Hi Dave, really enjoying the thread you started about individual HRT's affecting the tonality of the phantom centre image. Good work, food for thought. Likewise, extending a cliche just a bit further, still chewing on the whole min-phase diffraction business.
There's the direct-arrival sound, which arrives first. Then there's everything else, diffracted, reflected, combinations of both, standing waves if enough intervals pass, which are all summed at the detector. Whether or not this mess retains its minimum-phase character is something I'll have to read more about in the literature, make measurements, and do a little more more poking around.
Setting aside directivity - which is a separate discussion - the most obvious difference between OB's and boxes is replacing box coloration with diffraction artifacts. Box coloration is dominated by standing waves in a region where the driver is primarily in the piston-band, so it is readily audible once you're sensitized to it. That's easy to do by simply pressing your ear against the side of the box when music is playing - step back a few feet, and the coloration is still audible. Unfortunately, once this little trick has taught you the sound of the box, you'll never be able to forget it again, diminishing your listening pleasure for the rest of the time you own the speaker.
That doesn't let OB's off the hook. Now we have a new, unfamiliar type of coloration compared to ol' familiar box-sound - and the cute box-trick doesn't work this time. The only way to sensitize the ear might be to make a worst-case OB - a circular disk - and something better, with an irregular lossy edge. Since standing waves sound so different than diffraction, getting to know the subjective qualities of these known defects might be a good idea.
There needs to be a subjective cross-check on the improvement potential of the lossy-baffle and wiggly-edge baffles. It's one thing to get it to measure pretty, it's another to correlate it with different, hopefully improved sound quality. I'm not too sure my old familiar pink-noise would tell me what to listen for - maybe clicks or chirps? Or perhaps a well-known voice recorded with the same 1/2" ACO Pacific condenser microphone that I use for measurements?
There's the direct-arrival sound, which arrives first. Then there's everything else, diffracted, reflected, combinations of both, standing waves if enough intervals pass, which are all summed at the detector. Whether or not this mess retains its minimum-phase character is something I'll have to read more about in the literature, make measurements, and do a little more more poking around.
Setting aside directivity - which is a separate discussion - the most obvious difference between OB's and boxes is replacing box coloration with diffraction artifacts. Box coloration is dominated by standing waves in a region where the driver is primarily in the piston-band, so it is readily audible once you're sensitized to it. That's easy to do by simply pressing your ear against the side of the box when music is playing - step back a few feet, and the coloration is still audible. Unfortunately, once this little trick has taught you the sound of the box, you'll never be able to forget it again, diminishing your listening pleasure for the rest of the time you own the speaker.
That doesn't let OB's off the hook. Now we have a new, unfamiliar type of coloration compared to ol' familiar box-sound - and the cute box-trick doesn't work this time. The only way to sensitize the ear might be to make a worst-case OB - a circular disk - and something better, with an irregular lossy edge. Since standing waves sound so different than diffraction, getting to know the subjective qualities of these known defects might be a good idea.
There needs to be a subjective cross-check on the improvement potential of the lossy-baffle and wiggly-edge baffles. It's one thing to get it to measure pretty, it's another to correlate it with different, hopefully improved sound quality. I'm not too sure my old familiar pink-noise would tell me what to listen for - maybe clicks or chirps? Or perhaps a well-known voice recorded with the same 1/2" ACO Pacific condenser microphone that I use for measurements?
DDF said:
Hi Dave,
Again, this will largely depend on the relative amplitudes. After all, the diffracted signal sources from a baffle edge or the signal wrapping around the baffle of a dipole have considerably different amplitude and phase in comparison to the direct signal. Yet in both cases the result is a minimum phase response on aixs.
Lynn Olson said:
Hi Lynn,
Here is something to consider from a theoretical point of view. For a dipole, it can be shown that if you mount a driver on ANY shaped baffle, at low frequency, where the wave length is significantly greater than the distance to the baffle edge, the baffle can be replaced by a circular baffle of some equivalent radius. Another other thing to consider with regard to diffraction is that if a flat baffle is used, over the frequency range where the front and rear radiation are the same, the front and rear diffracted signals set up secondary dipoles with separation equal to the baffle thickness. So, for example, if the baffle at its narrowest point is such that the path length different between front and rear sources is 1' and the baffle is 1" thick, these secondary dipole sources will be on the order of 20dB below the on axis response. As the frequency rises and symmetry between front and rear radiation form a cone driver is lost, as will occur due the effects of the inverted cone, motor, spider and basket blockage, directionality, etc, these diffraction source will behave differently. Ultimately, in the front hemisphere, edge diffraction at higher frequencies resembles diffraction form a sealed box cabinet edge.
soongsc said:
Hi sonngsc
Your first comment is interesting. I have measured many drivers and have yet to come across a case where the measured response of the driver was not minimum phase over its acceptable operating range. At higher frequencies, well into the breakup region, the response may depart from minimum phase but I haven't seen this behavior below breakup. A speaker system, on the other hand is a different issue. Perhaps you could give the link to the other thread?
