this brings back memories! I remember finding some kind of quadrophonic circuit which may have been mentioned before. Front l & r as usual, think the rear pair were in series in phase and either across the fronts or in the earth line of each. Sounded really rather good on some music. I think it wasnt objectionable when the rears were used as a centre channel. These were the cheapest of speakers but imaging was good. An lpad or 2 would perhaps help.
I thought I put this link in the 1st post:
Trinaural Processor
Don't know excatly how it's done, thought.
I was once convinced by its description, but later I thought, is it really possible or meaningful to extract a 'true center' ?
Except for some rare situations that panned to the extreme left (or right), there's almost always some R in the L, and vice versa. So, where should we draw the line for separating L/C/R ? Where the 'boundaries' should be ? I think it's no way. They should be in a continous transition.
So, "a true center" is probably a wrong thinking.
As to the linear-matrix connection mentioned above, I don't think it degrades the center image. Quite the opposite. It's obviously better than ordinary stereo setup. The more you leave that traditional sweet spot, the more obvious the effect is.
The problem in this method, as I've mentioned some time before, should be in the imperfection of load sharing. With the speakers wired in series, their actual levels are determined by their individual impedances, which are not perfectly matched and worse still, frequency depending.
And it's difficult to make any adjustment.
So a line level circuit (or DSP) for this purpose should be an improvement.
Actually when three channels (speakers) are derived from a stereo sound source of left and right speakers with a center speaker that is a true center there will be no soundstage, only extreme left and extreme right with center, each being a separate sound source with no modulation occurring between the channels which creates the sound stage.
The perception in your head as heard through headphones is the right chanel in relation to the center channel and the left channel in relation to the center channel (which is not true center) it is a combination of the center mixed with the right and the left channel, both are mixed separately. All combined forms stereo. In a practical three channel sound stage created from stereo, each channel can be complely isolated from the other when played individually, but when combined, it gets complicated. My processor in this section takes all into consideration and the center channel output is as fast as it could be, because it responds immediately to the center not removing center but removing left and right channels within one cycle of the inputted frequencies of the left or right channels; therefore allowing the three speakers to complete the front soundstage.
The perception in your head as heard through headphones is the right chanel in relation to the center channel and the left channel in relation to the center channel (which is not true center) it is a combination of the center mixed with the right and the left channel, both are mixed separately. All combined forms stereo. In a practical three channel sound stage created from stereo, each channel can be complely isolated from the other when played individually, but when combined, it gets complicated. My processor in this section takes all into consideration and the center channel output is as fast as it could be, because it responds immediately to the center not removing center but removing left and right channels within one cycle of the inputted frequencies of the left or right channels; therefore allowing the three speakers to complete the front soundstage.
The only real center channel possible that I know of is in the center of your head when listening to headphones. A real center channel is useless unless accompanied by the relevant material in a fourth and fifth speaker. The fourth would be placed between the center and left, the fifth would be placed between the center and right. That fourth and fifth speaker would contain the mix between the center and extreme right, center and extreme left speakers. Or process as I did with three speakers in post 31.
I've seen the link to the Trinaural processor, but I'm not convinced about what it does. I've been trying to figure out what you can do with analog stuff (opamps), but I'm pretty sure you can still only do linear matrixing with them, which still doesn't completely separate the three channels.
As for the sound stage degradation with matrixing, this is my personal experience but also it was found in a listening test with 'expert' listeners by Rumsey that upmixers in general degrade the front sound stage. Of course, in general means that there might be a good one out there.
It seems that new Creative products apply the frequency-domain approach proposed by Avendano and Jot, a technology they call Creative Multi-Speaker Surround (CMSS).
This paper might also interest you.
As for the sound stage degradation with matrixing, this is my personal experience but also it was found in a listening test with 'expert' listeners by Rumsey that upmixers in general degrade the front sound stage. Of course, in general means that there might be a good one out there.
It seems that new Creative products apply the frequency-domain approach proposed by Avendano and Jot, a technology they call Creative Multi-Speaker Surround (CMSS).
This paper might also interest you.
Last edited:
Ambiophonics. Try it with a divider, it costs nothing, stick your nose to the divider. Your center image will be right there in front of you, as plain as someone talking to you in the room, even with non time aligned speakers. Since then, I could not go back to stereo. I like headphones, but I'm used to hearing sounds around me, not in my head.
Another way is the mono center speaker idea. Maybe y-together outputs from a tape loop on the main receiver to another receiver to drive the center speaker. My buddy would do (as I saw earlier posts) in cars, bring the left and right negatives together from the front speakers, connect that to a + of the center channel, then the negative back to either negative of the amp. He'd parallel a pot around the center channel to turn it's level down. It works better with a flatish Z curve.
Norman
Another way is the mono center speaker idea. Maybe y-together outputs from a tape loop on the main receiver to another receiver to drive the center speaker. My buddy would do (as I saw earlier posts) in cars, bring the left and right negatives together from the front speakers, connect that to a + of the center channel, then the negative back to either negative of the amp. He'd parallel a pot around the center channel to turn it's level down. It works better with a flatish Z curve.
Norman
I'm interested in derivation of a true centre channel from two channels, of course!I've read this thread, but all the solutions in the links seemed to do about the same trick. Matrix solutions are simple but cause a degradation of the sound stage. I haven't seen a solution that creates a true centre channel which ONLY contains the correlated signals of L and R from a two channel recording. Or maybe I've missed something??
I'm very curious if someone would know of a system that is able to create a more ideal centre out of two channels...
A completely discrete center is also what I was looking for when I started on my 3-channel experiments, and I also didn't find one. From my thought experiments, I think it would need some sort of block-based convolution/correlation approach and it's not obvious to me that this is
a) practically feasible without artifacts
b) psyco-acoustically correct.
If you haven't already, read the Trifield paper. Their approach is pragmatic but compelling - they used panned pink noise to derive a 2-band matrix approach that led to seamless perception of L/R continuity AND let to no change in apparent image size/spread as the image was panned. While it is true that there is theoretical 'degradation' of the image due to L/R bleeding, due to the negative coefficient in the matrix this actually does some x-talk cancellation which psycho-acoustically restores the percieved soundstage.
So, while I still think a correlation-based approach to derive discrete channels is interesting, I'm not convinced it's necessary or even correct.
The Trifield paper is by Gerzon - 'Optimum Reproduction Matricies for Multispeaker Stereo' from Jul/Aug 1992.
Yes - by Meridian in their Trifield mode (I think they've added a couple extra tweaks, but the paper describes the foundation).
All Meridian pre/pro units implement it as far as I know. The 568.v2 is the minimum to get 96kHz capability though, and those still fetch a higher price on A'gon than I'm willing to pay. I implemented it in Reaper JS scripting pretty easily, though.
Trinaural
Hello fellow skater, I go skating twice a week nothing like it, I wish they had a trinaural at the rink. Here's an example of my version I have been working on of a trinaural decoder. I picked this song because it is very fast and a lot happening in the song. The left and right channels respond to remove the center in about 3 milliseconds. The center which is not included is even faster.
Hello fellow skater, I go skating twice a week nothing like it, I wish they had a trinaural at the rink. Here's an example of my version I have been working on of a trinaural decoder. I picked this song because it is very fast and a lot happening in the song. The left and right channels respond to remove the center in about 3 milliseconds. The center which is not included is even faster.
Hi, thanks for sharing.
One of the diagrams in the last few pages looks identical to this:
http://www.diyaudio.com/forums/mult...t-channels-out-stereo-signal.html#post2661028
"CMSS" looks familiar.
I got it. Actually I have one of their products: USB Sound Blaster Digital Music SX
(I bought it several years ago for the USB interface. Now it serves as a preamp for a pair of active mini monitors for TV and web-radio.)
There's a button marked CMSS. But this device has only 2-channel output, hmm....
For an unknown reason, this thread has stopped...
