Stereo Crosstalk Elimination (reduction) Par Excellence!

[Tim Link]

It's been a while since I posted anything so I thought it time for an update. I've been experimenting with various crossover points for the 3 speaker array, as well as employing absorption on the floor and behind my head. What I've come to now after comparing the benefits of standard 2 speaker stereo to the 3 speaker array is a 1000Hz crossover for the array, with regular stereo on widely spaced speakers below 1000 Hz. This makes sense for a couple of reasons. Above 1000 Hz is where the center image comb filtering becomes a problem and where we are mostly using level difference to detect direction. Below 1000 Hz it's more phase difference, and the widely spaced speakers can do that well. I was worried that such a high crossover with such large distances between drivers would create an incoherency but it doesn't, assuming everything is adjusted well and you are close to the center. If you are listening off center it's a bit odd but better than listening to the center array crossed over lower. At least you're still getting some stereo effect, and voices still stay located at the center because above 1000 Hz seems to dominate the direction cues when your brain has to choose between lower frequency phase differences and higher frequency level differences. The other good thing about the 1000Hz crossover is it mostly eliminates the issue with increasing lower frequency cancelation in the left and right difference channel, leading to a sense of increased brightness from sounds panned left or right. A third good thing is it minimizes the requirements for the 3 speaker array drivers - they only have to reach down to 1000 Hz, so that opens all kinds of opportunities.

The floor absorption is being accomplished with a 4" pile flokati rug between me and the center array. Behind my head I have a big 2' wide by 5' tall absorber panel that's positioned with it's top all the way to the ceiling and sits a couple inches off the back wall so I can reach the thermostat and light switches behind it.

The 1000 Hz crossover for the center array combined with the absorbers is producing the best imaging I've gotten yet. The soundstage is wide and solid across the center, the strange tonal anomalies with sound coming from the sides seeming brighter than sound in the center is gone. A sense of ambient depth is very present in recordings that have it, and recordings that have special effects are spectacular. Isao Tomita's Greatest Hits album seems to use HRTF type effects to create sounds that seem to circle around you. I hadn't noticed before how strong the effect of the sound going behind your head is. It's really working now. The 1000Hz crossover lets me use just CD horns in the center array, which gets rid of those open baffle 10" woofers I was using and instead employ the midbass horns in my main stereo speakers. Those are great from 400Hz to 1000Hz. The overall tone and clarity is excellent, and measurements show that, with lower distortion and higher clarity. It sounds great.

Another very peasant discovery is the use of equal loudness compensation. I don't know why I wasn't doing this. It's so easy with the EQ functions I have. When I want to turn it down and listen low I just employ the loudness curve and then turn it down. It sounds sublime at low level. I can turn it down very low so I can't hear it at all in the other room, and yet when I sit down I hear everything clear and sweet, with lots of ambience and stereophonic beauty. Highly recommended for low level listening!

The only problem I'm having with this setup is the horns are just big enough and sound best when placed high enough that they block part of my view to the lower edge of the TV screen. I'll raise the TV at some point. It's easier said than done.

 

[Tim Link]

An update to my last update: This time about refining the tonal response, and measuring at the ear offset position rather than the center.

Brief summary - The side speakers in the 3 speaker array need to be EQ'd down at around 4K to keep the tone consistent across the soundstage.

Tone ultimately rules the listening experience, and there's a tonal problem with this 3 speaker array. When sound is panned center, only the center speaker plays. When something is panned to the sides, all three speakers play. For a left signal it's L,L, and -L. This creates more energy in the room and directly at the ear, but more so for some frequencies than others. In my setup, if i measure at the ear position, I see a broad peak centering around 4000 Hz on the left or right panned signal that's a few dB higher than a centered signal once the overall levels are matched. That makes side panned signals brighter. An interesting and airy sound, but not ideal. It can be dazzling but fatiguing.

