Loudspeaker Imaging

[skatz]

I recently stumbled across a paper written by Jeff Bagby entitled Loudspeaker Imaging Theorem, (http://diyaudiocorner.tripod.com/imaging.htm#) that I thought was quite interesting. However it was written in 2001, and I wonder how much of it he still believes and what the current state of understanding of this is.

In the paper, he suggests that loudspeaker imaging, ie, the ability to recreate the original localization information in the recording, is fractal in nature. Why this is he doesn't really explain, except to say he believes it is true. He then elaborates that digital recording may cut off this information because it is below a "floor," and that analog is more able to preserve it. So he speculates that that may be why some see analog as superior. He also points out that with modern recording techniques, meaning that the final recording is a composite of many different takes and rooms and times, most of this is lost anyway. but there are techniques (Blumlein and Binaural two mic recording) that can preserve it

Assuming that the information is there in the recording, he then goes on to discuss various ways it can get lost or covered ("veiled" is the term he uses), by such things as room reflections, baffle diffraction, driver mounting (dipole he suggests may be better), lack of structural rigidity, and others.

Finally he notes crossover effects, with lots of reactive components and undue complexity, different path lengths from drivers at various depths, lack of phase coherence in the XO, etc. He suggests that for DIY, LR 2nd and 4th order XOs may be the best networks. But he also thinks series networks may help by reducing the number of reactive components.

So, where are we now with this?

[Sydney]

...digital recording may cut off this information...So, where are we now with this?

Sometimes in a short singular paper, it can be less than clear what is meant in short statements of summation and analogies.
That make analysis in today's terms difficult.
anyway...
Imaging improvement for me was diffraction control by making the baffle infinite ( ie in-wall ) with minimal side reflection.

[Guest]

In my experience, good imaging results from matched amplitude and phase between the two channels. More elaborate crossovers increase the probability that the difference will be greater. Reflections will also degrade imaging. Particularly, the reflections from the edges of the baffle can lead to a 'chaotic' sound field, within which it is very hard to get a precise match between the left and right. Decades ago, there was an article in Audio Magazine about use of felt or foam on the baffle to absorb the wave travelling along the baffle before it gets to the edge and diffracts, part of the energy comes back toward you. About the same time, the highly regarded Dunlavy speakers were using this technique.

About 20 years ago, I built some speakers that had amazing imaging. They were tall column types with 6 inch radiuses along the vertical edges, so from above they looked like a teardrop shape. They were wrapped with multiple layers of felt, with cutouts for the drivers, of course. They were triamped using a custom built active analog XO with precisely matched components. The mids and tweets were Accuton ceramics, my kids managed to crack some of them, so I junked them. My current speakers are nice, but don't image as well as those did.

[neildavis]

There are some well-researched articles and presentations at Dave Griesinger's site that address sound localization and imaging, but his focus is on listening rooms and the effects of wall reflections. His Powerpoint presentations usually have a number of embedded audio samples that illustrate his points. One of his conclusions is that phase coherency of harmonics of the fundamental frequencies are essential to localize sound.

Some of the issues that Jeff addresses in his paper also affect phase coherency of the harmonics--such as baffle diffraction, driver offsets and crossover phase response. However, some of these effects are difficult to visualize and quantify due to the lack of 3D modeling software. But there are some nice tools emerging that may help with this visualization--see the jl ohl website or the new version of XSim, which allows viewing the response while moving the loudspeaker in 3D. However, the holy grail of modeling loudspeaker imaging will require addressing all of these issues, along with room effects, and will require a lot of parallel processing. Maybe the latest generation of video cards will be good platforms for this type of modeling.

[Guest]

I should have also said the wall behind the speakers was covered with acoustic foam tiles.

About the analog vs. digital, perhaps the crosstalk that occurs on a vinyl record has a small effect on imaging. There was an article decades ago in Audio magazine about image enhancement. It described forming the sum and difference signals of a stereo pair, and applying different EQ to them, then recombining them to get left and right again. I experimented with this myself and it did mess with the stereo image. I believe this was the basis of Q Sound (anyone remember that?), but I think Q Sound also used delays, and was applied on separate tracks during mixdown.

[wogg]

....But there are some nice tools emerging that may help with this visualization--see the jl ohl website or the new version of XSim, which allows viewing the response while moving the loudspeaker in 3D...

What?!? New XSim with 3D?!?

Downloaded and playing with it immediately. So far it's a bit odd, throws some floating point errors before you start adding baffle data. Super neat if it works well though!

[Sydney]

...Dave Griesinger's site...

Great material - terrible web page design

[DE Focht]

About the analog vs. digital, perhaps the crosstalk that occurs on a vinyl record has a small effect on imaging.

Yes the LP's lack of channel separation or crosstalk is a nice band aid for the two channel playback system's center image. Lot's of failed attempts to make surround sound with 2 speakers, including these last dipole speakers wasting the considerable talents where a center channel and rears with a proper surround processor would do more better and cost less.

Playing digital back with surround speakers via Ambiance extraction algorithms (DPLIIz, Trifield) and the center channel greatly enhances imaging by dramatically reducing sidewall reflections from the primary source speaker and the algorithms removal of (Unintended) surround information created by complex digital processing, analog reverberation, and complex instruments like a cymbal. "Digital Sound" is mostly a 2 channel phenomena. Unfortunately two channel ideology and the anchored thoughts about audio in the home for music will keep us there for a while longer. Luckily there is enough technology available that surround still out paces stereo playback in my budget range. .

[neildavis]

What?!? New XSim with 3D?!?

Downloaded and playing with it immediately. So far it's a bit odd, throws some floating point errors before you start adding baffle data. Super neat if it works well though!

It's a beta release, and Bill knows there are still a lot of issues that need debugging. Right now he is focused on getting the math correct and validating the calculations. But if you load one of the provided files and turn on the 3D viewer (under Options), you can move the speaker around and immediately see the effect on the response--it's very cool and gives you a lot of useful information about driver placement and crossover behavior.

But that's still just a 3D walk around at a fixed distance from a single loudspeaker, without baffle diffraction or room reflections. There are a lot more features that a full-fledged loudspeaker imaging model should support. I actually bought some 3D goggles to experiment with modeling using Unify and VisualStudio. I'd love to see a fully immersive view of stereo loudspeaker response, where you can walk through a virtual room and visualize the localization values that Greisinger describes. For example, you could depict the phase coherency of harmonics using "fog" of various colors, and determine how well the stereo image was clearly maintained under a variety of conditions. But I'm concerned that the amount of processing would require using a high-performance GPU for real-time rendering, as that's a computing environment that will probably have a steep learning curve.

[philthien]

One of my favorite parlor games is getting people that applaud a setup’s imaging to be specific with details. Once put on the spot, people tend to provide very vague answers that amount to very little. When two or more play, there is very little correlation.

Binaural you say? Watch how Chesky does binaural, where the performers are in relation to one another, and the head/mics, and then come back and tell me if you still think the imaging is what you thought it was.

[wogg]

It's a beta release, and Bill knows there are still a lot of issues that need debugging. Right now he is focused on getting the math correct and validating the calculations. But if you load one of the provided files and turn on the 3D viewer (under Options), you can move the speaker around and immediately see the effect on the response--it's very cool and gives you a lot of useful information about driver placement and crossover behavior.

But that's still just a 3D walk around at a fixed distance from a single loudspeaker, without baffle diffraction or room reflections. There are a lot more features that a full-fledged loudspeaker imaging model should support. I actually bought some 3D goggles to experiment with modeling using Unify and VisualStudio. I'd love to see a fully immersive view of stereo loudspeaker response, where you can walk through a virtual room and visualize the localization values that Greisinger describes. For example, you could depict the phase coherency of harmonics using "fog" of various colors, and determine how well the stereo image was clearly maintained under a variety of conditions. But I'm concerned that the amount of processing would require using a high-performance GPU for real-time rendering, as that's a computing environment that will probably have a steep learning curve.

Didn't even try a sample, but as soon as I put my baffle and driver data in it started throwing graphs out that reacted well to scrolling around. The most surprising part for me was the midbass null from the floor and ceiling reflections. Very cool!

Folks that do this level of complex work in software and throw it out for free hobby use like Bill, Dave Ralph, yourself, and others are tremendous assets to the DIY community!

[JRT]

Lot's of failed attempts to make surround sound with 2 speakers...

...and at least one successful effort in generating 3D illusion from suitable two channel recordings.

Perhaps you should try to convince Edgar Choueiri and his students at Princeton’s 3D Audio and Applied Acoustics (3D3A) Lab.

Choueiri delivered the keynote address at AES 2017, NYC.

http://www.aes.org/events/143/press/?ID=427

https://www.princeton.edu/3D3A/

"...Edgar Choueiri, is a professor of applied physics at the Mechanical and Aerospace Engineering Department of Princeton University, and Associated Faculty at the Department of Astrophysical Sciences, Program in Plasma Physics. He is also Director of Princeton University’s Engineering Physics Program and Chief Scientist at the university’s Electric Propulsion and Plasma Dynamics Lab, a recognized center of excellence in research in the field of advanced spacecraft propulsion. He is also the director of Princeton’s 3D Audio and Applied Acoustics (3D3A) Lab."

Rocket scientist converts stereo into 3D sound

Princeton University professor Edgar Choueiri not only designs plasma rocket engines, he invented 3D stereo!

by Steve Guttenberg
November 12, 2010

https://www.cnet.com/news/rocket-scientist-converts-stereo-into-3d-sound/

Edgar Choueiri is a professor of applied physics at Princeton University, where he is the director of the Engineering Physics Program and the chief scientist of the university's laboratory for advanced spacecraft propulsion (the Electric Propulsion and Plasma Dynamics Lab). Right, he's a rocket scientist, but he's also an audiophile.

Professor Choueiri's Pure Stereo system is a "Revolutionary Technology for Audiophile-Grade 3D Audio." I was treated to a demonstration in Professor Choueiri's lab in Princeton, so I can tell you it really works. The professor played a variety of commercial classical and rock recordings, including Led Zeppelin, over a pair of closely spaced speakers (see photo). The sound spread to the full width of the room, and projected sound forward. He also played recordings he made, including a "haircut" with the sound of "scissors" snipping away all around my head. Professor Choueiri's 3D claim is no hype; Pure Stereo sounds amazing.

The technology can be used with any stereo system, and can operate in any resolution, including high, 192-kHz sampling rates, at 32-bit resolution. Pure Stereo is also compatible with analog sources like turntables and FM radios. Professor Choueiri doesn't equate Pure Stereo (two speakers only) with surround-sound systems that produce envelopment from multichannel music or movie soundtracks from five or more speakers. Pure Stereo's goal is to create more accurate spatial reproduction from two-channel recordings.

It's Professor Choueiri's assertion that stereo playback over loudspeakers is a deeply flawed concept because "Crosstalk corrupts the natural transmission to the brain of 3D cues that exist in all stereo recordings." That is, crosstalk occurs in normal stereo listening when your ears hear both the left and right channels. