Beyond the Ariel - Page 211 - diyAudio
Go Back   Home > Forums > Loudspeakers > Multi-Way

Multi-Way Conventional loudspeakers with crossovers

Please consider donating to help us continue to serve you.

Ads on/off / Custom Title / More PMs / More album space / Advanced printing & mass image saving
Reply
 
Thread Tools Search this Thread
Old 20th September 2007, 02:06 PM   #2101
soongsc is offline soongsc  Taiwan
diyAudio Member
 
soongsc's Avatar
 
Join Date: Mar 2005
Location: Taiwan
Quote:
Originally posted by john k...


That would be fine, but I don't see where this has to do with cepstral editing. If I could generate an impulse sequence that represents the room then it could be subtracted form the basic impulse response.
If the original signal is impulse separated by enough time so that the reflections do not get mixed with the subsequent impulses, then it's fine. But if the original source is MLS type sequence, editing out the tail means also editing out the low frequency decay characteristics.
__________________
Hear the real thing!
  Reply With Quote
Old 20th September 2007, 02:36 PM   #2102
soongsc is offline soongsc  Taiwan
diyAudio Member
 
soongsc's Avatar
 
Join Date: Mar 2005
Location: Taiwan
Quote:
Originally posted by DDF



I haven't been too forthcoming regarding the exact technique, but that's consciously. Fundamentally there is enough info known in perhaps 2 to 3 measures to be able to solve this.

The optimization would occur in the quefrency domain as it better suits the optimization method.

I appreciate we canít achieve closure on this discussion when the algorithm details aren't revealed, nor the background techniques. However, Iím surprised that it can be so readily dismissed and critiqued, without this information.
Knowing the concept, the math wizards can figure it out. Hope we get better tools.
__________________
Hear the real thing!
  Reply With Quote
Old 20th September 2007, 03:01 PM   #2103
gedlee is offline gedlee  United States
diyAudio Member
 
Join Date: Dec 2004
Location: Novi, Michigan
Quote:
Originally posted by DDF


I appreciate we canít achieve closure on this discussion when the algorithm details aren't revealed, nor the background techniques. However, Iím surprised that it can be so readily dismissed and critiqued, without this information.

I am surprised that you want us to take this discussion seriously when you won't tell us what you plan to do.

No disrespect intended, but signal processing is what I do too and what you are claiming just doesn't fit with the math capabilities of Spectral Estimation (the name of what you are talking about). Spectral Estimation is a highly evolved field due to its fundamental usage in sonar.

My paper was in the JAES about 15 years ago.

The bottom line here is that we can reconstruct missing (edited) data almost perfectly IF we know some parameters of the underlying system. But if it is that system is precisely what we are testing then our "estimate" can be no better than our "guess" as to what the undeying system parameters are.

If we know for example that the underlying system is a close box with a given Q and resonant frequency, then I can exactly reconstruct the impulse response, Ceptrum, whatever (linearity assumed of course) well into edited areas with a very high precision (that decreases with number of estimated points). This is done in Praxis and his implimentation is the best that I have seen.

But the point is that we have to make assumptions about the DUT. The better the assumptions the better the test. But how do we know how good our assumptions are? We don't. Hence we don't know how good the estimates (linear prediction whatever you want to call it) is. The techniques will always yield a result and virtually always that result will be better than had we done nothing, but we can never really know how accurate our predicted results actually are.

Of course we could make a prediction and then test that prediction - kind of an iterative adaptive approach, and Eric Benjamin of Dolby Labs has done some work in this area. Very interesting - holds promise -(an AES preprint from 2005, I believe).

But trust me, there is not likely to be something useful that has not been tried yet.
  Reply With Quote
Old 20th September 2007, 03:02 PM   #2104
diyAudio Member
 
john k...'s Avatar
 
Join Date: Aug 2004
Location: US
Quote:
Originally posted by soongsc

But if the original source is MLS type sequence, editing out the tail means also editing out the low frequency decay characteristics.
It would be the same with cepstral editing. And remember that the reflections are basically an impulse chain that applies to all frequencies and all time greater than the delay from source to reflection point.

Look at this image.

Click the image to open in full size.

This is the real cepstrum of a driver measured in a room with the time to the first room reflection just over 2 msec. So after 2 msec please identify what is room reflection and what is direct radiation. It doesn't matter how far out in time the plot is extended. The low frequency information is always contaminated by reflected sound. If the first reflection happens later, say at 6 msec, then the cepstrum and the original impulse would still be contaminated past 6 msec.

