• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). There are daily reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Stereo Crosstalk Elimination (reduction) Par Excellence!

OP
T

Tim Link

Senior Member
Forum Donor
Joined
Apr 10, 2020
Messages
320
Likes
292
Location
Eugene, OR
So, out of curiosity I did take some measurements of a mockup of this. I have two CS5's, but not three, so I can only simulate the L-R and -(L-R) which I treat as side and -side signals like a MS recording array, and having a hole in the middle where the center speaker would be = diffraction from hell. But anyhow here was the setup on the PC:

View attachment 245032

The measurement mic is located 3' from the front baffle. This seemed to give a somewhat approximate listening distance while still not being swamped with reflections from the room. Edit: Measurement axis was along the midranges, for what its worth. Oh yeah, and the T15s are just there to get them off the table. With that setup, I took measurements from on-axis to 70 degrees off-axis to see what the horizontal pattern looked like (Note: Not normalized):

View attachment 245033

Even though this is a very literal interpretation, there is an indication that there is a null in the middle with the formation of two lobes on either side. Its hard to say how much of the aberrations are due to diffraction (probably a lot) or are due to variations from the two speakers interacting with each other and experiencing constructive and destructive interference. Off the top of my head, one issue with conventional XTC filters was the resultant tonal distortion. I will leave the details to ASR's resident acousticians to elaborate, but Dr. Edgar Choueiri does have commentary on it in AES' Immersive Sound. Interesting read, but really heavy on the math as he goes over the theoretical basis for the BACCH filters. Still worth a look.

I did take a quick listen, and there was some sense of additional imaging beyond just stereo, but too hard to say without adding a center channel and applying corrections to fix any potential coloration from the method being used. The system behind it with the Polk S55 towers and sub is also a system that has XTC, but relying on passive head shadowing, application HRTFs via equalization, and RFZ (and yes the wall behind it needs to be treated. I've been meaning to finish it some day, but being lazy :() so I do have something to compare with. My usual experience with this system is that it has the tonality of headphones on the Harman target, but with much more accurate spatial effects. Its not simply the usual "instruments in a line" but an actual sense of accurate lateral position and depth, with the sources in the recording being perceived as separate entities in their own right.

Still, it looks like it does have merit. To me, it basically looks like an MTM in reverse, with the beaming being on the sides instead of on-axis. I would think with some measurements and tinkering it could work well. But this assumes that there is sufficient headroom if substantial corrections are needed.
Thanks for taking those measurements and posting! You can use one of the T15s as a center speaker and it should work fine for stereo sound field demonstration purposes. Since I used up 5 of my sscs5s at work I only have one left at home, so it is currently flanked by two Athena bookshelf speakers I happened to have. It still produces a fine wide stereo effect along the front of my room. I've been listening to it all weekend. I've got all the sound from my system routed through the Mac Mini

The issue with how to measure a setup like this gets interesting. How should the dispersion look? I think the way to test it would be to test each driver, or tweeter/woofer set individually for its total response, which should match the Harman target. When they're all playing together you get a mess, just like you do in stereo - interference patterns of all kinds. If you have 3 speakers that measure well individually, what happens when you jam them right into each other? The outside edge speakers are going to sound different than the middle speaker. Maybe there should be some extra baffle outboard to make all the drivers load up more similarly, and then some EQ adjustments to make up for the baffle change.
 
Last edited:
OP
T

Tim Link

Senior Member
Forum Donor
Joined
Apr 10, 2020
Messages
320
Likes
292
Location
Eugene, OR
I ordered a Toslink to USB sound input adapter from hifimediy.com so I could route sound from various sources through my Mac Mini and get the required channel summing and differencing done. I've watched movies, played games and listened to music with my set up enough now that I think I'm starting to get a good feel for its sound character compared to regular stereo. It produces a wide sound stage and I like that, with a really solid center image. All good. The downside I think is that it doesn't tickle the pinnae with any real side signals. Everything is coming from straight ahead. So my next experiment is to treat the 3 speaker array as a center channel and mix it with a standard two channel setup. This will require 4 channels of amplification so it starts to get more complicated. Another thing I could try is to modify the 5 speaker setup, leaving the 3 speakers together in the middle but moving the side channels further out. I think that's where I'll start because it's easier. It still only requires a two channel amp, but it's different because it still does the L-R and R-L instead of a plain L and R in the outside speakers. But at least it will send some sound waves at my head from the sides. I think it might be best to have sound coming from straight ahead and from the sides.
 

Cars-N-Cans

Addicted to Fun and Learning
Joined
May 19, 2022
Messages
556
Likes
681
Location
Dirty Jerzey
I was thinking of maybe trying it again later with a cut-out around a T15 to match the baffles of the CS5's. I suspect getting the three speakers to act as though they are on one baffle is probably at least somewhat critical to getting it to work.

