Optimizing my DIY full range esl

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

I've built both a "small" (40x120 cm) and big (68x178 cm) electrostatic loudspeaker. Currently I use the big panels as they go lower in frequency, although the small panels still perform quite good given theire relative small size.

I will post some near-field and far-field frequency repsonse measurements, but it turns out there is a very steep high (Q) resonance at 25 Hz on both panels. I used small rectangular spacers, two vertical rows, similar to silocone dots, so I guess the membrane have multiple resonance frequencies, but at 25 Hz there is a very clear resonance. Fortunately, only in very specific recordings this is audible - the membrane starts to rattle.

I used the esl_seg_ui software by Edo Huzlebos to simulate and design the ladder segmentation. I have 9 wire groups of different widths - the middle is about 1 cm width and the groups get wider towards the sides of the panel.

I have a few questions about frequency response, especially about low frequency repsonse.

My first question: does the esl_seg_ui software take into account dipole acoustic cancellation?

My second question: at what frequency starts acoustic cancellation? At Wikipedia (link) I read:

"The bass rolloff 3db point occurs when the narrowest panel dimension equals a quarter wavelength of the radiated frequency for dipole radiators, so for a Quad ESL-63, which is 0.66 meters wide, this occurs at around 129 Hz, comparable to many box speakers (calculated with the speed of sound taken as 343 m/s)."

Bass rolloff of my 68 cm wide panels would then start at 343 / (0.68 * 4) = 126 Hz at 6 dB/oct.
 
Last edited:
… does the esl_seg_ui software take into account dipole acoustic cancellation?
The software assumes that your segmented ESL is an infinite dipole line source with no baffles on the sides. The natural response of such an ESL driven with constant voltage is a response that slopes up at +3dB/oct. Segmentation not only improves the polar response, it flattens the rising response. The sloped response is due to both “dipole cancellation” at lower frequencies and directivity gain or “beaming” at higher frequencies. With no baffle, there is a smooth transition between these two ranges resulting in a constant +3dB/oct slope. If you have baffle edges that are of significant size relative to the ESL width, they will put a slight step or offset in the response separating the two sloped ranges.

Here is an example measurement of an un-segmented ESL line source showing the +3dB/oct slope.
Current vs voltage drive ESL?

Here is another example measurement of a nearly massless dipole ribbon showing the constant slopes and slight step offset from baffle edges. (Note the slopes are +6dB/oct since it is small and behaving like a point source. If it would tall, it would be have had a +3dB/oct slope of a line source)
DIY ribbon dipole tweeter, reductio ad minimum

More details on line source assumptions here:
Segmented Wire Stator ESL simulator (esl_seg_ui)

Example measurements-vs-modeling of what happens if your ESL is too short to act like a line source and transitions to a point source:
First ESL build - Full Range .


…at what frequency starts acoustic cancellation?
Dipole cancellation starts at about lambda/2 and by lambda/4 the response is -3dB down. But as mentioned above, with dipole ESLs the response does not flatten out above this point. Because of directivity gain(and the lightness of the diaphragm), the response has a constant slope over the whole audio frequency range.

More details here: Dipole phase-cancellation in ESL speakers
 
Hi Bolsert, thanks for your extensive and excellent reply. If I'm correct the slope rate depends on differnt facors. A few factors are (taken from your links)

* (far field) 3 dB/oct. for line source voltage driven esl's
* (far field) 6 dB/oct. for point source voltage driven esl's
* current drive (by placing a resistor at primary or secondary of the audio transfomer) changes contant voltage to constant current drive and flattens frequency response
* stator segmentation and ladder nresistor etwork adds area with decreasing frequency and flattens frequency repsonse

My esl doesn't reach the ceiling, there's at least 1 meter free space between the top of my els and the ceiling. And my listening position is bigger than the height of my esl. So I would think my esl is somehtng "in between" a point and line source - or at least not a true line source? That would mean that the slope would be in between 3 and 6 dB/oct (far field, voltage driven)?
 