To try to fix that the obvious thing is to EQ down the side channels. That idea has always irked me though, because then all the speakers in the array don't cancel perfectly. I tried pre-processing the stereo signal with mid/side processing. That works, but it creates leakage from left to right so it narrows the sound stage, and is unnecessarily complex. With that, might as well just EQ down the side channels. So I did, and found I needed a -5 dB parameter at 4000 Hz with 1.5 octave width to get a good spectral match between center and side panned sounds.

This, I believe is the right approach. The sounstage is still very wide with excellent separation, but now with a more lovely and natural tonal consistency. A big step forward for my ears.

Another thing I'm reconsidering is the high crossover point, which has problems of its own. It depends on how wide your main left and rights are apart I think. In my case the stereo speakers are all the way in the corners, so it's very wide, and that makes getting good algnment at the ears difficult between the 3 speaker array and the side channels. 600 Hz is a lot easier than 1200. I get no alignment dips at dead center or off to the sides where the ears are. So that has finally made up my mind for me to stick with the lower crossover.

So, there are trade-offs. For frequencies below about 2K, or wherever your system starts to develop strong comb filtering at the ears for center panned images, a regular 2 speaker system is pretty ideal for creating stereo phase shifts across the head. Above that, something has to be done about the comb filtering. It's not like the comb filtering from room reflections. It's very strong and degrades both centered tone and overall imaging. The problem is how to switch between standard stereo and crosstalk reduction. Moving the frequency down lower than needed helps it to integrate better. Running the 3 speaker array all the way down is reasonable, assuming you have standard subwoofers. Otherwise you're going to have some strange woofer interactions wasting a lot of energy, and often the center woofer is going to be working all alone.

Whatever specific crossovers are chosen, the range between 2000 and 10,000 will need some side channel EQ to keep the tone consistent across the soundstage.

An interesting thing to try is to compare the measurement of a side panned and centered signal at the ear offset position on a standard 2 speaker setup. I wonder if there's much of a measurable difference once the room is factored in and smoothing applied. My guess is not, which is why the tone overall is acceptable with regular stereo and good speakers. The total energy in the room works out correctly. Only the initial sound arrival is totally screwed up.

 

[LIΟN]

An update to my last update: This time about refining the tonal response, and measuring at the ear offset position rather than the center.

Brief summary - The side speakers in the 3 speaker array need to be EQ'd down at around 4K to keep the tone consistent across the soundstage.

Tone ultimately rules the listening experience, and there's a tonal problem with this 3 speaker array. When sound is panned center, only the center speaker plays. When something is panned to the sides, all three speakers play. For a left signal it's L,L, and -L. This creates more energy in the room and directly at the ear, but more so for some frequencies than others. In my setup, if i measure at the ear position, I see a broad peak centering around 4000 Hz on the left or right panned signal that's a few dB higher than a centered signal once the overall levels are matched. That makes side panned signals brighter. An interesting and airy sound, but not ideal. It can be dazzling but fatiguing.

To try to fix that the obvious thing is to EQ down the side channels. That idea has always irked me though, because then all the speakers in the array don't cancel perfectly. I tried pre-processing the stereo signal with mid/side processing. That works, but it creates leakage from left to right so it narrows the sound stage, and is unnecessarily complex. With that, might as well just EQ down the side channels. So I did, and found I needed a -5 dB parameter at 4000 Hz with 1.5 octave width to get a good spectral match between center and side panned sounds.

This, I believe is the right approach. The sounstage is still very wide with excellent separation, but now with a more lovely and natural tonal consistency. A big step forward for my ears.

Another thing I'm reconsidering is the high crossover point, which has problems of its own. It depends on how wide your main left and rights are apart I think. In my case the stereo speakers are all the way in the corners, so it's very wide, and that makes getting good algnment at the ears difficult between the 3 speaker array and the side channels. 600 Hz is a lot easier than 1200. I get no alignment dips at dead center or off to the sides where the ears are. So that has finally made up my mind for me to stick with the lower crossover.

So, there are trade-offs. For frequencies below about 2K, or wherever your system starts to develop strong comb filtering at the ears for center panned images, a regular 2 speaker system is pretty ideal for creating stereo phase shifts across the head. Above that, something has to be done about the comb filtering. It's not like the comb filtering from room reflections. It's very strong and degrades both centered tone and overall imaging. The problem is how to switch between standard stereo and crosstalk reduction. Moving the frequency down lower than needed helps it to integrate better. Running the 3 speaker array all the way down is reasonable, assuming you have standard subwoofers. Otherwise you're going to have some strange woofer interactions wasting a lot of energy, and often the center woofer is going to be working all alone.

Whatever specific crossovers are chosen, the range between 2000 and 10,000 will need some side channel EQ to keep the tone consistent across the soundstage.

An interesting thing to try is to compare the measurement of a side panned and centered signal at the ear offset position on a standard 2 speaker setup. I wonder if there's much of a measurable difference once the room is factored in and smoothing applied. My guess is not, which is why the tone overall is acceptable with regular stereo and good speakers. The total energy in the room works out correctly. Only the initial sound arrival is totally screwed up.

Interesting! Are there any measurements for this?

 

[tjcinnamon]

Did you see the new cross talk features in REW?

 

[Tim Link]

Here's some measurements at my ear (with my legs crossed, just so you know.) The left panned signal should be quieter than the center. Uncorrected you can see that the left pan actually gets louder at times. With side channels EQ it mostly behaves. Looking at this I can see that I need to set the PEQ frequency a little lower so it gets that spot around 2.5k a little more. This correction also reduces the effect above 10k where the sound seems to reverse and come from the right. I think it's still happening but it's much higher now, so I can barely here it happen during the sweep.

 

[Tim Link]

Did you see the new cross talk features in REW?

No!
Where is that?

 

[tjcinnamon]

No!
Where is that?

IACC​

The IACC (Inter-Aural Cross Correlation) panel is used to calculate the early, late and full IACC values for a pair of binaural measurements. The zero time for the measurements is taken as the earlier of the two measurement IR start times. The division between early and late is 80 ms later. The end of the "late" period is 500 ms or the end of the Schroeder integral (where the integral falls into the noise floor), whichever is later. REW does not change the alignment of the measurements, if they need to be aligned the Cross corr. align option may be suitable. The measurements are filtered with one octave, zero phase bandpass filters of the chosen order.

 

[Tim Link]

I haven't found that IACC panel yet, but I'm going to read up on it. edit; I found it. Now to learn what the numbers mean.

Here's an updated measurement with better controls, including a panel to sheild the reflections from behind my head. Despite the volumes all being lower on the opposite side, I still percieve some frequencies to be straight ahead or off to the wrong side when playing sine tones. This is true even when I'm just playing the one center speaker, so sine tones are tough on our sense of direction. I'm surprised that to get to this point I actually had to turn down the side channels by 3 dB. Now that I think of it, there are two of them and only one center. Sort of makes sense. Also a lot of EQ on top of turning them down. So they're way down at 3 kHz and 10 kHz.

The weird thing is that while the measurement looks better with the EQ adjustments, it actually seems to have better stereo separation without any adjustments, as I would have expected. This is a bit of a connundrum because there's really no way to EQ down the extra energy on the sides without compromising the imaging. I wasn't sure at first because it was imaging pretty good still. But with the panel behind my head I can hear a more significant difference.

 

[Tim Link]

IACC​

The IACC (Inter-Aural Cross Correlation) panel is used to calculate the early, late and full IACC values for a pair of binaural measurements. The zero time for the measurements is taken as the earlier of the two measurement IR start times. The division between early and late is 80 ms later. The end of the "late" period is 500 ms or the end of the Schroeder integral (where the integral falls into the noise floor), whichever is later. REW does not change the alignment of the measurements, if they need to be aligned the Cross corr. align option may be suitable. The measurements are filtered with one octave, zero phase bandpass filters of the chosen order.

Thank you! The IACC panel confirms that the correlation is lower when I don't use the EQ. Darn it! Actually, I'm not sure what to make of the numbers. I thought they'd get smaller when the level difference was higher, meaning less correlation. They don't seem to do that.

 

[Tim Link]

Managed to finally effectively mix RACE with my setup, so the side channel recursive cancellation is happening and actually working, but the center is still left alone. This decreases the IACC values, which is good. It sounds really good. Measurements with just regular stereo still show some frequencies being louder in the wrong ear! Something to do with HRTF and perhaps the directionality of my horns. So I'm not going to try so hard to fix that. The good news is that the RACE doesn't seem to be adding enough extra volume at my ear or in the room overall to make it louder than the center channel which isn't using it.

As I'm learning to use REW better, it was easy to determine that I needed a 7 sample delay and about a -2 dB attenuation, with -5 dB at a couple points to make the RACE work.

 

[tjcinnamon]

@Tim Link @STC @Lion

I was at Axpona last weekend and heard Dr Chourie's BACCH it was a great experience! Less wide soundstage than my Polks but more of a 3D presence (like the whispering demo, that could be based on room set up and speakers)

Do any of you have experience with BACCH or HomeAudioForum X-Talk Shaper?

I just tried uBACCH which is a windows VST based on BACCH4MAC. It definitely added to the effect of my L800's.

I'm curious if I could create a convolution file based on the output from those plug-ins to put in a MiniDSP?

Thoughts? I ran the uBACCH VST through JRiver.

 

[Keith_W]

I'm curious if I could create a convolution file based on the output from those plug-ins to put in a MiniDSP?

Thoughts? I ran the uBACCH VST through JRiver.

BACCH, HAF X-Talk Shaper, and other types of crosstalk cancellation are based on RACE - Recursive Ambiosonic Crosstalk Eliminator. The key word here is recursive. This is why they are all VST-based. The DSP unit on the MiniDSP is not recursive, not to mention it is extremely limited in CPU power. It's a nice idea but I don't think it would work.

 

[Tim Link]

@Tim Link @STC @Lion

I was at Axpona last weekend and heard Dr Chourie's BACCH it was a great experience! Less wide soundstage than my Polks but more of a 3D presence (like the whispering demo, that could be based on room set up and speakers)

Do any of you have experience with BACCH or HomeAudioForum X-Talk Shaper?

I just tried uBACCH which is a windows VST based on BACCH4MAC. It definitely added to the effect of my L800's.

I'm curious if I could create a convolution file based on the output from those plug-ins to put in a MiniDSP?

Thoughts? I ran the uBACCH VST through JRiver.

I have no direct experience with BACCH - yet. The convolution seems impossible since it involves two channels interacting with each other. Strangely, my 3 speaker setup could use a convolution because the inverse of the opposite channel is - the same thing it's already doing! So, I could create a convolution of echoes that are properly timed and equalized. Should work in my case. In normal stereo that will not work. If there is a convolver that allows stereo convolutions of the two channels against each other, that could work. The recursive part is ultimately impossible, but practically there's a limited number of cycles needed to get the crosstalk signals down to -60 dB, or however low you need to go until the diminishing returns have set in.

Boy, now that I've said that I'm going to have to try that. I'm going to be buried in REW experiments for the rest of my life!

 

[tjcinnamon]

I have no direct experience with BACCH - yet. The convolution seems impossible since it involves two channels interacting with each other. Strangely, my 3 speaker setup could use a convolution because the inverse of the opposite channel is - the same thing it's already doing! So, I could create a convolution of echoes that are properly timed and equalized. Should work in my case. In normal stereo that will not work. If there is a convolver that allows stereo convolutions of the two channels against each other, that could work. The recursive part is ultimately impossible, but practically there's a limited number of cycles needed to get the crosstalk signals down to -60 dB, or however low you need to go until the diminishing returns have set in.

Boy, now that I've said that I'm going to have to try that. I'm going to be buried in REW experiments for the rest of my life!

Nice! Thats a high quality problem.

How much latency does your audio high jack solution introduce? Could it be used with video. I’d consider adding a box after my AVR for my LR channels. My AVR might really push the distance on them but it would have a limit of 30ms

 

[Tim Link]

Nice! Thats a high quality problem.

How much latency does your audio high jack solution introduce? Could it be used with video. I’d consider adding a box after my AVR for my LR channels. My AVR might really push the distance on them but it would have a limit of 30ms

It's usually below 30ms. My setup gets close to that though on the tweeters because the are closer to me than the woofers, and there's a 6 foot folded path length for the woofer horns on top of that.

 

[STC]

I was at Axpona last weekend and heard Dr Chourie's BACCH it was a great experience! Less wide soundstage than my Polks but more of a 3D presence (like the whispering demo, that could be based on room set up and speakers)

If you are referring to Chesky’s demo tracks then it would be unfair comparison as those recordings made with BACCH in mind.

 

[tjcinnamon]

@Tim Link have you tried BACCH or HAFs plugin compared to your setup. I listened to the BACCH dedicated appliance (setup and with Dr Choiri) at Axpona and downloaded the VST version of BACCH and tried the HAF VST. So far the stereo style setup with my Polks are much better.

My concern is the Polks aren’t made anymore. So I could see having to build something eventually. I could never go back to a standard setup.

 

[Tim Link]

I tried to use BAACH but I was getting a lot of noise and distortion with my system. I'll look into the HAF VST. I've used recursive crosstalk reduction and it definitely widens the soundstage compared to regular stero. From what I could hear with BAACH, the soundstage width and depth was very similar to my setup. I

 

[Tim Link]

@Tim Link have you tried BACCH or HAFs plugin compared to your setup. I listened to the BACCH dedicated appliance (setup and with Dr Choiri) at Axpona and downloaded the VST version of BACCH and tried the HAF VST. So far the stereo style setup with my Polks are much better.

My concern is the Polks aren’t made anymore. So I could see having to build something eventually. I could never go back to a standard setup.

You got me thinking about Polk's SDA, so I've been spending a few minutes listening to something closer to their setup. I'm doing L+R in the center still, but just -L in the right and -R in the left. This relaxes the center image, making it not so intense. This setup is more responsive to time delay of the center speaker to create time aligned arrival at each ear for good cancellation of the crosstalk. I'll have to listen to this for a while. What I'm hearing so far is different, and actually quite nice! It does seem like the the center panned vocals have a slight coloration now, which changes with how I set the time delay. I think I'm hearing a different ratio of direct to reflected sound now. The cancellation is causing it be a little quieter at my listening position, but now the center voice is being played in all 3 speakers so it's traveling around the room more and getting to me later. The soundstage is definitely not lacking in width.

 

[tjcinnamon]

You got me thinking about Polk's SDA, so I've been spending a few minutes listening to something closer to their setup. I'm doing L+R in the center still, but just -L in the right and -R in the left. This relaxes the center image, making it not so intense. This setup is more responsive to time delay of the center speaker to create time aligned arrival at each ear for good cancellation of the crosstalk. I'll have to listen to this for a while. What I'm hearing so far is different, and actually quite nice! It does seem like the the center panned vocals have a slight coloration now, which changes with how I set the time delay. I think I'm hearing a different ratio of direct to reflected sound now. The cancellation is causing it be a little quieter at my listening position, but now the center voice is being played in all 3 speakers so it's traveling around the room more and getting to me later. The soundstage is definitely not lacking in width.

Oh, so you just modified your Mac Mini and are playing through 3 speakers -R; L+R; -L

Nice! Do you have a 4th speaker and amp channel? If so, you could copy it completely. -R L; very wide space in between; then R -L

Where the inverted channels are on the outside. Also, you would take the inverted channels and frequency limit them (my guess is about 400Hz to 5000Hz).

I think a 4 speaker set would provide much more flexibility than the Polks because the speakers could be time aligned to not require being so close together and allow some “toe-ing”