The only way to get the room reflections out of either is if you can define the impulse chain of the room. This would need to include all the time delays for 1st, 2nd,...,nth order reflections as well as the characteristic of each reflection (that is the frequency dependent attenuation). For example, a perfectly reflective surface would have a perfect, but delayed impulse. A carpeted floor would have the impulse of maybe a delayed low pass filter. Now assuming that the correct impulse chain can be identified then it could be removed from either the original impulse, or by computing the transfer function of the room in the frequency domain, and then the cepstrum of that, from the cepstrum of the original system. Of course, this will also be dependent on the room, the position of the source in the room and the position of the microphone, the off axis response of the driver.....

All I can say is don't tell me about it, show me! As they say, talk is cheap. I don't mean to sound arrogant or even nasty here, but this isn't a new subject. People have looked at it over the years and without much to show for it. I think there is plenty of incentive too. After all what is the other choice? A $250,000 (or more) anechoic room? Speculate all you like but if you want to convince me show me results.
__________________
John k.... Music and Design NaO Dipole Loudspeakers.
  Reply With Quote
Old 20th September 2007, 03:13 PM   #2105
gedlee is offline gedlee  United States
diyAudio Member
 
Join Date: Dec 2004
Location: Novi, Michigan
Quote:
Originally posted by john k...

People have looked at it over the years and without much to show for it.

John - this is quite true, but I do think that some of the implimentations work quite well. Praxis, as I stated, does a very good job.
  Reply With Quote
Old 20th September 2007, 04:10 PM   #2106
soongsc is offline soongsc  Taiwan
diyAudio Member
 
soongsc's Avatar
 
Join Date: Mar 2005
Location: Taiwan
Quote:
Originally posted by john k...


It would be the same with cepstral editing. And remember that the reflections are basically an impulse chain that applies to all frequencies and all time greater than the delay from source to reflection point.

Look at this image.

Click the image to open in full size.

This is the real cepstrum of a driver measured in a room with the time to the first room reflection just over 2 msec. So after 2 msec please identify what is room reflection and what is direct radiation. It doesn't matter how far out in time the plot is extended. The low frequency information is always contaminated by reflected sound. If the first reflection happens later, say at 6 msec, then the cepstrum and the original impulse would still be contaminated past 6 msec.

The only way to get the room reflections out of either is if you can define the impulse chain of the room. This would need to include all the time delays for 1st, 2nd,...,nth order reflections as well as the characteristic of each reflection (that is the frequency dependent attenuation). For example, a perfectly reflective surface would have a perfect, but delayed impulse. A carpeted floor would have the impulse of maybe a delayed low pass filter. Now assuming that the correct impulse chain can be identified then it could be removed from either the original impulse, or by computing the transfer function of the room in the frequency domain, and then the cepstrum of that, from the cepstrum of the original system. Of course, this will also be dependent on the room, the position of the source in the room and the position of the microphone, the off axis response of the driver.....

All I can say is don't tell me about it, show me! As they say, talk is cheap. I don't mean to sound arrogant or even nasty here, but this isn't a new subject. People have looked at it over the years and without much to show for it. I think there is plenty of incentive too. After all what is the other choice? A $250,000 (or more) anechoic room? Speculate all you like but if you want to convince me show me results.
The one who shows it in a product will be leading the technology.
As described somewhere in the SE manual, the Cepstral functionality is a work in progress.

I think you have already addressed the key issues. Each time one changes measurement envoronment, it's best to redo a room signature capture. Once the method of room signature capture is derived, the other parts are pretty common to most software out there. The question really would be what signal is best to extract room signature?
__________________
Hear the real thing!
  Reply With Quote
Old 20th September 2007, 04:24 PM   #2107
mige0 is offline mige0  Austria
diyAudio Member
 
mige0's Avatar
 
Join Date: Jan 2005
Location: Austria, at a beautiful place right in the heart of the Alps.
Hi

Wouldn't a simple firecracker (or a pinned balloon ) at the position of the speaker do it for a good room capturing ?

Same can be done for RT60 measurement BTW.

How precise could such a speaker measurement become when done in domestic rooms ?
How many octaves down the usual 200-300 Hz ?


Greetings
Michael
  Reply With Quote
Old 20th September 2007, 04:28 PM   #2108
ucla88 is offline ucla88  Tahiti
diyAudio Member
 
Join Date: Dec 2001
Location: rocklin, CA
Quote:
Originally posted by DDF



Hi Michael: last wizzer I measured was 25 years ago, with swept sine, so, sorry, don't know. I strongly suspect it would be min phase as I don't see any physical mechanism in a wizzer which would add non-min phase delay.

Cheers,
Dave
Hi Dave,

The mechanism postulated is that a whizzer cone effectively acts as two different drivers. While each may be min phase, the system might be slightly nonmin phase.