Last edited:
…I would think my esl is somehtng "in between" a point and line source - or at least not a true line source? That would mean that the slope would be in between 3 and 6 dB/oct (far field, voltage driven)?
When your ESL is a finite line source, it will behave like an infinite line source(+3dB/oct) at high frequencies, and a point source (+6dB/oct) at low frequencies. The frequency at which the response transitions between these two slopes is dependent both on the line length and the listening distance.
F = c * r / d^2

where:
c = speed of sound
r = listening distance
d = panel dimension (height)

You can use the latest version of the segmented ESL spreadsheet(experiences with ESL directivity?) to plot this be choosing Line Source = Finite, setting the number of segments to n=1, the low frequency break point fL = 100000, and the diaphragm resonance frequency Fs=0. The solid green line would be the response for an infinite line. The dashed green line shows the response for the finite line.

The ESL spreadsheet uses an average trend line of the response. If rigorous calculations are performed you will see the response has ripples superimposed on the response, with a hump at the transition from line source to point source behavior.

An example calculation of this is attached. (taken from here: Help with esl simulator)
An example measurement-vs-theory can be found here: Vertical dispersion on planars. How much?
 

Attachments

  • ESL_Finite_Line_Transition.png
    ESL_Finite_Line_Transition.png
    90.2 KB · Views: 1,097
  • ESL_Linemodel_on-axis.jpg
    ESL_Linemodel_on-axis.jpg
    90.6 KB · Views: 1,086
Hi Bolsert,

again, thans for your helpful reply and example calculations, this is really appreciated.

If I'm correct my stator simulated by Edo Hulzebos his esl-seg-ui software, is calculated with the assumption of an infinite line-source whereas my esl that has a height of 180 cm that is smaller than the ceiling height of 280 cm is a point-source for lowest frequencies (below (343 * 3) / 1.8^2 = 318 Hz at 3 m listening distance). If I'm correct Edo's software compensates for a +3dB slope whereas my slope below 318 Hz is a +6dB slope, so the final result after calculated segmentation / correction by the esl-seg-ui software would be +0 dB/oct above 318 Hz and +3dB/oct slope below 318 Hz?

If my statement is correct, I could use a low-shelving filter of +3dB/oct. I tried that using my MiniDSP 2x4HD, but although there was a subtle audible boost at low frequencies, it was unlcear to me whether this was an improvement or not. So I decided to leave it as it is.
 
Your understanding is correct concerning the compensation of the ESL itself. Of course, there may be room mode related problems below 300Hz that might need to be addressed as well. One other important thing is to handle the diaphragm resonance, which may be complicating the LF behavior. I forget if you are using acoustic mesh damping or not, but since you have a miniDSP, you could measure the resonance with near-field mic placement and then use a Linkwitz Transform to modify the resonance Q to something around 2.
Linkwitz Transform
Set f(0) and f(p) equal to the diaphragm resonance frequency.
Set Q(0) to match the diaphragm resonance Q
Set Q(p) = 2

This along with the shelving compensation will improve the fidelity of low frequency reproduction.
 
Hi Bolserst,


thanks for your reply and directions, I implemented the Linkwitz Transform filter in my MiniDSP and even though it lowers the resonant peak, the rattling sound still is there. (I have only 1 cd that is able to produce this problem).


Below is a measurement before and after applying the filter:


linkwitz-transform-fq.jpg



So in my opinion there is no audible improvement even though the peak is lowered by about 10 dB. Even when applying steep (48 dB/oct) high-pass filters the rattling sound is still there.


I can attenuate this sound quiete a bit by just clamping a wooden board aganst the back of one panel covering the whole vertical length of the panel. (using glue clamps) I measured the effect of clamping 1 panel of about 40x160 cm perpendicular against one side at the back of the panel. I got this idea because the panel is rather tall and wide but thin and it lacks a stiff supporting frame. So when I stamp my feet I can see the top of the panel moving back and forth.