I'm not sure I understand your question about cepstral analysis. The cepstrum of a system is directly related to the system frequency response much as the impulse response is. If you know the cepstrum you can obtain the frequency response form it just as you can obtain the frequency response form the impulse response. It's just a matter of the correct processing.
Hi John,john k... said:
Here is the link of the posted test results. Would appreciate any insight that I may have missed.
http://www.diyaudio.com/forums/showthread.php?postid=1227789#post1227789
In an example, from a group of people talking, can cepstral analysis be used to extract and reproduce the voice of a single person as if he/she were talking alone? I think not. But I think it's possible to extract certain characteristic of the conversation to identify what is being said by a specific person.john k... said:
In an example of a room with one signal source and one mic. If we know that the signal source is perfectly flat within the frequency range of interest, then we can extract the room mode/reflection response. If we know exactly what the room mode/reflection response is, then we can extract the response of a non-ideal speaker driver. If both are unknown, and we do Cepstral analysis, we can only identify certain characteristis that could belong to one or the other, but how can we specifically come up with a response that correctly follows the response of one or the other.
soongsc said:
I looked at you SPL plots and frankly, I would be suspicious of your phase data. Looking at the characteristis of the phase data it appears that you may be able to remove the observed differences by added a pure time delay. I wonder if this isn't related to the pulse delay problems you have been having with SE and your sound card?
Back to the cepstral analysis, the idea is that we know that the reflections won't appear before an easily determined delay. Same with an impulse response. Supposedly it is easier to edit out the reflections from the cepstrum than the impulse response. However, work I am doing currently suggests otherwise. Still, if you know something about the intended low frequency tuning of the speaker then I believe either cepstral editing or direct editing of the impulse response can give a good idea of the anechoic response. At the same time, I'm not 100% convinced (not even 50%) that cepstral editing is superior, in even the best cases, to merging near field and far field measurements.
This is the sound card without any interchannel delays. I think you may have heard about my other experiences with other cards. During these series of tests, the mic was never moved, the driver was never moved (all pattern changes were done with the driver fixed), the window beginning distance never changed. If you look at the data and explanations closely, you can see that the phase lead increases as the rings are added inward. I'm sure if one reads the data more closely there are other trends that can be observed. In fact, if someone actually tried to change the phase just by shifting the beginning of the window, one would find that such small changes cannot be acheived with a sample rate of 96KHz. The sample rate would probably have to be at least four times this value. Here is the link where I try to ovelap the charts.john k... said:
http://www.diyaudio.com/forums/showthread.php?postid=1229981#post1229981
On the Cepstral analysis, I think it finally needs to be mathematically processed rather than manually edited. It should be something like a calibration file, and I think it just might be possible. This way, once you are able to obtain the room characteristics, it cant be subtracted from the total measurement to get the pure driver data.
john k... said:
Yes, exactly. This is an area where cone speakers behave rather differently than electrostats - front-to-rear symmetry is lost at higher frequencies. That alone is an excellent reason to use a moderately low crossover for the dipole cone driver, to keep it out of the asymmetric region. For this project we're contemplating 1~2 kHz, wavelengths in the 7~14 inch range, where asymmetry is just starting to appear (and is obviously much worse at higher frequencies).
What has your experience been with MMT's, where the lower mid has a crossover about the half the frequency of the upper driver? Does this 2.5 shading technique work, or does it sound (and measure) oddly?
I ask partly because the awkwardness of working with MTM's, with their tricky vertical polar patterns. I've never been completely happy with the sonic portrayal of MTM's - there's a subtly unnatural quality in the midrange, and I put that down to ripples in the power response vs direct-response curves.
Lynn, don't know if you saw this but 96dB/watt and 14mm throw...and I think you'll only need one per baffle
http://www.hempacoustics.com/Tsuzureko_25_-_Hemp_Spec_Sheet_ready.pdf
http://www.hempacoustics.com/Tsuzureko_25_-_Hemp_Spec_Sheet_ready.pdf
Hmm the all-hemp speaker system won't die! anyone ask for the response curve? Sometimes those huge speakers aren't very cost effective though, but maybe the hemp cone doesn't have the screwy upper bass response that a lot of them do..On the other hand, probably that's why they aren't showing it.. Still, a 25" woofer, who wouldn't want one of those!
Hi
For those not so familiar with the 21W54 some pictures that give some insights to a roughly two decades old speaker
What can be seen is that the dust cap is inverse and much bigger than VC. It was glued to the membrane AND the voice coil with very different glues. For the VC the glue is still elastic in the same way as I remember the LX2 sealing recommended to glue the speaker to the baffle. For the outer diameter a brittle glue was used that is much like the one made from bones which is used for classic instruments.
There is quite a lot of Xmax (around 1 cm each direction ). When the VC bottoms there is only 3-4 mm space to the dust cap. All the air behind the dust cap has to be blown through the hole in the inner pole piece ( nothing I would like to have for OB's ) – no additional path due to ferrofluid.
The pole piece has a massive chopper shortening ring and a felt ring to dampen resonance's behind the dust cap.
John, I didn't understand completely.
Where is it that the 1" baffle thickness establishes the secondary dipole? At the baffle edge - if it is 1" there and sharp edged ?
Greetings
Michael
For those not so familiar with the 21W54 some pictures that give some insights to a roughly two decades old speaker
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
What can be seen is that the dust cap is inverse and much bigger than VC. It was glued to the membrane AND the voice coil with very different glues. For the VC the glue is still elastic in the same way as I remember the LX2 sealing recommended to glue the speaker to the baffle. For the outer diameter a brittle glue was used that is much like the one made from bones which is used for classic instruments.
There is quite a lot of Xmax (around 1 cm each direction ). When the VC bottoms there is only 3-4 mm space to the dust cap. All the air behind the dust cap has to be blown through the hole in the inner pole piece ( nothing I would like to have for OB's ) – no additional path due to ferrofluid.
The pole piece has a massive chopper shortening ring and a felt ring to dampen resonance's behind the dust cap.
John, I didn't understand completely.
Where is it that the 1" baffle thickness establishes the secondary dipole? At the baffle edge - if it is 1" there and sharp edged ?
Greetings
Michael
Lynn Olson said:
I have only worked with MMT's in a "normal" speaker I designed for E-Speakers.
All my NaO dipole diesgns are all MTM.
The power response works out very differently with an MTM dipole. In general, for a single driver on an open baffle, as the frequency rises towards the dipole peak the power response starts to increase before it drops off due to the increasing driver directionaliy.
An externally hosted image should be here but it was not working when we last tested it.
The cancelation between the mids in an MTM format can therefore result in more uniform power response. Of course all this is heavily dependent on driver spacing, crossover point and baffle dimensions. (The blue line in the plot is the power response of two uncorrelated sources with the same directional characteristics for reference).
Hi
John, how is this happening, that you get a roughly 8dB INCREASE from decorrelated sources ?
What is cancelling here when correlated?
The only simulations I can do is with EV's ArraySHOW ( freeware ) witch is NOT exactly drivers in an open baffle but rather ommnidirectional sources arranged as dipoles ( as suggested by bjorno from post 732 up )
The upper pictures show the directivity from in phase operation and the lower one the out of phase operation with two sources at 30 cm distance in the vertical direction at 1600 Hz.
How did you do your simulation or was it a measurement you made ?
Greetings
Michael
John, how is this happening, that you get a roughly 8dB INCREASE from decorrelated sources ?
What is cancelling here when correlated?
The only simulations I can do is with EV's ArraySHOW ( freeware ) witch is NOT exactly drivers in an open baffle but rather ommnidirectional sources arranged as dipoles ( as suggested by bjorno from post 732 up )
The upper pictures show the directivity from in phase operation and the lower one the out of phase operation with two sources at 30 cm distance in the vertical direction at 1600 Hz.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
How did you do your simulation or was it a measurement you made ?
Greetings
Michael
soongsc said:
I have no argument that the phase does change. What I am suggesting is that it appears that the manner in which the phase changes is the result of a change in the linear phase component. That is, since you were careful not to change the measurement distance or window, then perhaps the explanation is that some how the AC is shifted by you cone mods. I would look at the data using the HBT and see if you can bring the different phase responses into closer agreement by adding or removing delay.
Looking at the overlaid plot: http://www.diyaudio.com/forums/showthread.php?postid=1229981#post1229981
If I look at curve 2, clean, and curve 5, enabled 3, what I see is that at 20 K the phase difference is about 6 divisions. At 10 K it is about 3, at 5 K it's about 1.5. That suggests a differences in the linear phase component.
Looking at the 6 divisions at 20k point, 6 divisions is 120 degrees or 1/3 of a cycle at 20k, or 0.01666 msec, which is less than a sample length at 48k sampling.
Re: Yust gotta know.
IF the baffle edge is a knife edge with 0 thickness, and IF the front and rear responses are symmetric then the diffractions sources from the front and rear will be identical but 180 degrees out of phase and they will cancel.
I wrote the software myself. I use directional or omnidirectional sources and numerically integrate the intensity over the surface of a sphere enclosing the sources. This is the definition of radiated power. You can get the basic idea here: http://www.musicanddesign.com/Power.html
Consider the power of a single omni source as the 0dB reference. When correlated we know that two omni sources operating in phase radiate 6 dB greater power than a single source. When uncorrelated they radiate only 3dB more power. A dipole composed of two omni sources, at least 1/2 octave below the dipole peak, has a directivity factor of approximately 3 which means it radiates 4.7dB LESS power than a single omni source with the same on axis response, or 7.7dB less power than TWO uncorrelated sources. As the frequency rises towards the dipole peak the directivity factor decreases. At the dipole peak the directivity factor is 1.414 (the dipole radiation pattern broadens out and fills in) and a dipole radiates only 3dB less power than an omni source with the same on axis response. Above the dipole peak two omni dipole sources behave as two uncorrelated sources. It no longer matters if they are in or out of phase. The radiated power is 3dB more than a single omni source. Hopefully this provides the explanation you need. Directivity complicates the issue which is why I applied numerical integration.
Below are two images I prepared with Array Show. I used 4 omni sources set up to mimic the NaO baffle. Obviously the lack of directionality makes these highly approximate but I think they show the effect I was referring to. The top figure shown the H and V polar responses and the intensity at 200, 400, 800 and 1K Hz. Above that we are getting into the crossover region effects. The lower figure shows the 100 Hz and 1k Hz result back to back. Notice how the intensity plot widens in the horizontal direction at 1K compared to 100 Hz while at the same time it narrows vertically. The broadening represents the increase in power from the dipole as the dipole peak is approached and the narrowing is the cancellation due to the MTM configuration. At the same time there is very good horizontal dispersion while the reduced vertical dispersion is such that there is still very good coverage for listeners sitting on the floor, on a chair, or standing.
Obviously these are highly aproximate figures but hopefully they provide insite into why I feel the MTM dipole is a different animal than the MTM direct radiator format.
Zene Gillette said:
IF the baffle edge is a knife edge with 0 thickness, and IF the front and rear responses are symmetric then the diffractions sources from the front and rear will be identical but 180 degrees out of phase and they will cancel.
mige0 said:
I wrote the software myself. I use directional or omnidirectional sources and numerically integrate the intensity over the surface of a sphere enclosing the sources. This is the definition of radiated power. You can get the basic idea here: http://www.musicanddesign.com/Power.html
Consider the power of a single omni source as the 0dB reference. When correlated we know that two omni sources operating in phase radiate 6 dB greater power than a single source. When uncorrelated they radiate only 3dB more power. A dipole composed of two omni sources, at least 1/2 octave below the dipole peak, has a directivity factor of approximately 3 which means it radiates 4.7dB LESS power than a single omni source with the same on axis response, or 7.7dB less power than TWO uncorrelated sources. As the frequency rises towards the dipole peak the directivity factor decreases. At the dipole peak the directivity factor is 1.414 (the dipole radiation pattern broadens out and fills in) and a dipole radiates only 3dB less power than an omni source with the same on axis response. Above the dipole peak two omni dipole sources behave as two uncorrelated sources. It no longer matters if they are in or out of phase. The radiated power is 3dB more than a single omni source. Hopefully this provides the explanation you need. Directivity complicates the issue which is why I applied numerical integration.
Below are two images I prepared with Array Show. I used 4 omni sources set up to mimic the NaO baffle. Obviously the lack of directionality makes these highly approximate but I think they show the effect I was referring to. The top figure shown the H and V polar responses and the intensity at 200, 400, 800 and 1K Hz. Above that we are getting into the crossover region effects. The lower figure shows the 100 Hz and 1k Hz result back to back. Notice how the intensity plot widens in the horizontal direction at 1K compared to 100 Hz while at the same time it narrows vertically. The broadening represents the increase in power from the dipole as the dipole peak is approached and the narrowing is the cancellation due to the MTM configuration. At the same time there is very good horizontal dispersion while the reduced vertical dispersion is such that there is still very good coverage for listeners sitting on the floor, on a chair, or standing.
Obviously these are highly aproximate figures but hopefully they provide insite into why I feel the MTM dipole is a different animal than the MTM direct radiator format.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
soongsc said:
When you see an in-room cepstral view, the first reflection sticks out like a sore thumb, and your fears would be allayed. The only other possible reason for the cepstral spike is if the transducers are misaligned by 3 to 4 ms, which we'd know about ahead of time.
Cheers,
Dave