I suppose it's a reasonalbe hypothesis. I've never tested any whizzer cones (not really high on my list) so I don't have any experimental data one way or the other.
__________________
http://www.audioheuristics.org/ aka Mark's Speaker Page
  Reply With Quote
Old 20th September 2007, 07:22 PM   #2109
diyAudio Member
 
Lynn Olson's Avatar
 
Join Date: Jul 2004
Location: Northern Colorado
Default Re: Re: Raven

Quote:
Originally posted by Jaco Pastorius

Hi Lynn Iím from Italy.

A friend of mine (another fanatic like diyAudio title says ) has a great experience with Eighteen Sound Drivers.

He obviously has contacts with the company.

If you need information we will be proud to help you.

Iím a fanatic too, no business interest in my proposal!

Ciao
I've sent an e-mail to Mr. Previ, we've made contact, and I'm awaiting a reply. He's probably talking to the technical staff as we speak. Additional contacts within 18Sound are welcome, though, and feel free to point them to the most recent posts on this thread.

The 8-inch spot looks good for the 8NMB420 - an additional 3 dB of efficiency would be welcome, along with retaining the neodymium magnet structure. Any data on impulse, CSD, and swept distortion would be most welcome.

The 12-inch spot looks good for the 12NDA520, with the 12ND710 as a second choice. The efficiency is good as it is, and like the 8NMB420, any data on impulse, CSD, and swept distortion is welcome.

The 18-inch spot looks problematic for 18Sound drivers, due to the very large 1.5 kHz peaks. I know these are intended for professional subwoofer use, and pros use 24 dB/octave LR active crossovers and EQ. I won't be using such steep crossover, and want something more like the conventional woofer response of the RCF L18S800 or L18P300, or better yet, the response of the 18Sound 15NMB420. The rough response of 18-inch drivers is what makes me think of using a pair of 12-inch drivers side-by-side instead of single 18-inch driver.

The system will NOT be using 24 dB/octave LR crossovers, so smooth out-of-band performance is a requirement, along with low distortion in the desired frequency range, which is 60 Hz ~ 2 kHz for the 8-inch, 60 Hz ~ 300 Hz for the 12-inch, and 60 Hz ~ 150 Hz for the 15 or 18-inch.


Quote:
Originally posted by john k...


Correct me if I misread this and your earlier post on aligning the drivers for first arrival but that sounds like what Spica did back in the late 70's. There are a couple of caveats here though. Considering the 2.5k crossover, the group delay of the 2nd order Bessel would be about 0.1 msec so the tweeter would have to be offset about 1.35" behind the acoustic center of the 8". No big deal. Then the 250 Hz crossover would have a GD of 1.0 msec and the 8"/tweeter combo would have to be offset 13.5" behind the 12" AC. Now there's a bit of a problem, no? Lastly, for the 100 Hz x-o the GD is 2.2 msec requiring an offset of almost 33" relative to the 12"/8"/tweeter combination. These offsets are in addition to the offsets of the driver ACs.

The Spica approach was fine for a small 2-way box speaker with crossover in the 2 to 3 k range, but using physical offset to time align 3 or 4 way systems isn't the way to go, IMO. Additionally, with a dipole system if you physically offset the drivers to align the 1st arrivals from the front you make a mess out of the rear response.

My feelinng is if you want time coherence in a dipole the driver AC's must be aligned and then time coherent crossover must be used as I am doing in my ICTA design development.

http://www.musicanddesign.com/icta_cross.html
The problem with the ICTA is the large inter-driver phase angle, as indirectly indicated by the magnitude graph - if my eyes don't deceive me, I'm guessing an inter-driver angle of about 120~135 degrees, right? As an old-time quadraphonics guy, phase angles this large are directly audible in their own right as a "phasey" sensation. although not everyone is sensitive to this. Inter-driver phase angles of 90 degrees or less seem more acceptable, but this depends on auditioner thresholds.

This is not the same as midrange lobing, which sounds quite different. I'm talking about a lack-of-cohesion sensation that is audible at any distance, and in fact gets worse as you approach the loudspeaker. To really create a sensation of a single driver, I've found I have to get the phase angles (between drivers) to 30~40 degrees or less. All subjective here, but that's how I hear things. I am probably more sensitive than most since I worked for many years on SQ and QS matrix decoders that intentionally created a 90-degree phase difference between front and rear speakers.

Now this phase-sensitivity between drivers mostly applies for equal magnitudes - reduce the magnitude of one driver by 3 dB or more, the sensitivity drops off pretty fast, and at 10 dB difference, the "acceptable" phase angles open up to 120~140 degrees. I've played with this a fair amount, listening to pink-noise and music at different distances, and have come up with a rough sort of "acceptable" phase angles between drivers.

Oddly enough, the MTM mounting scheme isn't necessarily an improvement - the "incoherent" sensation becomes less acceptable over a large area and with more drivers. That's what I especially dislike about large vertical WMTMW or line arrays - a strong sense of incoherence, of things not working together, and not like real musical instruments at all.