Below: mic @ 3 meter from panel, right channel, in room:


fq-l-bracing.jpg



Below: left channel mic @ 7cm from panel:


fq-r-bracing.jpg



To conclude, it seems that the frame itself resonates as well, resulting in a higher Q.
 
Last edited:
.... I implemented the Linkwitz Transform filter in my MiniDSP and even though it lowers the resonant peak, the rattling sound still is there


Your measurements are very interesting. Im just a noob so forgive my blunders but it looks like your (L) and (R) membranes resonate freely below 40Hz. The frame resonance seems like a secondary issue with more essential issues to sort out? The filter you tried seems like it wouldn't work as the resonance after its applied was still up by nearly 10dB from 20-30Hz where the high Xmax at such a low frequency could create slapping.


When you tried the steep high-pass filters what frequency did you use? You might need to have it up at 46Hz or higher to attenuate enough at 30Hz and below to work?



Would tensioning up the membrane (eg heat shrinking ) help to push the resonance freq. higher?
 
… I implemented the Linkwitz Transform filter in my MiniDSP and even though it lowers the resonant peak, the rattling sound still is there. (I have only 1 cd that is able to produce this problem). Below is a measurement before and after applying the filter:
The rattling sound you mention, does is sound like the diaphragm is hitting the stators?

Wow! That is some peak you have there, +25dB. With Q>20, it will be very difficult to get the Linkwitz Transform filter aligned exactly right. Your best option might be to use some acoustic damping to lower Q of the resonance to the point where you can more easily take care of the rest with the L-T miniDSP. I know you had experimented unsuccessfully with acoustic damping in the past, but you might want to revisit it as a partial solution to use in combination with the miniDSP.

To conclude, it seems that the frame itself resonates as well, resulting in a higher Q.
To avoid the problem of frame resonance, you need to make the natural resonance frequency of the frame(fF) significantly different from the primary resonance of the diaphragm(fD) so one doesn’t easily excite the other. With large full range ESL, this can be difficult. You need to either stiffen the frame (like you did with clamped board) to increase fF, or add mass to lower fF. Perhaps if you can move fF away from fD the Q of resonance will be lowered enough to handle more easily with the miniDSP, acoustic damping, or a combination.
 
Hi bolsert,

sorry for my (very) late reply, I missed it. Thanks again for your extensive and insightful repy. I think I will keep this stat as it is and design and build a new version that hopefully solves the limitations / problems of my current model.

This is a rough picture just to demonstrate my current idea:

ESL-v4-diagram.png


(the areas are just an indication, not exact dimensions. The red / blue colored areas constitute the front view of the complete stat, the drawing below is to demonstrate the electrical ladder segmentation)

design decisions:

* separate bass panel in order to use damping screen exclusively for bass unit so the mid / treble panel keeps it's "transparent" sound (I didn't like the sound of my current stat with damping screen because of deteriorated mid / treble sound quality
* the bass panel is less (at least I intent) sectioned compared to the mid / treble panel in order to get a lower resonant frequency for the bass panel. This will increase output at lower frequencies compared to a bass unit that exists of multiple smaller sections (is this true?)
* the bass panel will still get a few silicone dots in order to keep the big diaphragm from hitting the stators
* the mid / treble panel is physically separated from the bass panel by spacers. (actually it is a full-range panel as it gets LF as well just like the bass panel) The mid / treble panel is mechanically divided in different areas in order to spread resonance energy over a broader frequency range. This panel lacks damping screen in order to keep mid / high frequencies 'transparent' sounding.
* all panels are connected to one pair of step-up transformers (2*1:75=1:150 step up ratio) in order to keep things simple
* electrical ladder segmentation is applied over the full surface of the complete stat: one wide section for the bass panel and multiple, width-decreasing sections for the mid / treble panel in order to get both good dispersion and impedance load

I guess this idea isn't unique as if I remember correctly Martin Logan has done similar designs. But before starting to make drawings and implementing this design, what do you think about this idea? Do all points made above make sense? Are they correct? Could this model become an improvement to the current model which is only electrically segmented and has a few silicone dots?
 