As for what the overall pattern should look like, probably something similar to what a mid-side recording array looks like:

blog_mid_side_mic_recording_feat2.jpg


In this case, the front speaker is the red cardioid and the side speakers are the blue figure 8. However the side speakers will tend to project lobes forward instead due to the cabinet and baffle (what we actually want). This is seen if I extract a polar plot at 1 kHz where the pattern is fairly uniform:

1 kHz Polar Plot, Sony CS5 Proposed XTC.png


There are clearly two distinct lobes of sound (with the respective left and right stereo information) off to the side, with a null in the middle where the center speaker provides the material common to both channels. Note that there will be interaction with the center speaker once its in place, so the pattern may end up changing a bit depending on how all the speakers interact with each other.

Note though that this requires the use of semi-anechoic (i.e. gated) measurements to properly see the pattern since you have to exclude the reflections. I use ARTA and my PC's soundcard to do gated measurements. You can also do them in REW, but I think all it can really do is a spinorama and not the directivity and polar plots of other packages, so you would need to know how to interpret it. Also you will need to mock the setup on some kind of turn-table to take incremental angular measurements relative to the microphone. On mine, I turn the speakers in 10 degree increments and take a series of measurements to get a family of gated responses progressively more off-axis.

As for an in-room measurement, I would assume the overall response would be similar to what you would normally see, but there are a lot of factors that go into that such as listening distance, room characteristics, dispersion pattern of the speakers, etc. Edit: Its worth noting that the pattern will get messier as you go up in frequency. This is where measurements and the use of PEQ could help fill in some of that so there aren't as many aberrations in the tonality. Some of the messiness naturally could be alleviated with a purpose-built speaker that is designed to ensure a smoother dispersion pattern at all frequencies, but that could potentially be hard to design without the use of some form of simulation.
 
Last edited:
OP
T

Tim Link

Senior Member
Forum Donor
Joined
Apr 10, 2020
Messages
320
Likes
292
Location
Eugene, OR
I was thinking of maybe trying it again later with a cut-out around a T15 to match the baffles of the CS5's. I suspect getting the three speakers to act as though they are on one baffle is probably at least somewhat critical to getting it to work.

As for what the overall pattern should look like, probably something similar to what a mid-side recording array looks like:

View attachment 245241

In this case, the front speaker is the red cardioid and the side speakers are the blue figure 8. However the side speakers will tend to project lobes forward instead due to the cabinet and baffle (what we actually want). This is seen if I extract a polar plot at 1 kHz where the pattern is fairly uniform:

View attachment 245242

There are clearly two distinct lobes of sound (with the respective left and right stereo information) off to the side, with a null in the middle where the center speaker provides the material common to both channels. Note that there will be interaction with the center speaker once its in place, so the pattern may end up changing a bit depending on how all the speakers interact with each other.

Note though that this requires the use of semi-anechoic (i.e. gated) measurements to properly see the pattern since you have to exclude the reflections. I use ARTA and my PC's soundcard to do gated measurements. You can also do them in REW, but I think all it can really do is a spinorama and not the directivity and polar plots of other packages, so you would need to know how to interpret it. Also you will need to mock the setup on some kind of turn-table to take incremental angular measurements relative to the microphone. On mine, I turn the speakers in 10 degree increments and take a series of measurements to get a family of gated responses progressively more off-axis.

As for an in-room measurement, I would assume the overall response would be similar to what you would normally see, but there are a lot of factors that go into that such as listening distance, room characteristics, dispersion pattern of the speakers, etc. Edit: Its worth noting that the pattern will get messier as you go up in frequency. This is where measurements and the use of PEQ could help fill in some of that so there aren't as many aberrations in the tonality. Some of the messiness naturally could be alleviated with a purpose-built speaker that is designed to ensure a smoother dispersion pattern at all frequencies, but that could potentially be hard to design without the use of some form of simulation.
Getting the wide stereo effect to work doesn't seem to require the baffle to be continuous, although I'm sure minimizing diffraction is going to sound better overall. I had gaps between the speakers and still got a really good sound stage. I'm not sure if it sounded better or worse with the half inch gaps. Maybe I should try again. In experiments using four speakers to recreate the Polk SDA effect I was surprised that I had to space them apart a bit more than what I thought was ear distance to get the effect to be strongest. It got me thinking that the ideal cancel point isn't actually right at the ear canal opening but in the space just outside the ear canal, like half an inch or so outboard. Maybe that's practically the only place it can be. Your T15s are actually what I'm thinking are the perfect width at 6.5". That gives you a bit of room to experiment. Average head width is 6.5 to 7 inches, so if you start at 6.5 you can then space them further apart until the stereo effect is maximized, and then just make spacers to go in between them to achieve a continuous baffle. My Athenas are more than 7" so I'm over 7 inch center to center on my current setup. Still it's producing a very wide sound stage.
The overall dispersion pattern is going to change drastically depending on whether the speaker array is fed something mono, or something panned to the sides to various degrees. That's why I think it's best to think of this as three speakers that just happened to be connected, and check the response of each speaker individually. What's going to happen with the side channels is that a mono signal will produce no sound at all! That's about as poorly measuring a speaker as I can imagine. So I think for measuring purposes we should test the side channels with just one playing at a time, and feed it a left or right signal only, and disconnect the other speakers in the array. If that gives a satisfactory response curve down to whatever bass it can produce, and if it is very similar to the center speaker played alone, then I think the overall energy response of the system will be off to a good start.