I agree completely about the need for precision in the mid/HF region. Absolutely. That's why the tweeter needs to be user-adjustable for listening height and distance. Even small adjustments produce large subjective changes in image focus, shape of soundstage, and realism in vocal quality. This requires a combination of acoustic measurement and subjective "trimming-in" - probably in the mm range for the final subjective adjustments.

I also agree about the concerns of the LF region - if these were conventional crossovers, switching from one driver to the next, stepping down from 8 to 12 to 18, then both the overall phase response and inter-driver phase angles become a major concern. If phase distortion is audible anywhere, it's audible in the 100 Hz to 1 kHz region.

But in my experience - which all I'm qualified to talk about - INTER-driver phase is much more audible than anything else, and it's not an artifact of polar patterns, but directly audible as "lack of coherence" and an artificial-sounding disjointed "multidriver system" sound. The opposite of a single-driver system sense of wholeness and coherence, in other words.

You walk towards the speaker system, and everything falls apart sonically - the illusion of a single source is destroyed. Nearly all physically large, complex, multiway high-end systems sound like this - the sound is incoherent and artificial, even if they "measure well" and fit the modern Toole criteria for "good sound". The reason for the incoherence comes down to large inter-driver phase angles, driver separation, and narrowband phase shifts close to the crossover region. None of this shows up in conventional FR measurements, impulse response, or the CSD.

That's a subtle reason I'm keeping the bass drivers all working together in the bass region, with only half-crossovers. I want to keep the inter-driver phase angles low and controlled - thus the choice of 1st-order or low-Q Bessel lowpass filtering for the lowpass functions for the 12 and 18-inch drivers. I don't want the angles to exceed 60 degrees, and preferably a lot less, so there's going to be some careful trimming of crossover behaviour in this region.

It's also why I don't want narrowband response deviations in the mutual-response region, since these cause sharp phase deviations as well, and will destroy the sense of coherence. This is part of the reason for my concern about the 18-inch driver - because of inter-driver phase concerns, a conventional sharp crossover is not an option. At higher frequencies, all of the drivers will collectively share an additional sharper lowpass filter at the tweeter crossover, to get rid of them once and for all above 2 kHz. But all of the drivers have to be reasonably well-behaved below that - which just may be asking too much of an 18-incher, but not a 15-incher.

The subjective goal is to have all three cone drivers sound like one driver, at any listening distance, including no more than few inches away from the front panel. My speakers have always done this, and the new system must do this as well. Nearly every other subjective goal is secondary.

Now will it succeed or not - well, I don't know yet. It is the direction I'm aiming for.
  Reply With Quote
Old 20th September 2007, 07:44 PM   #2110
DDF is offline DDF  Canada
diyAudio Member
 
DDF's Avatar
 
Join Date: Sep 2004
Location: Ottawa
Quote:
Originally posted by gedlee



I am surprised that you want us to take this discussion seriously when you won't tell us what you plan to doÖ..
I donít want to give away the idea.

Thereís also a thread in these boards, where an author is taking the same somewhat secretive approach with his HOM based loudspeaker system. So why be surprised when I feel less than inclined to give away all the cookies myself, without full discosure?

My concept is different than Benjaminís work, if youíre meaning his technique of using low frequency pre-emphasis and de-emphasis pre and post FFT. Good idea, and it can be applied independently and in concert with what Iím investigating.

In any estimation technique, a reference is required. A suitable reference can be formed here with a couple additional measurements, some basic assumptions about the acoustical properties of the room, and by intelligently bounding the convergence algorithm, in parameters adapted, and boundaries for them, along with degrees of freedom alowed for adjustment.

Iíll also say it one more time, I heard the same nay-saying when developing an ecan verification methodology. It was successful enough that a group audio terminal company we were in talks with to partner, threatened to sue us as they believed we cracked their f/w code. All from measurements people said couldnít be done.

I may have some spare time coming up to investigate this further (Iím not retired, I donít work in audio any longer and I have 3 young children), but this is my last post describing details of the technique.

Iím hoping we can progress this to a more respectful tone. Skepticism is warranted, outright closed minded dismissal isnít.

Without my oferring more detail, maybe there's not much else to discuss on this topic.
  Reply With Quote

Reply


Hide this!Advertise here!
Thread Tools Search this Thread
Search this Thread:

Advanced Search

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off



New To Site? Need Help?

All times are GMT. The time now is 06:23 PM.


vBulletin Optimisation provided by vB Optimise (Pro) - vBulletin Mods & Addons Copyright © 2014 DragonByte Technologies Ltd.
Copyright ©1999-2014 diyAudio

Content Relevant URLs by vBSEO 3.3.2