Last edited:
Hi,

I just came up with a few new questions regarding my previous post:

* instead of using damping screen to damp the bass panel resonance, I could use mechanical segmentation for the bass panel as well. I'm aware some people don't like the idea of physical segmentation as it would remain high-Q and uncontrolled motion, distorting a wider bandwidth. Furthermore, if I'm correct, a mechanically segmented bass panel has lower efficiency (sound output) and less extended low frequency range compared to a non-segmented bass panel. That's why I lean toward a mechanically unsegmented (maybe I would still use silicone dots so it is still somehow segmented) bass panel and use fine mesh silk screen to lower Q / damp resonant frequency. Because one design goal is to get a deep / high output LF production.

I'm wondering if I would gain deeper and more low frequency output when using this approach compared to mechanically segmented stators?

* Another question I have is: does it make sense to have a phycical separate bass panel so I can cover the whole back stator with silk screen in order to efficienctly damp resonant frequency? Would this be more efficient compared to 1 total diaphragm that plays all frequencies and partly cover its surface? I would think in the latter option air would more easily leak to the open area(s) compared to the first mentioned option where the whole back of the stator is covered with screen?
 
@ silvershadelynx,

You have put forth a lot of ideas and questions. You must know from your experience that designing a full range ESL with high output and well damped bass is very difficult…something I have not done. But, I have performed many experiments and measurements that may be helpful in your quest.

Concerning mechanically segmented panels and bass output:
You are correct that mechanically segmenting a given panel area will reduce low bass output capability compared to if the entire panel has the same resonance. For example, the Sound Lab speakers only have about 1/3 of the total panel producing useful output down at 30hz.

But, as you have found out, having a large panel area with a single resonance frequency results in very high Q. If use of damping mesh is not desirable, having a vertical stack of panels(or single mechanically segmented panel) with slowly varying sizes as you suggest will result in lowered Q for each of the panel resonances. Measurements of the Sound Lab A-1 can be seen here: Full Range Electrostatic Question They have recently modified their mechanical segmentation scheme to have large areas both at the top and bottom of the panels to increase radiating area at lower frequencies. This increases the amount of panel area contributing to output in the bottom octave. They call it “bass-focus”. (see attached pic and *.pdf)

However, placing a single large bass panel with a much lower resonance next to the line of higher resonance panels (like in your drawing) will likely cause some undesirable interaction. The resonance frequencies of adjacent panels need to be close together to affect some amount of damping without ill effects. See attached measurements from (20 years ago :eek: )

Concerning diaphragm resonance damping:
Notice in the measurements of the Sound Lab speakers linked to above that the resonances are of lower Q than when you use one large diaphragm area with a single resonance frequency. This technique may be all you need to lower the Q enough to make DSP equalization of the milder peaks possible.

One other method of damping diaphragm resonance uses a passive “bass” panel set next to your ESL. The passive panel reduces excursion of the driven ESL as well as contributing to the total output, not unlike vented enclosures used with dynamic drivers. At least one company uses this method. (Pic attached in case website is removed in the future)
Electrostatic loudspeakers model Magic Music - DEA Electrostatics | Audio systems design and production

Measurement of this technique show it to work much better than I would have expected. The only trick would be to get the frequency of the passive matched to the active panel. Attached is a plot comparing measurements of a large ESL: undamped, with acoustic mesh damping, and undamped but with a passive damping panel.
For more details and many other measurements using this technique, see attached Passive_Damp_v2.pdf file.

Concerning resonance of stator structure :
In my own speakers I chose to limit ESL output below 75Hz and use dynamic subwoofers to handle the bottom 2 octaves. This allowed me to mass load the stator frames to push their resonance below that of the diaphragm. In your case you will need to make the stators as stiff as possible to keep their resonance above the diaphragm resonance. For your consideration, the Sound Lab ESLs use stator stiffeners that are over 10cm deep to get the stiffness required. Attached are 2 pics of the front stator and one of the rear. The resulting stator structure is incredibly solid.
 

Attachments

  • SL_Bass-Focus.png
    SL_Bass-Focus.png
    531.5 KB · Views: 492
  • Bass-Focus_Technology.pdf
    192.9 KB · Views: 146
  • ESL_Resonance_Interaction.png
    ESL_Resonance_Interaction.png
    23.7 KB · Views: 516
  • Passive_Damp_v2.pdf
    732.3 KB · Views: 208
  • Passive_Damping_Comparison.png
    Passive_Damping_Comparison.png
    177.3 KB · Views: 515
  • DEA-Magic-Music-72.jpg
    DEA-Magic-Music-72.jpg
    45.6 KB · Views: 505
  • soundlab2.jpg
    soundlab2.jpg
    775.9 KB · Views: 488
  • Majestic 945-4.jpg
    Majestic 945-4.jpg
    91.4 KB · Views: 282
  • Majestic stator.jpg
    Majestic stator.jpg
    69 KB · Views: 265
Last edited:
Hi Bolserst,

thanks again for your great reply and sorry for my late reply. It seems more difficult than expected to get enough LF range without resonant peaks. I tried a passive element by using one channel as a passive element with varying distances to the neighbouring active element, but without much effort. (it only dampens a few dB's in my test case).

It seems, according to your old tests, that putting mechanically seperated elements of wide varying sizes results indeed in unwelcome results / frequency resonse. I think it's not worth the effort to try - to much work.

I'm not convinced the sould labs solution of distributed resonance is would result in an improved sound quality. It's a lot of effort to build and who knows one exchanges one problem for a different one. (less output in the lowest octaves).

To conclude, maybe my "full range" stats are not that bad at all. The only option to improve LF / resonant peak left would be to make an even wider and taller stat, but getting it mechanically stable will be more difficult I guess. Maybe I will try this in future. If so, then I will keep you informed!
 
…I tried a passive element by using one channel as a passive element with varying distances to the neighbouring active element, but without much effort. (it only dampens a few dB's in my test case).
Did you have the bias supply on your passive element powered up? Because of the negative compliance associated with bias voltage, it is possible that if you were not using any bias voltage on the passive panel its resonance frequency was higher than the active element and was therefore not able to provide significant damping. Also, did you have felt or other damping material on the passive element? For all of the panels I experimented with if 1) resonance was matched, 2) damping material used on the passive element, and 3) passive element placed with minimal space beside the active element, significant damping was achieved.

If unfamiliar with bias voltage related negative compliance:
https://www.diyaudio.com/forums/pla...agm-resonance-change-hv-bias.html#post1884466
Current vs voltage drive ESL? Post#66
Current vs voltage drive ESL? Post#29


…I'm not convinced the sound labs solution of distributed resonance is would result in an improved sound quality. It's a lot of effort to build and who knows one exchanges one problem for a different one. (less output in the lowest octaves).
After experimenting with several prototypes, that was my conclusion as well.
 
Hi,

A couple of years ago I built a pair of ESL's using Sound Labs' principle of distributed resonance. And at least for me, they were well worth the time and the effort. However, as you mentioned , they are a lot of work, and if not for a heck of a lot of guidance from Bolserst, they would now probably be out in the patio used as bug zappers. I don't know if this is a pro or con, but the stator resonances are also distributed. Good luck with all your tinkering!
 
... However, placing a single large bass panel with a much lower resonance next to the line of higher resonance panels (like in your drawing) will likely cause some undesirable interaction. The resonance frequencies of adjacent panels need to be close together to affect some amount of damping without ill effects. See attached measurements from (20 years ago :eek: )

It has been a while since posting my idea of a single large bass panel next to another panel which has a (much) higher resonance panel(s). This could lead to ill effects. But the idea still keeps in my mind. I'm just wondering if commercially planar loudspeakers like the Apogee Duetta:

Apogee Duetta Signature – info | HFA - The Independent Source for Audio Equipment Reviews

...would suffer from these ill effects as, if I'm correct, share a same design? (multiple panels next to each other having a higher resonance?
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.