However, there are problem with tonality changing with panning in the recording. Sounds that are dead center will all come through the center speaker only, but doubled up with a 3db gain. A sound that is panned hard left will play through the left and center speaker without being doubled, but since there are two speakers it still effectively doubles up to match sound energy of the center speaker playing by itself. But then the right speaker is also playing inverted left, which is adding more sound energy to the room in the high frequencies, but increasingly cancelling out the bass as the frequency goes down due to its proximity to the other drivers. So that leads to a brighter sound from the sides than from the middle, which is something I think I'm hearing.

What to do about the bass? I think before the summing and differencing is applied the bass should be separated out, summed to mono, and then fed back into the speakers afterwards. So a high pass on the summing/differencing matrix and then mix it back in with the bass. Polk does that. What should the crossover frequency and slope be? And how to deal with the extra treble energy in left or right panned signals? The best I can think is to turn down the side channels slightly, which will diminish the stereo width a bit to preserve tonality. We could also leak a wee bit of the center channel in to them. Sounds like the center wide setting in some other 3 channel up-mixers. I was deriding that idea earlier but now I can see why it might be important.

You know, a regular stereo system has an opposite problem, panned signals get less tweeter action than center signals. Do mixing/mastering artists subdue the center panned sound slightly because of this? If so, that makes the problem with this array worse.
 
OP
T

Tim Link

Senior Member
Forum Donor
Joined
Apr 10, 2020
Messages
320
Likes
292
Location
Eugene, OR
More experimenting this morning. I tried accurately spacing the speakers apart at various distances and compared the 3 speaker array to the 5 speaker array in the same room.

1. Spacing - It turns out that spacing them out to about 8" center to center is actually about best. That seems to be close to ideal for my ears. The imaging starts to fall apart if I go further, and doesn't work as well if I squish them closer. The gaps seem to have no stereo degrading effect. If you have to err it's better to err a little narrow than too wide. It gets really weird if you go too far with the spacing. Too narrow just seems to degrade width but it's still cohesive. The good news is that this makes many more speakers useable in the array. Limiting the width to 6.5 inches really puts a damper on speaker choices.

2. More speakers is better - The simple 3 speaker array works but is not as good as the 5 speaker array. Makes me wonder about a 7 or 9 speaker array...

With the 8" spacing and the 5 speaker array I'm getting close to 180 degree stereo spread. On recordings that do ultra wide stereo effects I'm hearing sounds almost completely off to my left or right. It also sounds smoother and more detailed, natural, and clearer all around, which further tells me the wider spacing is adding/canceling more precisely like it should. With more speakers in the array you can get further back and still have a wide stereo spread. The distance is important, just like a standard stereo setup. There's some distance that's ideal but it's not a really sharp Q. You can move forward or back pretty generously and get a slightly different but still very good presentation.

The big sound field makes me want to crank it up but the little SSCS-5s start to sound like they're struggling. So it's a big sound but has to remain relatively low volume. Need subwoofers and better speakers.
 
Last edited:
OP
T

Tim Link

Senior Member
Forum Donor
Joined
Apr 10, 2020
Messages
320
Likes
292
Location
Eugene, OR
JBL Stage A130 looks like an excellent candidate for a ready made good speaker that could work really well in this kind of array. They're currently $199.00 a pair.
 

Cars-N-Cans

Addicted to Fun and Learning
Joined
May 19, 2022
Messages
556
Likes
681
Location
Dirty Jerzey
I don't know if there is a real easy answer to how to quantify what sort of target response you would need from each speaker in the array. Since the two side speakers are coupled together, they have to be measured as such. Separating them and measuring individually breaks the relationship that you have established via the signal chain. Essentially the measurement sweep I did simulates the |(L-R)| difference component of the stereo music content to see how the acoustic energy is distributed around the listening position. The cancellation is not full XTC per se, I don't think, but rather the lobes processing quite steep gradients with respect to lateral position forming a null in the center, minimizing how much acoustical energy that can diffract around to the opposing ear. If the sound field drops off steeply, there will still be crosstalk from each lobe, but this will be dramatically less than what you would get with a speaker that has a wide listening window. There, both ears will hear each speaker at nearly the same SPL level, leading to the contralateral ear hearing essentially the same thing the ipsilateral ear will hear, but with the usual delays from path differences and having to diffract around the head.

An additional wrinkle is that the ITDs and ILDs that we want to segregate via the XTC are being somewhat counter-acted by the fact that they are being combined in the center channel, leading to deliberately induced XTC. Now, if this is about 8-10 dB or more below the difference components it may not pose an issue as its attenuated enough to potentially be ignored by the auditory center. Additionally, the common content in the center channel may help "fill in" the gaps for the sound that arrives within the fusion window. But this is largely speculation on my part. Having the coupled side channels and a dedicated center derived from conventional stereo does complicate things a bit, to say the least. :)

Re: The tonality itself, obviously we want something that is largely flat at the ears, but due to the deliberate formation of lobes in the response, you have to first determine where you are in the sound field before any sort of corrections could be applied. And then these corrections have a high sensitivity to both error in the positional measurements as well as the listener moving about the listening position to a different part of the sound field. I don't know if there is an easy answer, especially since its not as conducive to being measured in the usual ways. Below is some spin data showing the various responses as one moves off-axis. As you can see, the situation is quite messy, so some form of compromise and optimization would be needed to get the best results. Also this will be quite sensitive to how far out you listen as once the reflections become significant there is no longer any meaningful correlation between the two side speakers.

Sony CS5 XTC Spin Data.png
 
OP
T

Tim Link

Senior Member
Forum Donor
Joined
Apr 10, 2020
Messages
320
Likes
292
Location
Eugene, OR
I don't know if there is a real easy answer to how to quantify what sort of target response you would need from each speaker in the array. Since the two side speakers are coupled together, they have to be measured as such. Separating them and measuring individually breaks the relationship that you have established via the signal chain. Essentially the measurement sweep I did simulates the |(L-R)| difference component of the stereo music content to see how the acoustic energy is distributed around the listening position. The cancellation is not full XTC per se, I don't think, but rather the lobes processing quite steep gradients with respect to lateral position forming a null in the center, minimizing how much acoustical energy that can diffract around to the opposing ear. If the sound field drops off steeply, there will still be crosstalk from each lobe, but this will be dramatically less than what you would get with a speaker that has a wide listening window. There, both ears will hear each speaker at nearly the same SPL level, leading to the contralateral ear hearing essentially the same thing the ipsilateral ear will hear, but with the usual delays from path differences and having to diffract around the head.

An additional wrinkle is that the ITDs and ILDs that we want to segregate via the XTC are being somewhat counter-acted by the fact that they are being combined in the center channel, leading to deliberately induced XTC. Now, if this is about 8-10 dB or more below the difference components it may not pose an issue as its attenuated enough to potentially be ignored by the auditory center. Additionally, the common content in the center channel may help "fill in" the gaps for the sound that arrives within the fusion window. But this is largely speculation on my part. Having the coupled side channels and a dedicated center derived from conventional stereo does complicate things a bit, to say the least. :)

Re: The tonality itself, obviously we want something that is largely flat at the ears, but due to the deliberate formation of lobes in the response, you have to first determine where you are in the sound field before any sort of corrections could be applied. And then these corrections have a high sensitivity to both error in the positional measurements as well as the listener moving about the listening position to a different part of the sound field. I don't know if there is an easy answer, especially since its not as conducive to being measured in the usual ways. Below is some spin data showing the various responses as one moves off-axis. As you can see, the situation is quite messy, so some form of compromise and optimization would be needed to get the best results. Also this will be quite sensitive to how far out you listen as once the reflections become significant there is no longer any meaningful correlation between the two side speakers.

View attachment 245457
I really appreciate your thoughts and investigations in to this. It looks like it's just going to take some experimenting to find what sounds good, and maybe try to make some sense of it after the fact. There is some reinforcement of bass from having multiple speakers near each other, but I think the higher frequencies will fill the room with energy equivalent to the character of each speaker working alone. As you have predicted, there are definite changes in tone when you move back and forth in the room. It gets duller and narrower as I move further back from the array, sharper and crisper and wider at the ideal distance, which isn't too far from the array. Maybe 5 feet or so. It's also as picky about room acoustics as any system. The Sony's sound VERY bright in the room at work. In my carpeted front room they are much more mild.
 

tmuikku

Member
Joined
May 27, 2022
Messages
58
Likes
58
As per Cars-N-Cans measurement on center panned (mono) signal the side channel speakers output cancel out around 0 axis, when one is equidistant from both speakers. Especially below ~<2kHz where the wavelength is long enough and acoustic output of both correlates well, and since they are opposite phase, they cancel out.

If you calculate angle to ears when nose is at 0-axis, at ~1meter away angle to each ear it is about ~10 deg, and around 2 meters ~5deg, so the further you are from speakers the less you have direct sound from side channel speakers as the whole head fits the null and the more one hears direct sound from the center channel only. Sidechannels hear through lateral reflections though.

I'm not sure how one should EQ the system, perhaps just by experimenting. If you want to split to highs and lows, do it with the speaker distance in mind. For example lows of side speakers cancel each other out as explained above, guestimating you could mono the system up to frequency whose wavelength is roughly speaker separation distance * 2. Perhaps 1/4wl would be better so if side speakers are 40cm apart they are opposite phase and effectively cancel out completely to all directions with mono signal up to 160cm wavelength, above which a combfilter starts. I would try mono the system bass up to ~200Hz or so and the figure out what to do with the highs, whats the balance of the system now?
 
OP
T

Tim Link

Senior Member
Forum Donor
Joined
Apr 10, 2020
Messages
320
Likes
292
Location
Eugene, OR
As per Cars-N-Cans measurement on center panned (mono) signal the side channel speakers output cancel out around 0 axis, when one is equidistant from both speakers. Especially below ~<2kHz where the wavelength is long enough and acoustic output of both correlates well, and since they are opposite phase, they cancel out.

