Its long past time I shared my ESL experience with fellow DIYers here:
Background:
Many, many, years ago, before I was married, I had the pleasure of owning a set of Quad ESL-57s. Of course they were intolerable objects because they would not support potted plants. So under considerable duress I sold them and replaced them with KEF CS7s, the largest of the constructor series KEF made, which had ample room for potted plants.
Needless to say, I always regretted selling the 57s; stereo was never quite as addictive after that. years later, in late 2007, a work colleague asked me a silly question that prompted me to look into ESLs again. To my great surprise there was a good sized DIY community, and so I began …..
Acknowledgements:
In addition to the many people who post here, I owe special thanks to Rob McKinlay of ER Audio, whose notes on his ESL kitsets convinced me that a DIY project was feasible. Also, Rob introduced me to the Loudspeaker and Headphone Handbook (Borwick) chapter on ESLs by Baxandall, which was a revelation. I also owe thanks to colleague PM, whose encyclopaedic memory I pick, and MP for helping me when I get stuck on some basic acoustics theory. Finally, thanks to Bolserst for the many off-line discussions and advice.
The design:
Towards the end of Baxandall’s chapter he derives the ‘Walker Equation’ relating SPL to stator current, for a line source – actually a floor-to-ceiling doublet. The equation has an awkward sqrt(f) dependence, which superficially at least, looks hard to equalise. However, the input impedance of an RC transmission line also has a sqrt(f) dependence. That means that an ESL comprised of many segments of equal capacitance, each linked to its neighbour by a single resistor, will have a flat frequency response. Following some computer modelling, I discovered that the symmetric version of this ESL has some additional magic – a polar response that varies as a second-order Butterworth response versus angle – that means (i) a broad listening area near the on axis position, (ii) no zeros and phase reversals in the frequency response (causing the phasiness that wide-segment ESLs are known for) and (iii) a graceful decrease in treble as the listener moves of axis. The results of the study were published in JAES as “A wide range electrostatic loudspeaker with a zero free polar response”. Contact me if you want a pre-print.
Practical build:
There are unfortunately a few beginners’ mistakes – I’ll explain as I go.
The finished ESL is about 500 mm wide by 2320mm tall – designed to fit under a 2400 mm ceiling. According to the design equations, the ESL has a nominally flat response down to 80 Hz, and then falls a way as sqrt(f) below that. In practice diaphragm resonance will boost the low frequency end – see later. I have chosen the smallest segments to be 12 mm wide to ensure minimal diffraction effects from the finite segment width.
The stators are manufactured from eight sections of 1100 mm x 384 mm x 0.8 mm PCB with 3 mm wide slots. They are reinforced with acrylic ribs glued in between the slots, every 40 mm. The spacers are also acrylic, 3.0 mm. The final stator-diaphragm spacing is less than 3mm due to half a dozen coats of acrylic lacquer. I have used copper tape as the stator ring on the rear stator, 3.5 um Mylar (supplied by Rob McKinlay), and spray-on Licron (NZ supplier available). I used the technique recommended by Rob to attach the diaphragm – stretched on an old glass shower panel using duct tape, to constant tension, and glued using slow setting epoxy.
Mistake 1: I designed the mechanical construction, in particular the split down the middle of each panel, before I discovered that an asymmetric ESL has a terrible polar response. So on each panel, one half is symmetric transmission line (for treble) and the other (bass end of the transmission line) is asymmetric. Not too much damage done – within a couple of dB of ideal.
Mistake 2: I assembled the PCB stators with the copper surface inwards, and the machined edges (followed by sanding and etching to minimise the problems) mean that I begin to get corona discharges at about 4 kV, and I’m having to run the ESL with half of the polarising voltage that I hoped for – 2.6 kV. However, I find I have sufficient area to still get them comfortably loud. The HT supplies have 100 M ohm in series to ensure constant charge operation (minimise distortion).
Mistake 3: For each of the two transformers in each ESL I used eight 20VA toroidal power transformers rated for 230Vac at 50 Hz. That gives me secondary voltages of up to plus-and-minus 2.6 kV at 50 Hz. I stripped the secondary windings off the transformers and put a single winding through a group of eight to make a sausage-like transformer. I did this with the hope of reducing the secondary winding thickness and hence reducing leakage inductance - but actually made no practical difference. I did a whole bunch of experiments on some 15 VA toroids (fo = 126 kHz), which suggested that a transformer of this design would have an unloaded resonance at 44 kHz. Unfortunately I spotted some 20VA transformers going cheap – but they turned out to have two 115Vac windings that had been wound simultaneously – hence very high winding capacitance, and the unloaded resonant frequency of the multicore transformer turned out to be only 23 kHz, and about 18 kHz with the ESL attached. So I lost a bit of the upper end (I can’t hear it anyway, but others might).
With the primaries of the transformers in parallel, I have a nominal step up ratio of 120. This gives the ESL a very low sensitivity, but on the plus side, my amplifier saturates before breakdown occurs, so I don’t need a protection circuit.
Conclusion:
More by good luck than good management, I finished up with a diaphragm resonance at 37 Hz – which means my ESLs have a -3dB cut-off a touch lower (response falls as f^2.5 below this). I used screen printing mesh to damp the resonance from a Q of about 30 down to perhaps 2 or 3 - this tops up the SPL below the nominal electrical cutoff at 80 Hz. So I estimate the final frequency response from 35 Hz to 18 kHz (guess + modelling – I hope to do measurements soon).
I am really pleased with the results, especially considering that I did not know what I was doing when I fixed most of the design parameters. Most of all I had not expected the fantastic bass – it goes further down than the KEFs did and with no crossovers etc, it’s seamless - fantastic. The midrange and tops are every bit as good as I hoped for – vocals, acoustic guitars, etc give eerily good imaging, and metallic instruments sound metallic. Most telling I think, on really dense, heavily orchestrated music, the music opens up – things I never heard on other systems.
They’re not as loud as the KEFs, but loud enough that I never have to turn them up to full volume. On first hearing, the line-source behaviour is weird – with the pair operating, the volume through my lounge is almost uniform – which gives the illusion of them getting louder as you move away. I could do with more cores in the transformers or higher polarising voltage – I have a couple of tracks with a very deep bass note where the core saturates and the amplifier protection circuit kicks in, but they have to be quite loud for this to happen.
The WAF (wife acceptance factor) is truly appalling, especially with the day-glow yellow screen printing mesh on the backs. It’s probably a good thing there is no room for pot plants. However, the poor WAF appears to have been offset by the considerable improvement in sound quality – if I do a good job of the final clothing, I might yet be able to keep them in the lounge.
If I have managed to attached the picture correctly it shows he ESLs with one of the KEFs.
Thanks again to everyone at DIY, I’ve learned a lot.
Background:
Many, many, years ago, before I was married, I had the pleasure of owning a set of Quad ESL-57s. Of course they were intolerable objects because they would not support potted plants. So under considerable duress I sold them and replaced them with KEF CS7s, the largest of the constructor series KEF made, which had ample room for potted plants.
Needless to say, I always regretted selling the 57s; stereo was never quite as addictive after that. years later, in late 2007, a work colleague asked me a silly question that prompted me to look into ESLs again. To my great surprise there was a good sized DIY community, and so I began …..
Acknowledgements:
In addition to the many people who post here, I owe special thanks to Rob McKinlay of ER Audio, whose notes on his ESL kitsets convinced me that a DIY project was feasible. Also, Rob introduced me to the Loudspeaker and Headphone Handbook (Borwick) chapter on ESLs by Baxandall, which was a revelation. I also owe thanks to colleague PM, whose encyclopaedic memory I pick, and MP for helping me when I get stuck on some basic acoustics theory. Finally, thanks to Bolserst for the many off-line discussions and advice.
The design:
Towards the end of Baxandall’s chapter he derives the ‘Walker Equation’ relating SPL to stator current, for a line source – actually a floor-to-ceiling doublet. The equation has an awkward sqrt(f) dependence, which superficially at least, looks hard to equalise. However, the input impedance of an RC transmission line also has a sqrt(f) dependence. That means that an ESL comprised of many segments of equal capacitance, each linked to its neighbour by a single resistor, will have a flat frequency response. Following some computer modelling, I discovered that the symmetric version of this ESL has some additional magic – a polar response that varies as a second-order Butterworth response versus angle – that means (i) a broad listening area near the on axis position, (ii) no zeros and phase reversals in the frequency response (causing the phasiness that wide-segment ESLs are known for) and (iii) a graceful decrease in treble as the listener moves of axis. The results of the study were published in JAES as “A wide range electrostatic loudspeaker with a zero free polar response”. Contact me if you want a pre-print.
Practical build:
There are unfortunately a few beginners’ mistakes – I’ll explain as I go.
