Crosstalk Cancellation (XTC) Attenuation and Colouration Measurements – CTXMatrix

[STC]

​

For a long time, I have been trying to understand how crosstalk cancellation (XTC) attenuation measurements are actually made in practice. Despite plenty of simulated plots and theoretical charts, I have struggled to find real-world, empirical measurements and claims that XTC introduce severe coloration.

I have always thought that a simple, in-ear binaural measurement should show the actual difference. Since it doesn't seem to match expectations, there must be something else that I am overlooking.

Of course, the measured attenuation can look much greater if we look at an impulse response, where levels exceeding 20 dB of cancellation at certain frequencies are quite common. But does a single-point impulse response actually reflect the real-world effectiveness of crosstalk cancellation when listening to music?

Recently, I’ve been using Claude AI to help analyze actual WAV files captured in my room to see if we can extract meaningful, real-world measurements rather than relying on software simulations. It has generated some interesting graphs, and I am still processing the rest of the data.

For context, my speakers are set up at a narrow 20-degree span, and the recordings were captured using DPA binaural microphones in-ear. Below is the first chart analyzing the classic Sheffield Lab channel test, where the source signal only comes out of the left speaker. It is just a short speech clip saying, "Channel test... Left channel."


In this graph below you can see the additional attenauation that XTC can produce. I will produce another imporatnt chart later once I figure out a way to generat them accurately where phase can play tricks to enahace soundstage.

Fig1 -LR channel separation

Here is the actual leak (crosstalk between XTC and stereo)

Fig2 - Attenaution difference between XTC and Stereo


And finally, the colouration. I have previously post videos of actual sound capture that shows colouration is negligible but here are the measurents.

Fig3 - Frequency response of colouration

And finally here is the snippet of the IR measurements which can further be refined to show attenation of more than 30db at some frequencies.

Fig4- IR graph



Cheers!
ST

 

[Keith_W]

Thank you for continuing your investigations into XTC. I had a bit of a lightbulb moment in our last email exchange, when you said that the point of XTC is not to widen the soundstage, but to improve clarity. It made me wonder whether all those subjective impressions of BACCH demonstrations where people said you could hear people whispering into your ear and the soundstage extending all the way to 180deg and beyond were actually in the recording, and was the effect of improving clarity rather than XTC actually widening the soundstage by itself.

Every time I see one of your posts or guest articles in Archimago's site I feel a bit guilty at my own laziness for not investigating it further. I guess I really should get off my arse and look into it a bit more

 

[STC]

Actually, I had a light-bulb moment about what you mentioned regarding not being able to sense the shift with the occlusion method. I found that I also had a hard time detecting the shift after taking a long break from listening. So I guess we really rely on our prior audio memory and expectations when localizing sounds. Once we re-familiarize ourselves and learn to hear the difference, it becomes much easier. Just a random thought.

I can’t really comment on whether you should hear sound beyond 180 degrees with unknown commercial recordings, since we don’t know what kind of effects or processing they contain. I always use my own binaural recordings to ensure that over-cancellation doesn’t occur, which can artificially widen the stage due to excessive phase manipulation.

I still remember one Madonna track that had a kind of whooshing sound moving to the back (even in the old stereo days). So it’s possible some commercial tracks do contain extreme panning or effects that go beyond 180 degrees.

Having said that, the goal of crosstalk cancellation is to make the speakers as transparent as possible so we can rely purely on the encoded ITD and ILD cues for positioning. At 60 degrees, where the pinna’s natural direction-finding mechanism is strongest, it’s particularly hard for the brain to ignore the physical speaker location. That may require stronger cancellation. How exactly that affects the perceived soundstage is anyone’s guess.

Regarding the improved clarity, it’s simply that without the crosstalk, fine details emerge more clearly.

I realized that calibration process is difficult with CTX and for big setup it is extremely difficult to calibrate if the monitor is not Infront of you and you hold you head steady for a few seconds during the process. A partner would be great but is it practical? NOpe. So I understand your hesitation.

BTW, another 3D product from BACCH. https://www.ecoustics.com/products/canvas-l-audiophile-soundbar/

ST

 

[Keith_W]

If you go to Tidal and search for "8D" you will find a collection of albums remixed for spatial effects. With some of them, if I listen with headphones, I can hear sound coming from behind me. It's a pretty remarkable effect. But with my main rig, the best it can do is slightly more than 180deg, even with crosstalk cancellation. I would be curious what your system could achieve with those recordings.

I don't know how keen you are to keep coding your XTC, but a potential solution to your "monitor not in front of you" problem would be to code up a web interface. Then you could use a phone or a tablet to change CTXMatrix settings.

 

[Redacted]

Sorry to jump in and not be helpful, and I'm not having go, but does crosstalk need attenuating if, as most measurements show, it's not really audible? On real, recorded music, I doubt you'd notice much, since most instruments are panned to be in certain places*, and very rarely at extreme left or right.

*I have a modest rig, and yet with many recordings I can identify the position of each piece in a drum kit across my back wall - or even up/down, as kick drum has quite a bit coming out of L&R subs, and there's a sensation of front-back too. Sometimes I 'hear' extreme panned sounds as being way to the left or right of the speakers - almost in a different room. Most other instruments not mic'd are mono, and get placed L-R in the mix as a choice. I'm not seeing how crosstalk is affecting listening, nor how some kind of processing to reduce it, won't risk introducing new artifacts that make people think a track sound 'clearer' of whatever,

 

[STC]

Sorry to jump in and not be helpful, and I'm not having go, but does crosstalk need attenuating if, as most measurements show, it's not really audible? On real, recorded music, I doubt you'd notice much, since most instruments are panned to be in certain places*, and very rarely at extreme left or right.