If you calculate angle to ears when nose is at 0-axis, at ~1meter away angle to each ear it is about ~10 deg, and around 2 meters ~5deg, so the further you are from speakers the less you have direct sound from side channel speakers as the whole head fits the null and the more one hears direct sound from the center channel only. Sidechannels hear through lateral reflections though.

I'm not sure how one should EQ the system, perhaps just by experimenting. If you want to split to highs and lows, do it with the speaker distance in mind. For example lows of side speakers cancel each other out as explained above, guestimating you could mono the system up to frequency whose wavelength is roughly speaker separation distance * 2. Perhaps 1/4wl would be better so if side speakers are 40cm apart they are opposite phase and effectively cancel out completely to all directions with mono signal up to 160cm wavelength, above which a combfilter starts. I would try mono the system bass up to ~200Hz or so and the figure out what to do with the highs, whats the balance of the system now?
I've been mixing the mono bass back in with the side channels about where you suggested. I forget exactly what frequency I landed at after some experimenting but it is sounding good. I was surprised how much it degraded the side channel width if I took the mono bass crossover frequency a little too high. I also tried high passing the L-R channel before mixing the bass back in but it seems to sound better if I don't do the high pass. I'm thinking a better solution might be to cross all the speakers over at some point and just move to widely separated regular L, R stereo for the mid-bass and below down to wherever the subs come in.
Last night's listening was the best I've heard yet. With the speaker spacing more optimized and the bass better managed it's becoming increasingly pleasing.

Good point about the distance determining the crossover point. I can hear the side channels canceling at higher and higher frequencies with increases in listening distance. This makes sense because the difference in timing between the intended signals and the crosstalk, which is out of phase, gets smaller and smaller as you move away from the speakers, so the array has to have more speakers added to it to restore the side channel info by repeatedly canceling the cancelations. I think this is why going from 3 to 5 speakers causes the panned sounds to seem crisper and more detailed than with just 3 speakers. With my 5 speaker array it seems the ideal listening distance is maybe 50% more than the total width of the array. So I think this is starting to answer the question of how many speakers should be in this array. It depends on how far away you are. On a very closely placed desktop system 3 is probably plenty. In a mid-sized room 5 is good. Maybe 7 or 9 would be better in a bigger room. If the array gets too wide for how close you are, HRTF EQ will be needed and maybe a slight increase in spacing on the furthest outboard channels to prevent them from becoming too loud and for not compensating for the extra distance of the crosstalk having to wrap around the head to reach the opposite ear.
 
Last edited:
OP
T

Tim Link

Senior Member
Forum Donor
Joined
Apr 10, 2020
Messages
320
Likes
292
Location
Eugene, OR
Playing with the ripple tank simulator http://www.falstad.com/ripple/ I superimposed a couple of multi pole sources on to a point source, which simulates the right channel playing on the 5 speaker array in this case. What is interesting along with the cancelation patterns is the consistent phase lag of the signal between the right ear and the left ear. The right signal reaches the left ear but it is lagging. The wave front is effectively bent across the head as if it was arriving from the right. It's bent more severely up closer and then starts to straighten as distance is increased

Screenshot (18).png
 

Cars-N-Cans

Addicted to Fun and Learning
Joined
May 19, 2022
Messages
556
Likes
681
Location
Dirty Jerzey
Since we know that there is the formation of lobes in the side channels, I think an obvious question is have you tried to use speakers with a narrow dispersion pattern? An example would be the T15s that were seen under the CS5's. While these have a crummy 1st order crossover network and flipped tweeter to keep the cost down, the inverted waveguide and high crossover frequency means they have a fairly narrow dispersion pattern. Below is the normalized horizontal directivity plot:


T15 Normalized Horizontal Directivity.png


If we ignore the messiness from the tweeter polarity being flipped, the listening window is about +/- 40 degrees. In polar form:

T15 Polar Pattern, 2 kHz.png


While the lobe would need to be about half of this to get roughly the same response as the CS5's in your configuration, I would think with a suitable horn speaker that has rather narrow directivity located about 60-70 degrees apart you could get the same effect you are getting now with the side array with simply using the left and right channels as-is with a summed center channel for infill. The addition of the center channel could help to balance out the soundstage and provide a nice sense of spaciousness if it has a wide listening window of around +/-60 to 70 degrees. An additional benefit would be the ability to delay the center channel by about 10-15 ms from the left and right speakers. This would help ensure that the direct sound from the left and right speakers is not obscured by the combined center channel. The timing, however, is fairly critical as we want to be far enough away from the direct sound to ensure most of the cues get conveyed properly, but not so far that the center channel's content falls too far outside the fusion window and ends up being heard as an echo. This would be more well behaved, and be able to be measured in the usual ways we normally do using the in-room response and a spectrogram to image the reflections. Maybe something to try if time/finances permit as it could give the same result but without the need to resort to complex arrays and EQ strategies.
 
OP
T

Tim Link

Senior Member
Forum Donor
Joined
Apr 10, 2020
Messages
320
Likes
292
Location
Eugene, OR
Since we know that there is the formation of lobes in the side channels, I think an obvious question is have you tried to use speakers with a narrow dispersion pattern? An example would be the T15s that were seen under the CS5's. While these have a crummy 1st order crossover network and flipped tweeter to keep the cost down, the inverted waveguide and high crossover frequency means they have a fairly narrow dispersion pattern. Below is the normalized horizontal directivity plot:


View attachment 245662

If we ignore the messiness from the tweeter polarity being flipped, the listening window is about +/- 40 degrees. In polar form:

View attachment 245663

While the lobe would need to be about half of this to get roughly the same response as the CS5's in your configuration, I would think with a suitable horn speaker that has rather narrow directivity located about 60-70 degrees apart you could get the same effect you are getting now with the side array with simply using the left and right channels as-is with a summed center channel for infill. The addition of the center channel could help to balance out the soundstage and provide a nice sense of spaciousness if it has a wide listening window of around +/-60 to 70 degrees. An additional benefit would be the ability to delay the center channel by about 10-15 ms from the left and right speakers. This would help ensure that the direct sound from the left and right speakers is not obscured by the combined center channel. The timing, however, is fairly critical as we want to be far enough away from the direct sound to ensure most of the cues get conveyed properly, but not so far that the center channel's content falls too far outside the fusion window and ends up being heard as an echo. This would be more well behaved, and be able to be measured in the usual ways we normally do using the in-room response and a spectrogram to image the reflections. Maybe something to try if time/finances permit as it could give the same result but without the need to resort to complex arrays and EQ strategies.
I haven't tried any other speaker yet because I don't have anything else. I do have some very narrow dispersion tweeter horns and drivers - just two of them. I could buy a few more and try it. Expensive experiments, at least for me. I was thinking something with a better directivity would be good. Limited by the 8" width though as far as keeping the directivity down to a lower frequency.

There's a guy in town that has a bunch of speakers, some of them older, still in boxes. He may let me try some others. I'll go over to his place with my computer and hopefully I can demonstrate this array and get his thoughts on it. More people need to listen to it to see if it's really all that great, or just "interesting."

One thing I've noticed with this array is that the volume needs to be turned up a bit to get the width to flesh out better. I think at low volume there aren't enough details audible to give the timing cues, or whatever. I just hear sound coming from the middle with a little stereo effect. Turn it up a bit and things get deeper and wider. With a real multichannel array you can tell something is coming from the direction of the speaker even at lower volumes because all the timing cues and HRTF are perfectly consistent. So, in the end I still think something like Dolby Atmos multi-channel mix is the way to go. Or a really good up mixer for two channel. I've heard that Dolby Atmos has good upmixing but I've not yet heard it myself. I'd have to get a new receiver. At least I've already got plenty of speakers!
 

Cars-N-Cans

Addicted to Fun and Learning
Joined
May 19, 2022
Messages
556
Likes
681
Location
Dirty Jerzey
One thing I've noticed with this array is that the volume needs to be turned up a bit to get the width to flesh out better. I think at low volume there aren't enough details audible to give the timing cues, or whatever. I just hear sound coming from the middle with a little stereo effect. Turn it up a bit and things get deeper and wider. With a real multichannel array you can tell something is coming from the direction of the speaker even at lower volumes because all the timing cues and HRTF are perfectly consistent. So, in the end I still think something like Dolby Atmos multi-channel mix is the way to go. Or a really good up mixer for two channel. I've heard that Dolby Atmos has good upmixing but I've not yet heard it myself. I'd have to get a new receiver. At least I've already got plenty of speakers!
Regarding the need for volume my own listening experience with my setup is that the binaural aspect allows the system to maintain full detail and stable imaging down to very low SPL levels since it does a better job of conveying the imaging cues (a great benefit for late-night listening). My thoughts would be that the variations in the response cause some of the imaging cues to fall below the threshold of hearing. Turning up the volume brings these out of the "noise floor", so to speak, and allows the cues to become audible. Getting the response to be flatter will ensure the cues don't drop out at lower SPL levels.

This would be one potential advantage to going with a pair of horn speakers or something similar that have a narrow pattern along with the center channel. You can have better control over the response and ensure its fairly flat on-axis. I think Earl Geddes had the same thoughts as well as to trying to provide increased direct to indirect sound levels for improved imaging in one of his white papers. I believe it was this one on speaker directivity: Speaker Directivity. My personal preference is for speakers with a wide dispersion pattern, but in conventional stereo setups this often results in a lower ratio of the direct to indirect sound heard at the listening position.
 
OP
T

Tim Link

Senior Member
Forum Donor
Joined
Apr 10, 2020
Messages
320
Likes
292
Location
Eugene, OR
While the lobe would need to be about half of this to get roughly the same response as the CS5's in your configuration, I would think with a suitable horn speaker that has rather narrow directivity located about 60-70 degrees apart you could get the same effect you are getting now with the side array with simply using the left and right channels as-is with a summed center channel for infill. The addition of the center channel could help to balance out the soundstage and provide a nice sense of spaciousness if it has a wide listening window of around +/-60 to 70 degrees. An additional benefit would be the ability to delay the center channel by about 10-15 ms from the left and right speakers. This would help ensure that the direct sound from the left and right speakers is not obscured by the combined center channel. The timing, however, is fairly critical as we want to be far enough away from the direct sound to ensure most of the cues get conveyed properly, but not so far that the center channel's content falls too far outside the fusion window and ends up being heard as an echo. This would be more well behaved, and be able to be measured in the usual ways we normally do using the in-room response and a spectrogram to image the reflections. Maybe something to try if time/finances permit as it could give the same result but without the need to resort to complex arrays and EQ strategies.
I re-read this and now think I understand you are suggesting using the narrow horns just for the side channels, with a delayed center that may also be narrow, or could it be wide? I definitely have used the narrow horns with them pushed all the way out to the corners of my room, but haven't tried adding a center channel with them. I have corner horns. Actually two sets of corner horns, a pair of Klipschorns and a pair of DIY much larger corner horns. To get the 60-70 degree spread I'd have to play the room sideways, or sit very close to the front wall, or move the tweeters and mids forward from the front wall. I was able to get a very narrow dispersion pattern down to a little below 1000 Hz with the narrow horn tweeters and midrange horn I was using. I decided I didn't like that sound overall all though it worked really well for me on some orchestral mixes. I ended up with a somewhat wider dispersion horn crossed over down at 600Hz to get the dispersion to match the midrange horn at the crossover. That sounds better to me. It's still pretty narrow dispersion so early room reflections are reduced, which does help clarity and imaging as Geddes suggested. But in that situation crosstalk of the standard type is still rampant across the head. There's no way to get a speaker's dispersion narrow enough to beam sound at just one ear with any distance. I've thought about that problem a lot, and even ordered some ultrasonic speakers that beam encoded high frequency sound that deconstructs into audible band sound when it bounces off a hard surface. That can beam sound right at each ear, but the sound quality is not high fidelity by any standard. It's just good enough for clear, intelligible speech. In my experience so far, whenever there's standard 2 channel inter aural crosstalk the imaging is not in the same league as I get from my 5 speaker array. And the 5 speaker array still lags behind a true crosstalk elimination scheme, like using a divider wall coming right up to your face. That's still the best 2 channel imaging I've heard so far. The 5 speaker array is the next best thing to it I've heard so far. Recursive crosstalk elimination schemes also produce some great imaging with just 2 channels but it doesn't sound clean enough to me and uses up a lot of system headroom because of the recursive signal being overlayed onto the soundtrack.

A wide stereo array with a delayed center channel I think is a great candidate for some kind of up mixed matrix, but I really think the center information should be removed from the side channels in that case. And preferably the side channel information from the center as well. Otherwise that's sending a more confused, messier set of signals to the ears. My 5 speaker array produces a stereo effect that I think is substantially different than a wide standard stereo with a summed center channel added. Not to say that the standard stereo with summed center might not sound really good. Paul Klipsch used something like that. I really should try adding a summed center channel with the Klipschorns. That will require an additional amp channel as well. FWIW, I don't notice any hole in the center problem with the Klipschorns or my other corner horns, and I did have the DIY corner horns spread very wide apart at one time - close to 60 degrees at my friend's house where they were built. He had a really big room and those big horns worked great in there. I thought they produced a typical stereo 2 channel presentation with some extra width because they were so far apart, but not anything like the kind of effect a separator wall can produce. I wasn't expecting them to compete with a crosstalk elimination setup so there was no disappointment.

What I think could ultimately make me happiest would be a really good up-mixer that would take a 2 channel signal and mix it appropriately to maybe 9 speakers placed widely across the front of the room, and do so with very high fidelity. The up mixed signal should be capable of being down mixed back to 2 channel again and sound just the same as it did originally. Maybe an AI up-mixer with high powered hardware could do this in real time someday. More likely it would be a re-mastering effort. In my opinion, more channels in the recording played back through more speakers is best. I really like some of the multichannel mixes on Apple Music. Crosstalk reduction or mitigation with just 2 channels is next best thing, and any standard 2 channel configuration that doesn't mitigate crosstalk comes in last place for imaging.
 

beagleman

Addicted to Fun and Learning
Joined
Sep 24, 2020
Messages
784
Likes
973
reducing stereo crosstalk that occurs from the sound of the speakers reaching both ears instead of just the intended ear.



I mean great work and all but.....but...but....

That is how music is mixed and recorded to begin with. They realize it will be heard that way, and it is mixed and panned and so on with that all taken into account on the recording end of things.

Polk made this somewhat famous in the 80s as the SDA series of speakers, but never explained WHY it was needed, other than making the sound Gimmicky.

For TRUE Binaural records, YES it works to a great degree, just like how headphones do also.
But REAL stereo recordings you are supposed to hear the crosstalk, and it is intentional.

Am I the only one thinking crosstalk cancellation is being very unfaithful to the intended sound>?
 
OP
T

Tim Link

Senior Member
Forum Donor
Joined
Apr 10, 2020
Messages
320
Likes
292
Location
Eugene, OR
reducing stereo crosstalk that occurs from the sound of the speakers reaching both ears instead of just the intended ear.