The finished ESL is about 500 mm wide by 2320mm tall – designed to fit under a 2400 mm ceiling. According to the design equations, the ESL has a nominally flat response down to 80 Hz, and then falls a way as sqrt(f) below that. In practice diaphragm resonance will boost the low frequency end – see later. I have chosen the smallest segments to be 12 mm wide to ensure minimal diffraction effects from the finite segment width.
The stators are manufactured from eight sections of 1100 mm x 384 mm x 0.8 mm PCB with 3 mm wide slots. They are reinforced with acrylic ribs glued in between the slots, every 40 mm. The spacers are also acrylic, 3.0 mm. The final stator-diaphragm spacing is less than 3mm due to half a dozen coats of acrylic lacquer. I have used copper tape as the stator ring on the rear stator, 3.5 um Mylar (supplied by Rob McKinlay), and spray-on Licron (NZ supplier available). I used the technique recommended by Rob to attach the diaphragm – stretched on an old glass shower panel using duct tape, to constant tension, and glued using slow setting epoxy.
Mistake 1: I designed the mechanical construction, in particular the split down the middle of each panel, before I discovered that an asymmetric ESL has a terrible polar response. So on each panel, one half is symmetric transmission line (for treble) and the other (bass end of the transmission line) is asymmetric. Not too much damage done – within a couple of dB of ideal.
Mistake 2: I assembled the PCB stators with the copper surface inwards, and the machined edges (followed by sanding and etching to minimise the problems) mean that I begin to get corona discharges at about 4 kV, and I’m having to run the ESL with half of the polarising voltage that I hoped for – 2.6 kV. However, I find I have sufficient area to still get them comfortably loud. The HT supplies have 100 M ohm in series to ensure constant charge operation (minimise distortion).
Mistake 3: For each of the two transformers in each ESL I used eight 20VA toroidal power transformers rated for 230Vac at 50 Hz. That gives me secondary voltages of up to plus-and-minus 2.6 kV at 50 Hz. I stripped the secondary windings off the transformers and put a single winding through a group of eight to make a sausage-like transformer. I did this with the hope of reducing the secondary winding thickness and hence reducing leakage inductance - but actually made no practical difference. I did a whole bunch of experiments on some 15 VA toroids (fo = 126 kHz), which suggested that a transformer of this design would have an unloaded resonance at 44 kHz. Unfortunately I spotted some 20VA transformers going cheap – but they turned out to have two 115Vac windings that had been wound simultaneously – hence very high winding capacitance, and the unloaded resonant frequency of the multicore transformer turned out to be only 23 kHz, and about 18 kHz with the ESL attached. So I lost a bit of the upper end (I can’t hear it anyway, but others might).
With the primaries of the transformers in parallel, I have a nominal step up ratio of 120. This gives the ESL a very low sensitivity, but on the plus side, my amplifier saturates before breakdown occurs, so I don’t need a protection circuit.
Conclusion:
More by good luck than good management, I finished up with a diaphragm resonance at 37 Hz – which means my ESLs have a -3dB cut-off a touch lower (response falls as f^2.5 below this). I used screen printing mesh to damp the resonance from a Q of about 30 down to perhaps 2 or 3 - this tops up the SPL below the nominal electrical cutoff at 80 Hz. So I estimate the final frequency response from 35 Hz to 18 kHz (guess + modelling – I hope to do measurements soon).
I am really pleased with the results, especially considering that I did not know what I was doing when I fixed most of the design parameters. Most of all I had not expected the fantastic bass – it goes further down than the KEFs did and with no crossovers etc, it’s seamless - fantastic. The midrange and tops are every bit as good as I hoped for – vocals, acoustic guitars, etc give eerily good imaging, and metallic instruments sound metallic. Most telling I think, on really dense, heavily orchestrated music, the music opens up – things I never heard on other systems.
They’re not as loud as the KEFs, but loud enough that I never have to turn them up to full volume. On first hearing, the line-source behaviour is weird – with the pair operating, the volume through my lounge is almost uniform – which gives the illusion of them getting louder as you move away. I could do with more cores in the transformers or higher polarising voltage – I have a couple of tracks with a very deep bass note where the core saturates and the amplifier protection circuit kicks in, but they have to be quite loud for this to happen.
The WAF (wife acceptance factor) is truly appalling, especially with the day-glow yellow screen printing mesh on the backs. It’s probably a good thing there is no room for pot plants. However, the poor WAF appears to have been offset by the considerable improvement in sound quality – if I do a good job of the final clothing, I might yet be able to keep them in the lounge.
If I have managed to attached the picture correctly it shows he ESLs with one of the KEFs.
Thanks again to everyone at DIY, I’ve learned a lot.
