Full range drivers have a dispersion that narrows without the abrupt change that can happen with multi-way speakers at crossover points, so the "room response" seems likely to sound more natural. Plus, with passive crossovers there may be a significant mismatch in the phase response between the 2 speakers at the crossover points, which would damage imaging in the upper midrange. Above about 1kHZ we sense image location primarily by amplitude comparison, so any artificial deviation of amplitude balance in that frequency region (1kHZ - 6kHZ) would be damaging to imaging.
Lately I've been tri-amping with 4 pole active xovers, and using the Peerless (Tympany) TG9FD1008 three inch driver in a well damped tubular sub-enclosure to handle from 500HZ - 7kHZ, so I can keep crossover point anomolies out of the critical imaging frequencies. I haven't done any kind of fair comparison, but it sure does sound good.
Can we start calling this inter-aural comb filtering.. IA combing?
Thanks for directing me to the "new" shuffler you are trying. (ShuffleXTC.zip) I will give that a whirl.
I still don't understand what was asymmetrical left/right about the rephase shufflers though... I could not hear the difference on hard panned left/right material from the unprocessed versions. (ie. nancy sinatra "bang bang"). The effects were only apparent in centrally-mixed material, as I would expect. How then could it be different on one side from the other? Do you mean your physical listening position moving left-right? Are you saying that reversing left-right of the shuffler makes a difference?
It's a bit more complicated. Hard panned signals will sound the same, as there is no direct counter part to comb with at the ear.
But in the center, you first hear the imaging queues in both ears before that sound from each speaker reaches the opposite ear.
The way the phase was shuffled, one ear partly cancels the combing because the timing has changed of the sounds reaching your ear, the other ear has a bigger time error as a consequence. So while one side improves, the other gets something close to the original combing. You can see the different patterns in my example as presented in the early waterfall plots.
While imaging remains very close to the same on my phase aligned system I did get tonal differences I could not solve. Apparently the direction and distance are figured out by our ears in a short time span. Tonality is determined over a longer period, Averaged over a longer time span.
I did notice some difference in the exact placement of the center by switching the phase bumps. But I could not tell which side got the better IA combing canceling. Just an imbalance.
This new XTC-shuffler manipulates the signal over time and works exactly the same on both sides. I've changed a few things around in my processing and did not give this one a fair shot yet. I've had some fun with it for about an hour, but need way more to draw any conclusion. Its probably a good idea to convolve a few songs offline and mix them with originals in a play list (keeping volume level exactly the same) to see if I can pick them out that way.
But in the center, you first hear the imaging queues in both ears before that sound from each speaker reaches the opposite ear.
The way the phase was shuffled, one ear partly cancels the combing because the timing has changed of the sounds reaching your ear, the other ear has a bigger time error as a consequence. So while one side improves, the other gets something close to the original combing. You can see the different patterns in my example as presented in the early waterfall plots.
While imaging remains very close to the same on my phase aligned system I did get tonal differences I could not solve. Apparently the direction and distance are figured out by our ears in a short time span. Tonality is determined over a longer period, Averaged over a longer time span.
I did notice some difference in the exact placement of the center by switching the phase bumps. But I could not tell which side got the better IA combing canceling. Just an imbalance.
This new XTC-shuffler manipulates the signal over time and works exactly the same on both sides. I've changed a few things around in my processing and did not give this one a fair shot yet. I've had some fun with it for about an hour, but need way more to draw any conclusion. Its probably a good idea to convolve a few songs offline and mix them with originals in a play list (keeping volume level exactly the same) to see if I can pick them out that way.
With regard to: "Comb filtering in a stereo setup is very real. I bet any step we can achieve to reduce that combing helps to clear up that phantom center."
It seems analogous to what happens with room reflections; where if there's going to be ANY reflections at all, then the best thing to do is to have many other reflections with different delays, so they can all largely fill in each others cancellations. In the case of interaural crosstalk, which I think you're dealing with here, there is probably a strategic ratio of the delays that would give best results. If you could largely separate out the center image signal (L+R) from the rest, and only process it, then mix it back in, that's what I would try.
It seems analogous to what happens with room reflections; where if there's going to be ANY reflections at all, then the best thing to do is to have many other reflections with different delays, so they can all largely fill in each others cancellations. In the case of interaural crosstalk, which I think you're dealing with here, there is probably a strategic ratio of the delays that would give best results. If you could largely separate out the center image signal (L+R) from the rest, and only process it, then mix it back in, that's what I would try.
So maybe using something like a MiniDSP 2x4 to add some delay and band pass filter the L+R signal.
Bob, F1 FAN,
That's what I experimented with and posted about a few pages back. And started to do that again late last week. Though I used JRiver and a couple of spare channels.