*I have a modest rig, and yet with many recordings I can identify the position of each piece in a drum kit across my back wall - or even up/down, as kick drum has quite a bit coming out of L&R subs, and there's a sensation of front-back too. Sometimes I 'hear' extreme panned sounds as being way to the left or right of the speakers - almost in a different room. Most other instruments not mic'd are mono, and get placed L-R in the mix as a choice. I'm not seeing how crosstalk is affecting listening, nor how some kind of processing to reduce it, won't risk introducing new artifacts that make people think a track sound 'clearer' of whatever,

I like your tagline — “Addicted to Fun and Learning” is a great one!

Back in the early 2000s I was heavily into multichannel SACD and wondered how different Ambiophonics crosstalk cancellation could sound compared to regular stereo. So for the sake of fun and learning, I’d encourage you to experiment with crosstalk cancellation as well.

You might find this paper interesting. It explains the longstanding crosstalk problem in stereo reproduction and how it can be properly addressed:

https://ambiophonics.wordpress.com/2023/08/17/true-to-life-sound-reproduction/

There are also several other implementations worth trying, such as those from Home Fidelity Audio, SoundPimp, and BACCH.

The most revealing test is to make a few of your own spaced-pair recordings and compare how they sound with versus without crosstalk cancellation.

ST

 

[STC]

Sorry to jump in and not be helpful, and I'm not having go, but does crosstalk need attenuating if, as most measurements show, it's not really audible? On real, recorded music, I doubt you'd notice much, since most instruments are panned to be in certain places*, and very rarely at extreme left or right.

*I have a modest rig, and yet with many recordings I can identify the position of each piece in a drum kit across my back wall - or even up/down, as kick drum has quite a bit coming out of L&R subs, and there's a sensation of front-back too. Sometimes I 'hear' extreme panned sounds as being way to the left or right of the speakers - almost in a different room. Most other instruments not mic'd are mono, and get placed L-R in the mix as a choice. I'm not seeing how crosstalk is affecting listening, nor how some kind of processing to reduce it, won't risk introducing new artifacts that make people think a track sound 'clearer' of whatever,

Not really a fan of 8D audio — it feels more like a gimmicky effect than natural spatial reproduction. I tried a piano sonata on Apple Music processed this way and found the constant left-to-right swinging quite distracting and unnatural.

They rely heavily on artificial phase manipulation and rotating effects specifically designed for headphone listening. Expecting speakers — even with strong XTC — to fully reproduce those extreme artificial cues at or beyond 180° is probably asking too much. Headphones have a huge inherent advantage here since there’s zero crosstalk by design and pinna is suppressed.

CTXMatrix is aimed at cleaner, more accurate reproduction of real binaural cues (ITD/ILD) rather than these kinds of DSP tricks. A musical performance is fundamentally a frontal event. I don’t expect it to outperform headphones on heavily spatialized 8D content, and I think we may not be seeing eye to eye on what good “3D” reproduction should mean in this context.

Thanks for the web interface suggestion though — that’s a good usability idea. If I continue developing CTXMatrix, I’ll definitely consider adding a web-based control panel for phone/tablet use in future versions.

ST

 

[Redacted]

I like your tagline — “Addicted to Fun and Learning” is a great one!

Back in the early 2000s I was heavily into multichannel SACD and wondered how different Ambiophonics crosstalk cancellation could sound compared to regular stereo. So for the sake of fun and learning, I’d encourage you to experiment with crosstalk cancellation as well.

You might find this paper interesting. It explains the longstanding crosstalk problem in stereo reproduction and how it can be properly addressed:

https://ambiophonics.wordpress.com/2023/08/17/true-to-life-sound-reproduction/

There are also several other implementations worth trying, such as those from Home Fidelity Audio, SoundPimp, and BACCH.

The most revealing test is to make a few of your own spaced-pair recordings and compare how they sound with versus without crosstalk cancellation.

ST

"Addicted to etc" is added by the forum to people who post a bit!
My tagline or signature is "He used dramatic irony, metaphor, bathos, puns, parody, litotes and… satire. He was vicious."

I'm interested in what people experiment with, but for my listening, it's just stereo, and as "un-messed with" as possible. I don't even use tone or EQ, and am very happy with the sound I hear!

 

[STC]

Relooking at my data, I repeated the measurements using the same channel‑test speaker track to verify the accuracy of the crosstalk timing. My speakers are placed at a 20° total angle (10° each side), and using the occlusion method I perceive the acoustic image shifting to about 15°.

With this in mind, I ran the measurements again to check two things:

1. Whether the first‑arrival delay is correctly suppressed by the crosstalk cancellation
2. Whether the remaining delay corresponds to the perceived ~15° acoustic origin

Below are the actual measurement screenshots.

Fig 1 -This is the speach track without crosstalk (plain stereo)

Fig2 - This is with XTC

Fig 3 - Kuhn Modal values.

As you can see, the measured values are very close to what the Kuhn model predicts. However, the actual calibrated ITD values I get during listening are quite different from both the stereo measurement and the Kuhn model. My guess is that the measurements are slightly off because I’m wearing the microphones while doing the occlusion test, so small head movements introduce timing errors. whatever it is perceived XTC position and Kuhn modal are pretty close.

Even so, notice the difference between the model and the measured values during normal stereo playback.

ST