Completely agreed!
The whole doc (I don't think one could call it a paper) has so many basic errors in it, it appears to have been written by a 1st year student and hasn't been reviewed at all.
The main question is in the fact that comb filtering is inherent to a stereo system in a room and our auditory system is able to deal with. 2 examples :
- lets imagine a stereo triangle (left and right loudspeakers, listener) of 2.5m. For a second listener just beside, let say 0.5m out off the triangle tip, the difference of path for him for the sound coming from one loudspeaker to the other one is 0.5m
- With the same listening distance 2.5m and a 0.9m listening height (so 0.9m above the floor), the difference of path between the floor reflection and the direct signal is 0.6m.
An effect might appear in the polar response leading to a change of spectrum of the reflected sound compare to the direct one.
So, to create delayed sources is not difficult but is it really the right set up?
I'd say yes and no. The stuff coming out of left and right can differ. Maybe why centre speakers have been added. DSP's can do interesting things. Cross overs can probably mimic this too. But ok take your point. There are pure listening problems and nothing can be done about some of them. Maybe all listeners should wear headphones.
Are you using these exciters perhaps?
https://www.ecplaza.net/products/50-100w-powerful-48ohm-flat-panel-4599230
https://www.ecplaza.net/products/50-100w-powerful-48ohm-flat-panel-4599230
Comb filtering is simply what happens to the FR when out of phase signals combine. It happens all the time more or less whenever you have multiple sources or reflections.
I'm not great with physics, but sound travels faster in solid materials than in air, hence the phase change per distance will decrease. I have the exciters a few cm apart, lets say 5cm and assuming a speed of sound in EPS is similar to balsa wood at 5000m/s, you get a wavelength of 5 cm at 100000 Hz. So at half of that, 50000 Hz, you will get maximum cancellation. So well outside the range of both what the exciter can reproduce and what we can hear. You might still get some effects down in the hearing range, but if you instead have two separate plates, in my case that would mean 35 cm distance between sources, and with speed of sound through air that would mean maximum cancellation at around 500 Hz, smack bang in a critical range in our hearing.
However, DML will not produce a phase coherent signal. Instead of a coherent wave front spreading from a source, you have a complex web of waves with different phases coming from many different sources, the nodes and anti-nodes appearing on the plate, generating the same frequency from multiple points at the same time with but with different phase from each other. This makes it behave very differently than regular speakers when it comes to things like how we perceive also the relative phase between multiple sources or reflections.
However, in the end the theory doesn't really matter, only the results. I think DML should sound really bad in theory, and I was a bit concerned that multiple exciters would be a bad idea initially. But I tried it, and found that 4 exciters sounds great in a cluster configuration, but also never observed that the FR looked worse compared to when using a single exciter.
I'm not great with physics, but sound travels faster in solid materials than in air, hence the phase change per distance will decrease. I have the exciters a few cm apart, lets say 5cm and assuming a speed of sound in EPS is similar to balsa wood at 5000m/s, you get a wavelength of 5 cm at 100000 Hz. So at half of that, 50000 Hz, you will get maximum cancellation. So well outside the range of both what the exciter can reproduce and what we can hear. You might still get some effects down in the hearing range, but if you instead have two separate plates, in my case that would mean 35 cm distance between sources, and with speed of sound through air that would mean maximum cancellation at around 500 Hz, smack bang in a critical range in our hearing.
However, DML will not produce a phase coherent signal. Instead of a coherent wave front spreading from a source, you have a complex web of waves with different phases coming from many different sources, the nodes and anti-nodes appearing on the plate, generating the same frequency from multiple points at the same time with but with different phase from each other. This makes it behave very differently than regular speakers when it comes to things like how we perceive also the relative phase between multiple sources or reflections.
However, in the end the theory doesn't really matter, only the results. I think DML should sound really bad in theory, and I was a bit concerned that multiple exciters would be a bad idea initially. But I tried it, and found that 4 exciters sounds great in a cluster configuration, but also never observed that the FR looked worse compared to when using a single exciter.
Last edited:
This is exactly correct.
The whole issue of supposed "comb filtering" on DML panel boils down to two things:
1. Speed of sound in the substrate. Anybody can do the very basic sums to confirm that IF this phenomenon actually did exist on DMLs then it would manifest at much much higher, inaudible frequencies, than in pistons.
2. Lack of coherent wave front.
This blows the whole issue right out of the water. THERE'S NO COHERENT WAVEFRONT!
I see mention made of physical principles, and that's exactly the point. If one refuses to understand the underlying science behind DML then, just like any other layman, one would erroneously assume they are the same as cones simply because the make sound.
I see now that you explained how the speed of substrate affects where comb filtering happens a couple of posts ago, but more clearly
I often feel there is a bit of disconnect between what theory tells us about how we should perceive the signal and what we get with DML.
The reason is probably that our hearing is not an objective microphone. We do not hear the actual waveforms, but a version that is modified so suit our needs. For example, our FR is optimized around the mids, extra much so at low volume, since that helps us hear what is important for communication and survival.
Psychoacoustic effects like that are really prominent also when it comes to how we handle phase correlation and spatial information. Our ears are shaped to give phase inconsistencies, resulting in comb filtering, which then our brains can parse to provide us with information about the location a sound is coming from.
So only looking at signal theory and acoustics will not give you the complete picture, and our hearing is made to handle phase incoherent signals, since that is how sound is emitted in nature, so we are doing a lot of processing that is "designed" to handle exactly the type of signal a DML emits. A regular piston speaker on the other hand will expose our hearing to phase coherent signals which we are not really prepared to handle. That will make multiple sources or reflections much more obvious and disturbing since it confuses our hearing.
Yes
I would say no... The speed of sound is not in my understanding in the game. The reason is exciters are driven by the same signal so their coils are in phase. It is in the air that the waves are combined. So the player should be the speed of sound in the air. But here, I reach the limits of my understanding of the topic because I have no idea of what happen in the area of the voice coils in this case.
Yes, we might expect that.
Hmm... It is when comparing the signal in different directions that the phase differences appear (I haven't tried to get evidence of that... should be possible as I have somewhere FR of panels in different directions (the FR are very similar, I might have a look to the IR). So to come back to the topic, 2 panels side by side should show the same web of phase as they have the same dimensions, exciters, design...
Trying to explain the results by the theory of the moment is a way to improve the design or correct the theory. A way of progress.
I posted about that. Same conclusion about the FR : no effect on a FR with a very long observation time (time window)... but using the time dependent function from REW, differences appeared that I haven't be able to explain neither to understand if they have an impact. If I have the opportunity, I would try to check if I can see a difference in the FR according to the direction (directionality?). If there is one, the impact according the volume of the space we want the sound (in house, large room with public, open space...) might be different.
To come back to the example from Audiofranzy with 2 tweeter : in the axis there is no consequence. The problem occurs out of axis when the speaker network starts creating a loss of HF.
Christian
If we are talking about comb filtering in the air, at a particular listening position, I completely agree.
But some here appear to be talking about comb filtering at positions on the panel instead.
Whether or not such happens in practice, or is a bad thing if it does, I will not try to address, except to say that I'm skeptical of both. However, if indeed what we (or at least some of us) are talking about is comb filtering on the panel, caused by waves emanating from two exciters located on the same panel, then I think that the speed of sound in the panel is indeed relevant.
But I fear some are confused about which "speed of sound" they should be thinking about in this case. The most common meaning of "speed of sound" in a solid, is the speed of longitudinal waves.
This speed is a constant for a material, independent of the panel thickness and wave frequency, and largely irrelevant, because these wave produce virtually no significant displacement of the air and hence produce no significant sound.
The relevant waves, on the other hand, are transverse bending/shear waves which are the waves that actually produce lateral/transverse displacement of the panel and hence produce sound. The speed of these waves are given by these relations:
This speed is different from the longitudinal wave speed in that it depends on the panel thickness (h) and frequency of the wave. Further, as shown in this graph below by Hambric, this speed is bounded at high frequencies by the speed of shear waves and is hence always less than the speed of longitudinal waves, especially at lower frequencies. Further, due to the thickness effect in the equation above, for thin panels (like most DMLs) the speed of the transverse waves is typically far, far below that of the longitudinal waves.
So we should all be careful of which "speed" we are referring to when we are talking about the speed of waves on a panel. It's the speed of bending/shear waves that is relevant, and not the commonly referred to "speed of sound" that applies only to the irrelevant longitudinal waves.
Eric
Steve, I had a major restart on my computer and when I reopened this site, it landed on page 478 and your post #9599. Evo-stik is very expensive. Would a different glue work as well? By the way, how did it turn out? I went ahead 3 or 4 pages and did not see where you said if it worked well or not.
Last edited:
Maybe it would help if some knew what audio comb filtering is
https://blogs.qsc.com/live-sound/what-is-comb-filtering-and-how-to-avoid-it/
However lets get shut of one misleading common comment. This is the spec of an AR-LST. Widely used in serious listening conditions, opera houses and all sorts. Pictures grill off can be found on the web.
Power Handling: 64W
Crossover Frequency: 575, 5000Hz
Impedance: 4Ω
Sensitivity: 89.5dB
Bass: 1 x 12"
Midrange: 4 x 1.5" hemispherical
Tweeter: 4 x 0.75" hemispherical
Dimensions: 689 x 508 x 248mm
Weight: 40.5kg
They key to the whole thing is that comb filtering is caused by phase shift. Just where does that cause problem when a speaker such as above where all are driven by the same signal? Actually in practice the speakers should be designed such that the phase shift in each cross over band is negligible but that is an entirely different subject. The phase of reflections will figure. Any degree of phase shift will have an effect. That's why measurements are generally made in an anechoic chamber.
Wide field set ups - distinct difference. Signals to each are not the same.
https://blogs.qsc.com/live-sound/what-is-comb-filtering-and-how-to-avoid-it/
However lets get shut of one misleading common comment. This is the spec of an AR-LST. Widely used in serious listening conditions, opera houses and all sorts. Pictures grill off can be found on the web.
Specifications
Type: 3 way, 9 driver loudspeaker systemPower Handling: 64W
Crossover Frequency: 575, 5000Hz
Impedance: 4Ω
Sensitivity: 89.5dB
Bass: 1 x 12"
Midrange: 4 x 1.5" hemispherical
Tweeter: 4 x 0.75" hemispherical
Dimensions: 689 x 508 x 248mm
Weight: 40.5kg
They key to the whole thing is that comb filtering is caused by phase shift. Just where does that cause problem when a speaker such as above where all are driven by the same signal? Actually in practice the speakers should be designed such that the phase shift in each cross over band is negligible but that is an entirely different subject. The phase of reflections will figure. Any degree of phase shift will have an effect. That's why measurements are generally made in an anechoic chamber.
Wide field set ups - distinct difference. Signals to each are not the same.
I finally got around to building my rig for impedance testing using REW. It's little crude but seems to work great.
.
One thing I was very interested in was whether or not the peaks in the impedance curve would correspond to the natural frequencies of the panel. The results of some of my very first tests confirm that they absolutely do.
The panel I used in the tests below was an 8" x 37" balsa core with fiberglass skins, attached to a frame using small pieces of 3M indoor mounting tape around the entire perimeter, spaced about 2-3 inches apart. I chose this panel for the test because I know it has very little damping, so the resonance frequencies would be well defined, and because I knew the high aspect ratio panel with a fixed perimeter would have a nice series of roughly evenly spaced natural frequencies.
The curve in blue below is the result of an impulse excitation test (i.e. tap test), which is a method I have described before, and have been using for some time now to evaluate the natural frequencies of my panels. In a "tap test" I strike the panel with a small rubber mallet and monitor the response with a mic placed as close as possible to the surface of the panel. Peaks in the frequency response measured by the mic indicate natural frequencies of the panel. Striking (and mic-ing) different locations on the panel excites different sets of natural frequencies (or modes), so usually it requires a series of several taps at different locations to determine, say, the first 10-20 natural frequencies of the panel. But for high aspect ratio panels like the one I'm using here, the situation is particularly simple, and a single "end tap" reveals nine very clearly defined natural frequencies between about 80 and 450 Hz. (The one frequency not revealed by the "end tap" is the fundamental, which is easily revealed by a "center tap" to be at about 70 Hz)
The curve in orange below is the impedance curve I measured for the same panel, using my new rig. For this test, I used a DA EX25VT-4 exciter. In order to try to excite the same frequencies as I had in the tap test, I placed the exciter close to one end of the panel. (Of course, I wouldn't normally put the exciter there, but for the purposes of this test, it made the most sense). As you can clearly see, the peaks in the impedance curve match exactly with the natural frequencies observed in the tap test. Interestingly, the impedance test shows two additional peaks (500 and 616 Hz) which were not clearly resolved in the tap test, but are in the impedance curve.
The single peak at 30 Hz in the impedance curve (but not the tap test) is apparently related to the natural frequency of the exciter itself (magnet mass and suspension stiffness). The plot below shows the impedance curve for the same exciter, attached instead to a heavy block of metal. That result shows a single impedance peak at a similar frequency. Presumably the slight shift is due to the compliance of the panel and mounting, relative to the fixed metal block.
Eric
.
One thing I was very interested in was whether or not the peaks in the impedance curve would correspond to the natural frequencies of the panel. The results of some of my very first tests confirm that they absolutely do.
The panel I used in the tests below was an 8" x 37" balsa core with fiberglass skins, attached to a frame using small pieces of 3M indoor mounting tape around the entire perimeter, spaced about 2-3 inches apart. I chose this panel for the test because I know it has very little damping, so the resonance frequencies would be well defined, and because I knew the high aspect ratio panel with a fixed perimeter would have a nice series of roughly evenly spaced natural frequencies.
The curve in blue below is the result of an impulse excitation test (i.e. tap test), which is a method I have described before, and have been using for some time now to evaluate the natural frequencies of my panels. In a "tap test" I strike the panel with a small rubber mallet and monitor the response with a mic placed as close as possible to the surface of the panel. Peaks in the frequency response measured by the mic indicate natural frequencies of the panel. Striking (and mic-ing) different locations on the panel excites different sets of natural frequencies (or modes), so usually it requires a series of several taps at different locations to determine, say, the first 10-20 natural frequencies of the panel. But for high aspect ratio panels like the one I'm using here, the situation is particularly simple, and a single "end tap" reveals nine very clearly defined natural frequencies between about 80 and 450 Hz. (The one frequency not revealed by the "end tap" is the fundamental, which is easily revealed by a "center tap" to be at about 70 Hz)
The curve in orange below is the impedance curve I measured for the same panel, using my new rig. For this test, I used a DA EX25VT-4 exciter. In order to try to excite the same frequencies as I had in the tap test, I placed the exciter close to one end of the panel. (Of course, I wouldn't normally put the exciter there, but for the purposes of this test, it made the most sense). As you can clearly see, the peaks in the impedance curve match exactly with the natural frequencies observed in the tap test. Interestingly, the impedance test shows two additional peaks (500 and 616 Hz) which were not clearly resolved in the tap test, but are in the impedance curve.
The single peak at 30 Hz in the impedance curve (but not the tap test) is apparently related to the natural frequency of the exciter itself (magnet mass and suspension stiffness). The plot below shows the impedance curve for the same exciter, attached instead to a heavy block of metal. That result shows a single impedance peak at a similar frequency. Presumably the slight shift is due to the compliance of the panel and mounting, relative to the fixed metal block.
Eric
Thank you for your sharing and reply
https://www.tectonicaudiolabs.com/audio-components/audio-exciters/
I have never looked for Tectonic information before
There are also many exciters on their website
But it's a pity that there are no curves in their specifications for reference.
Then "BCT-3" may not perform well (there is a big drop in 1K~10K)???
So Tectonic seems to be out of sale
And Tectonic's "CLASSIC" exciter looks like made in China???
https://www.daytonaudio.com/images/resources/295-269--dayton-audio-bct-3-spec-sheet.pdf
I've seen Dayton's "BCT-3" before
But China has many different versions similar to "BCT-3"
So I wanted to see if anyone has bought the version with better curve
https://billionsound.en.alibaba.com.../Flat_Speakers.html?spm=a2700.shop_plgr.88.20
I think some of Dayton's products may be made in China
This company has more types of exciters than Dayton and Tectonic
And the price looks good
But they have a minimum purchase quantity of at least 500 pieces...XD
I think some of Dayton's products may be made in China
This company has more types of exciters than Dayton and Tectonic
And the price looks good
But they have a minimum purchase quantity of at least 500 pieces...XD
Too simple a view. Some things are designed and made in China. A lot of stuff is designed elsewhere and then made in China. Stuff is also copied by China and made in China. LOL Cars are an interesting example of that.
Fact is that in many areas they can do state of the art work. Dayton probably design and have stuff made in China as many companies do in many areas.
Left, I used to use one. 4 ohm, 25W I believe.
Unfortunately, I didn't have a measurement mic back then.
It has surprisingly decent bass reproduction. I made a test speaker (canvas + thin plywood) and was shocked with how it reproduces gunfire and explosions in movies. As for mids and highs, it'll get you by (I doubt there's any usable output above 8-9k, judging by my ears).
Beware however, this thing literally jumps when enough power's applied.
Sadly mine died, and they aren't cheap ($15-20/piece).
Last edited:
Was the magnet free or did you have a spine (some mechanical solution to keep the magnet in place)?
There are 2 general recommendations in this thread
- a spine to avoid the magnet jumping and also the bad effect of the gravity on the spider
- a LF filter to limit the bass going into the panel. With a standard panel, this is quite nature as there is generally a bass limitation coming from the membrane size and dimensions; so no problem to loose something. With a canvas, as it can go lower in bass with modest dimension, this is a compromise to be done.