I've played around with Choueiri's IR and that one basically does the same thing. But it has a fixed EQ and isn't as clean in early waterfall plots as I'd like.
So I decided to make my own using lots of (linear) EQ, a delayed band passed (L+R) signal and lots of fooling around between JRiver and REW.
Here's a comparison:
I'll show early waterfall plots of my newest shuffler (I haven't tried this one "live" yet, but a raw pré version showed promise enough to spend some time on it)
First the pure dirac case (as any Stereo setup would exhibit):
Early waterfall of direct signal to ear:
(0.1 ms rise time)
(0.2 ms rise time)
(0.3 ms rise time)
After which we'd get the inter aural comb pattern (0.270 ms has passed)
(0.1 ms rise time of L + 0.270 ms delayed R)
(0.2 ms rise time of L + 0.270 ms delayed R)
(0.3 ms rise time of L + 0.270 ms delayed R)
(0.4 ms rise time of L + 0.270 ms delayed R)
After a lot of work and searching variables I came up with this:
Early waterfall of direct signal to ear:
(0.1 ms rise time)
(0.2 ms rise time)
(0.3 ms rise time)
After which we'd get the inter aural comb pattern (0.270 ms has passed)
(0.1 ms rise time of L + 0.270 ms delayed R)
(0.2 ms rise time of L + 0.270 ms delayed R)
(0.3 ms rise time of L + 0.270 ms delayed R)
(0.4 ms rise time of L + 0.270 ms delayed R)
With a bit more time I might be able to get an even flatter response at the ear, though I think this should do fine for preliminary testing. It seems to work just fine to get that longer "time frame" (for our brain) of the stereo signal at the ear with relatively flat FR balance.
I can translate this to a FIR filter if anyone wants to run some tests.
This one has some extra (very mild) EQ as it's my center signal from the mid/side processing chain. I still need to check/update my side signal to match.
That's what I experimented with and posted about a few pages back. And started to do that again late last week. Though I used JRiver and a couple of spare channels.
I've played around with Choueiri's IR and that one basically does the same thing. But it has a fixed EQ and isn't as clean in early waterfall plots as I'd like.
So I decided to make my own using lots of (linear) EQ, a delayed band passed (L+R) signal and lots of fooling around between JRiver and REW.
Here's a comparison:
I'll show early waterfall plots of my newest shuffler (I haven't tried this one "live" yet, but a raw pré version showed promise enough to spend some time on it)
First the pure dirac case (as any Stereo setup would exhibit):
Early waterfall of direct signal to ear:
(0.1 ms rise time)
(0.2 ms rise time)
(0.3 ms rise time)
After which we'd get the inter aural comb pattern (0.270 ms has passed)
(0.1 ms rise time of L + 0.270 ms delayed R)
(0.2 ms rise time of L + 0.270 ms delayed R)
(0.3 ms rise time of L + 0.270 ms delayed R)
(0.4 ms rise time of L + 0.270 ms delayed R)
After a lot of work and searching variables I came up with this:
Early waterfall of direct signal to ear:
(0.1 ms rise time)
(0.2 ms rise time)
(0.3 ms rise time)
After which we'd get the inter aural comb pattern (0.270 ms has passed)
(0.1 ms rise time of L + 0.270 ms delayed R)
(0.2 ms rise time of L + 0.270 ms delayed R)
(0.3 ms rise time of L + 0.270 ms delayed R)
(0.4 ms rise time of L + 0.270 ms delayed R)
With a bit more time I might be able to get an even flatter response at the ear, though I think this should do fine for preliminary testing. It seems to work just fine to get that longer "time frame" (for our brain) of the stereo signal at the ear with relatively flat FR balance.
I can translate this to a FIR filter if anyone wants to run some tests.
This one has some extra (very mild) EQ as it's my center signal from the mid/side processing chain. I still need to check/update my side signal to match.
Magnepan have been successfully demonstrating a 3ch stereo setup the last few years. Bryston/Magnepan SP3 - The Audio Beat - www.TheAudioBeat.com.
They have been using a Bryston pre/pro but the specifics as to how they have it set up are not clear.
They have been using a Bryston pre/pro but the specifics as to how they have it set up are not clear.
Last edited:
It may be obvious, but personally I'd like to avoid the center channel. From what I've heard and experimented with I'm pretty sure we can do better than standard stereo (by using 2 speakers) with a little help.
(I'll not count the 2 ambient rear/side channels as they only provide me with a virtual room that's better than the actual room I'm in
)
(I'll not count the 2 ambient rear/side channels as they only provide me with a virtual room that's better than the actual room I'm in
)
Or one channel of the original shuffler on the L+R signal.
- Derive L+R signal
- Convolve that signal with the multi-tap impulse
- Mix signal back into stereo signals
This is what I've been saying (without the word "convolve" - not sure what that means exactly). If a waterfall plot is made using a stereo head mic, the waterfall should show the IA crosstalk cancellations (has anybody ever done that?). It seems that the only thing that will minimize their damage is adding in more "reflections", strategically, so most (hopefully all) cancellations get filled in by the presence of these additional "reflections" (electronic delays), each of which will have different cancellation frequencies due to the different delay times. If this could be done to only the L+R content, which would then be mixed back in with the L and R stereo effect content, the in between imaging might not suffer that much.
Wow, nice work on the graphs!!!!!!
Lessening side wall reflection helps make a better center image.
Maybe past 2khz anyway.
Seems that's what I heard last night having the movie "dark city" playing in the background.
My rat shack monopole linaeum tweeted center channels work better as mtm d'apolito than sideways when used as a stereo pair.
The ribbon used as a center channel (widest dimension left and right) it has wide dispersion but narrow vertical.
I like that mMG 3 speaker setup.
Lol, it would block my TV though !!!!!!
I bet 90% of people use their system setup around the tv, maybe not 7as many on these forums though.
Maybe you can take a 10 band eq, y some of the other outputs together to make a mono signal that can got out to another amp. But multiple volume controls gets to be a pain.
Norman
Lessening side wall reflection helps make a better center image.
Maybe past 2khz anyway.
Seems that's what I heard last night having the movie "dark city" playing in the background.
My rat shack monopole linaeum tweeted center channels work better as mtm d'apolito than sideways when used as a stereo pair.
The ribbon used as a center channel (widest dimension left and right) it has wide dispersion but narrow vertical.
I like that mMG 3 speaker setup.
Lol, it would block my TV though !!!!!!
I bet 90% of people use their system setup around the tv, maybe not 7as many on these forums though.
Maybe you can take a 10 band eq, y some of the other outputs together to make a mono signal that can got out to another amp. But multiple volume controls gets to be a pain.
Norman
Last edited:
I mentioned my hack-edit of the Choueiri XTC IR (posted earlier) being less clean in an early waterfall...
Why not let the graphs speak for themselves:
(0.3 ms rise time)
versus my new attempt:
(0.3 ms rise time, yes I cleaned the FR up a bit already)
I'm quite fanatical about these early waterfall plots. More than once I found some odd sounding things by looking at graphs in this way. I want them as clean as I can get (or keep) them. Time coherency is very important to get those clean looking plots. Though these tests are based on pure Dirac pulses in a direct loop.
Disclaimer:
The Choueiri IR I used as base material was an old IR dating back to 2011, floating around on the net, this is NOT related to any of the later work he did (as far as I know).
I just used that IR because I recognized the similarities in that particular IR for what I (or better said: we) want to achieve here. In other words, I'm not putting down any of his work in any way, instead I'd love to get to hear it someday. 
His paper has been inspirational and a great source of information.
P.S. I'll stop naming this a shuffler: let's call it Inter Aural Comb Delay, or should I go for: Aural Comb Delay Convolver, lust look at the capital letters in bold (lol). A nice couter part to Prof. Choueiri's BACCH filter name.
Why not let the graphs speak for themselves:
(0.3 ms rise time)
versus my new attempt:
(0.3 ms rise time, yes I cleaned the FR up a bit already
)I'm quite fanatical about these early waterfall plots. More than once I found some odd sounding things by looking at graphs in this way. I want them as clean as I can get (or keep) them. Time coherency is very important to get those clean looking plots. Though these tests are based on pure Dirac pulses in a direct loop.
Disclaimer:
The Choueiri IR I used as base material was an old IR dating back to 2011, floating around on the net, this is NOT related to any of the later work he did (as far as I know).
I just used that IR because I recognized the similarities in that particular IR for what I (or better said: we) want to achieve here. In other words, I'm not putting down any of his work in any way, instead I'd love to get to hear it someday
. His paper has been inspirational and a great source of information.
P.S. I'll stop naming this a shuffler: let's call it Inter Aural Comb Delay, or should I go for: Aural Comb Delay Convolver, lust look at the capital letters in bold (lol). A nice couter part to Prof. Choueiri's BACCH filter name
.
Last edited:
That I can understand, given the time and care you put into the making of your two wonderful line arrays!
This is of course completely OT here, but if you have a link to a post explaining how you feed these ambient speakers I would be interested to read it.(I'll not count the 2 ambient rear/side channels as they only provide me with a virtual room that's better than the actual room I'm in
Thank you
.I've got 2 links up in the first post of my thread: http://www.diyaudio.com/forums/full-range/242171-making-two-towers-25-driver-full-range-line-array.html
Look under: Adding Ambience! for those two links...
By the way, I've got to add, rePhase has been invaluable in all these experiments! Thanks, first of all for making it and secondly for sharing it with us!
Last edited: