.

1) The most obvious is that the LF impedance curve being strongly dependent on the level.
The only component to my feeling that can cause this effect, are the 12 delay line inductors.
If I had one, I could measure their behaviour at various levels with my vector network analyzer.

2) The other component is the used audio transformer, but without knowing the various parameters.
When you have the possibility, it would be great if you could measure:
1) the input inductance (L1=Ls+Lp) with the secondary open
2) input inductance (Ls) with secondary short circuited and
3) output inductance (L2) with the input open.
That would give me some important parameters of this trafo.

Hans
 
However there are still a few points to clear.
1) The most obvious is that the LF impedance curve being strongly dependent on the level.
The LF impedance changes with signal level because the primary inductance of the transformer is a function of core permeability which increases with signal level until core saturation is reached. I have measured the inductance as a function of Vrms for a few frequencies and will post once I track down my measurements.

2) The other component is the used audio transformer, but without knowing the various parameters.
I also measured step-up ratio and parasitics of the transformer(leakage inductance, winding capacitance, and resistance) and will post those as well.

For proper modeling of the HF roll-off behavior that stokessd mentioned, you would need to incorporate the losses intentionally introduced into the lattice inductors with shorted turns. You would also need to include the inherent winding capacitance in parallel with each inductor.

Peter Baxandall had a nice technical write-up of the ESL63 design, pages 179 – 194 of “Loudspeaker and Headphone Handbook”, J. Borwick, Ed., which is currently available for free preview on Google Books. Loudspeaker and Headphone Handbook - Google Books

Attached is a screen grab highlighting a couple sections related to your modeling effort.
 

Attachments

  • ESL63-Model_notes.png
    ESL63-Model_notes.png
    529.5 KB · Views: 574
The LF impedance changes with signal level because the primary inductance of the transformer is a function of core permeability which increases with signal level until core saturation is reached. I have measured the inductance as a function of Vrms for a few frequencies and will post once I track down my measurements.


I also measured step-up ratio and parasitics of the transformer(leakage inductance, winding capacitance, and resistance) and will post those as well.

For proper modeling of the HF roll-off behavior that stokessd mentioned, you would need to incorporate the losses intentionally introduced into the lattice inductors with shorted turns. You would also need to include the inherent winding capacitance in parallel with each inductor.

Peter Baxandall had a nice technical write-up of the ESL63 design, pages 179 – 194 of “Loudspeaker and Headphone Handbook”, J. Borwick, Ed., which is currently available for free preview on Google Books. Loudspeaker and Headphone Handbook - Google Books

Attached is a screen grab highlighting a couple sections related to your modeling effort.
Thank you for your feedback.
I’m looking forward to see the results of your measurements.

Hans
 
Your transient analysis wasn't long enough to let the DC offset die down, so the 117mV RMS valuse is erroneous. You would need to continue the transient analysis for a second or so and then only analyze the last few mSec. Alternatively, you could use AC analysis instead. Also, be sure to include the DC resistance of the inductors which is 5680 ohm, so Z = R + jwL = 5680 + jwL. Using the data posted, I calculate value of L = 2.651 H
 
Last edited:
Your transient analysis wasn't long enough to let the DC offset die down, so the 117mV RMS valuse is erroneous. You would need to continue the transient analysis for a second or so and then only analyze the last few mSec. Alternatively, you could use AC analysis instead. Also, be sure to include the DC resistance of the inductors which is 5680 ohm, so Z = R + jwL = 5680 + jwL. Using the data posted, I calculate value of L = 2.651 H

Thanks a lot for giving Rser, this changes a lot.
Again using Sheldons values of 2.08V and 117mV, 1010 Ohm and 1Khz, I get 2.67 Henry.
I will integrate this Rser in my LTSpice diagram.

Hans
 
… will post once I track down my measurements.
Finally located the measurements I took back in 2016. Thank again to stokessd for letting me borrow one of the interface units he was purchasing. My main interest at the time was the transformer design, but took some additional measurements of the LC transmission line with the intention of working up a correlated LTspice model at some point, but it has been low my the priority list. Hopefully you will find the measurements sufficient for the purpose.

Attachment #1: The transformer primary inductance was measured for increasing input voltages applied directly to the primary(ie bypassing the 1.65 ohm resistors) showing the expected trend of increasing inductance with a sudden drop off as core saturation is approached.
Attachment #2: Parasitics were also measured. Step up ratio was 1:122 and nominal values of primary and secondary inductance are used. Note the unusually large leakage inductance and unusually small winding capacitance. Golfnut had posted some comments regarding why the transformer was designed this way.
About to take the ESL plunge
esl transformer.

Attachment #3: Transfer function of the unloaded transformer in red shows the expected peak where the leakage inductance resonates with the winding capacitance. The green line shows the response with the LC transmission line circuit is attached, but no ESL panels are connected. The orange line shows the response when 22pF capacitors are used to load the line.
Attachment #4, #5: Impulse response trends for the LC transmission line unloaded and then loaded with 22pF capacitors. Notice the increase in delay when capacitors are added.
Attachment #6, #7: Frequency response trends for the LC transmission line unloaded and then loaded with 22pF capacitors.

I can post a set of *.wav files containing the impulse responses if that would be helpful, just let me know.

***NOTE***
Meant to mention that the transformer output and LC line measurements were taken with a balanced HV probe having an input impedance of 2.7Mohm.
 

Attachments

  • ESL63_TLine_04.png
    ESL63_TLine_04.png
    207.9 KB · Views: 212
  • ESL63_TLine_03.png
    ESL63_TLine_03.png
    207.1 KB · Views: 239
  • ESL63_TLine_02.png
    ESL63_TLine_02.png
    131.5 KB · Views: 233
  • ESL63_TLine_01.png
    ESL63_TLine_01.png
    132.8 KB · Views: 267
  • ESL63_Transformer_03.png
    ESL63_Transformer_03.png
    34.4 KB · Views: 474
  • ESL63_Transformer_02.png
    ESL63_Transformer_02.png
    22.7 KB · Views: 496
  • ESL63_Transformer_01.png
    ESL63_Transformer_01.png
    38.3 KB · Views: 539
Last edited:
In the version below of my LTSpice sim, I have made the following changes:

1: for the 12 Inductances: L=2.67H Rser=5680 Ohm and Cpar = 22p.
2: the first 6 HF sections are set to 12 x 44pF , giving a series value of 22pF.
3: I have changed the coupling factor of the audio transformer from 1 to 0.9998. Seemingly a small step, but this seems a correct value to take the serial inductance into account.

Although FR and step response are much better as before, I fail to get the impedance peak ar ca 23Khz of the real thing.

Hans
 

Attachments

  • Graph3.jpg
    Graph3.jpg
    612.6 KB · Views: 184