I mean great work and all but.....but...but....

That is how music is mixed and recorded to begin with. They realize it will be heard that way, and it is mixed and panned and so on with that all taken into account on the recording end of things.

Polk made this somewhat famous in the 80s as the SDA series of speakers, but never explained WHY it was needed, other than making the sound Gimmicky.

For TRUE Binaural records, YES it works to a great degree, just like how headphones do also.
But REAL stereo recordings you are supposed to hear the crosstalk, and it is intentional.

Am I the only one thinking crosstalk cancellation is being very unfaithful to the intended sound>?
True, the crosstalk is expected to be heard by those doing the mixing and mastering. But that's because there's usually no way around it. When all you've got is two speakers for playback, there are compromises that have to be made. I've read recording engineers words on this. It won't necessarily sound the way they'd ultimately like it to, and there's nothing they can do about it. 2 channel stereo is a gimmick too. But I like it better than mono, and it's the most practical arrangement beyond mono. My experience is that the vast majority of 2 channel stereo recordings benefit from having the crosstalk reduced or dealt with in some way. To my ears very few, if any, recordings actually sound worse unless the crosstalk scheme is adding distortion or tonal coloration. A lot of normal stereo systems add a lot of distortion and tonal coloration anyway. Get the tone right and distortion low and the standard 2 channel setup sounds great. Same is true with any crosstalk reduction or obfuscation scheme. I think this array of 5 speakers works great not because it actually gets rid of the crosstalk, but just makes it out of phase across the head so it doesn't compete with the initial image by trying to flip it backwards. Now for center panned stuff it actually does get rid of the crosstalk by playing it all out of the center speaker. But whoever is mixing and mastering can't be sure how far apart the listener will have their speakers, or if the recording will be mixed down to mono, or if people will be frequently listening off axis, so I doubt they actually color the center panned tone to account for crosstalk or other HRTF from a typical stereo spread. Maybe in some recordings. I'm not hearing anything but very clear and natural sounding vocals from the center, and I definitely prefer that to having the same signal hit both ears from the sides to create a center phantom image, which I hear moving with me if I move my head left or right. I'm learning that this 5 speaker array has a sweet spot which seems to be something like 1.5 to 2 times the width of the array. So that puts the best listening distance around 60 to 80 inches, which makes unsuitable for a lot of situations. It's kind of too close for my big screen TV. But closer listening has been my preference for standard 2 channel stereo configurations as well in order to reduce room colorations and optimize the stereo effect, so nothing new there. If you have multiple people listening in a big room most of them are going to miss out on the best sound, although it still has a benefit in that it keeps the center of the soundstage in place at the expense of stereo width. An alternate take on it with the outside speakers set further apart might be interesting. I guess I better try it.

Ultimately you have to decide with your own ears if you find the sound of a particular stereo configuration enjoyable or not. I think the problem with SDA was most people who owned it, like myself, never got it set up correctly. I think it's super finicky in terms of aligning the speakers with the listener's ears accurately, and that's hard to do when the speakers are spaced far apart like in a normal stereo system. And even when I've gotten their method to work well in my own experiments with the speakers all right next to each other, it tends to sound too "busy." Things sound doubled to my ears, which they are. Still, the imaging is improved. For some reason this 5 speaker array sounds much cleaner to me, in some ways even cleaner and purer than standard 2 channel stereo. It might be that everything is built around the summed center, just pulling the stereo width out of it only as needed. It lends itself well to long listening sessions, which is keeping me up too late.
 

tomelex

Addicted to Fun and Learning
Forum Donor
Joined
Feb 29, 2016
Messages
964
Likes
535
Location
So called Midwest, USA
very good thread and comments all along, the inventor of stereo knew and tried to deal with this from the begining but never could resolve it and died before having a chance to get that center channel right. crosstalk is part of the reason anything even slightly complex (like more than one instrument) is just not going to fly in plain old two channel stereo. It is an illusion (not a replication) but some folks brains just do not fall so easily to it.
 
OP
T

Tim Link

Senior Member
Forum Donor
Joined
Apr 10, 2020
Messages
320
Likes
292
Location
Eugene, OR
And additionally I don't see how the signals could combine at the ears with enough accuracy to make the cancellatoin work
Yes, it's surprising it works so well, and doesn't seem to require head-in-vice positioning accuracy. This came as quite a surprise for me. I expected this array to deliver some kind of stereo effect but the quality of it really astounded me. As I have stated earlier, I've come to conclude that this isn't so much crosstalk cancelation as perhaps crosstalk obfuscation. With the speakers all lined up in a straight row in front of you, as long as they are spaced apart appropriately for your ears, the timing will be correct at various distances as long as you are positioned in the center. So this makes it easier to get the speakers set up accurately and the listener positioned. If you look at the ripple tank simulation of the array you can see that the phase of the signal reaching the head lags at the opposite ear, with the wave form bent across the head. So a timing cue is maintained that isn't all that fragile in terms of location accuracy. If you're reasonably centered you get good timing differential for panned signals, which may explain why it can generate such a wide sound field with very good separation and solid imaging all the way across.
 
Top Bottom