I wonder if there are other approaches to this 3-channel audio ? I remember for example one old attempt named Salora Ortoperapekta HiFi stereo 3000 (SALORA orthoperspecta) with a center summed channel for 20Hz to 20kHz and two lateral speakers reproducing (L-R) and (R-L) from 300Hz to 3kHz !...
The MS recording method suggested by Blumlein allows the modification afterwards of the width of the soundstage. Several attempts of MS reproduction have been done but with mitigated results.
Stereolith developped and sells a 3-channels speaker giving a good stereo image anywhere in the room : the sweet spot is forgotten here !
I'm after a sort of combination of these previous principles, the goal is to remove the sweet spot and to recreate the original soundstage as close as possible. The initial observation is that when one listen to a violonist for example, the guy is not playing in stereo !... Anyway, we hear him as a 3D image...
I believe the design with "one-speaker" is a coherent direction, instead of three separated loudspeakers, even if there are 3 channels in the same loudspeaker...
All additional suggestion to reach these goals is welcome...
I wonder if there are other approaches to this 3-channel audio ? I remember for example one old attempt named Salora Ortoperapekta HiFi stereo 3000 (SALORA orthoperspecta) with a center summed channel for 20Hz to 20kHz and two lateral speakers reproducing (L-R) and (R-L) from 300Hz to 3kHz !...
The MS recording method suggested by Blumlein allows the modification afterwards of the width of the soundstage. Several attempts of MS reproduction have been done but with mitigated results.
Stereolith developped and sells a 3-channels speaker giving a good stereo image anywhere in the room : the sweet spot is forgotten here !
I'm after a sort of combination of these previous principles, the goal is to remove the sweet spot and to recreate the original soundstage as close as possible. The initial observation is that when one listen to a violonist for example, the guy is not playing in stereo !... Anyway, we hear him as a 3D image...
I believe the design with "one-speaker" is a coherent direction, instead of three separated loudspeakers, even if there are 3 channels in the same loudspeaker...
All additional suggestion to reach these goals is welcome...
Anyone still paying attention? Necro-posting on a 3 year old thread probably won't go anywhere, but just in case.....
I discovered that there IS in fact a (pseudo) correlation-based approach to center channel extraction that is (sort of) available. It looks like it was originally implemented in VirtualDub as the 'center cut' algorithm designed for removing vocals, but someone translated/ported it to a WinAmp plugin and gave it the option of either outputting just the extracted center or the 'cleaned' L/R.
The algorithm is described here: The "center cut" algorithm - virtualdub.org .
Summarized:
- FFT
- start with C=L+R
- compute a such that (L-aC).(R-aC) = 0 (i.e. scale C so that the result of subtracting C from L and R results in orthogonal signals)
- C' = aC, L' = L-aC, R' = R-aC
- inverse FFT
A WinAmp plugin isn't terribly useful anymore, but there is one report of someone successfully running this in a VST wrapper. I am trying to revive my 3-channel desktop experimental system and so I am hoping to get far enough to compare this to my trifield approach.
The next step is to see whether applying ambiophonic x-talk cancellation to the remaining L/R channels in a 3-channel setup works well. My old trifield experiments worked well, but fell short of the huge soundstage presentation of good ambio. I'm curious whether you can get a bit of the 'best of both worlds' with a hybrid approach.
I discovered that there IS in fact a (pseudo) correlation-based approach to center channel extraction that is (sort of) available. It looks like it was originally implemented in VirtualDub as the 'center cut' algorithm designed for removing vocals, but someone translated/ported it to a WinAmp plugin and gave it the option of either outputting just the extracted center or the 'cleaned' L/R.
The algorithm is described here: The "center cut" algorithm - virtualdub.org .
Summarized:
- FFT
- start with C=L+R
- compute a such that (L-aC).(R-aC) = 0 (i.e. scale C so that the result of subtracting C from L and R results in orthogonal signals)
- C' = aC, L' = L-aC, R' = R-aC
- inverse FFT
A WinAmp plugin isn't terribly useful anymore, but there is one report of someone successfully running this in a VST wrapper. I am trying to revive my 3-channel desktop experimental system and so I am hoping to get far enough to compare this to my trifield approach.
The next step is to see whether applying ambiophonic x-talk cancellation to the remaining L/R channels in a 3-channel setup works well. My old trifield experiments worked well, but fell short of the huge soundstage presentation of good ambio. I'm curious whether you can get a bit of the 'best of both worlds' with a hybrid approach.
I've been working on this kind of thing, on and off, since the early 1980's. It won't leave me alone... From reading all of these posts and more, I come up with the following:
If the center channel is mostly weakened in the upper-midrange frequencies by the arrival time differential at the listeners ears, due to the size of the wavelengths involved (makes perfect sense to me), then in an effort to do minimal damage to the overall stereo sound field, we might want to roll off the center channel drive below about 800HZ (perhaps very gradually), and above about 6kHZ (someone pointed out that cymbals got erroneously pulled into the center, or something like that).
It seems that it's all about correlation at the listener position (how the waves add and cancel), of the now 3 channels of sound. David Griesinger seems to have found that the trick to generating good stereo reverb, is to de-correlate the two outputs. How does one de-correlate one signal relative to other signals? I'm not an expert on this, but I suspect it's by time differential. A time delay of either the center channel or the L and R channels, relative to the center channel (?).
I've got an old Carver tuner/preamp that only has Dolby ProLogic 1... (so I just use it for regular stereo), but the center channel signal has a time delay knob, that allows you to move the center out either way, in time, from the rest (L and R and surrounds). I've never actually used this center output, but now I think I will, and see what it offers (as soon as I finish my center channel speaker project).
So now I'm thinking L+R with attenuation (maybe 6dB?) from 100HZ - 800HZ, and from 6kHZ to 20kHZ, and negative time delay (L and R delayed by a few mS ?). Letting it do bass frequencies might help with boomy bass room effects and not really affect bass imaging much. This certainly isn't perfect, but may be worth the trouble.
I might also add an L-XR and R-XL function to the L and R signals, to further cancel the center signal in the L and R outputs, but I'm not sure if that helps more than it screws things up. As you increase that effect, you may be introducing signals that don't add right off axis. This reminds me of speaker system design; perfection is not an option, but you may be able to create a great audio experience with the right psycho-acoustic tricks inspite of that.
If the center channel is mostly weakened in the upper-midrange frequencies by the arrival time differential at the listeners ears, due to the size of the wavelengths involved (makes perfect sense to me), then in an effort to do minimal damage to the overall stereo sound field, we might want to roll off the center channel drive below about 800HZ (perhaps very gradually), and above about 6kHZ (someone pointed out that cymbals got erroneously pulled into the center, or something like that).
It seems that it's all about correlation at the listener position (how the waves add and cancel), of the now 3 channels of sound. David Griesinger seems to have found that the trick to generating good stereo reverb, is to de-correlate the two outputs. How does one de-correlate one signal relative to other signals? I'm not an expert on this, but I suspect it's by time differential. A time delay of either the center channel or the L and R channels, relative to the center channel (?).
I've got an old Carver tuner/preamp that only has Dolby ProLogic 1... (so I just use it for regular stereo), but the center channel signal has a time delay knob, that allows you to move the center out either way, in time, from the rest (L and R and surrounds). I've never actually used this center output, but now I think I will, and see what it offers (as soon as I finish my center channel speaker project).
So now I'm thinking L+R with attenuation (maybe 6dB?) from 100HZ - 800HZ, and from 6kHZ to 20kHZ, and negative time delay (L and R delayed by a few mS ?). Letting it do bass frequencies might help with boomy bass room effects and not really affect bass imaging much. This certainly isn't perfect, but may be worth the trouble.
I might also add an L-XR and R-XL function to the L and R signals, to further cancel the center signal in the L and R outputs, but I'm not sure if that helps more than it screws things up. As you increase that effect, you may be introducing signals that don't add right off axis. This reminds me of speaker system design; perfection is not an option, but you may be able to create a great audio experience with the right psycho-acoustic tricks inspite of that.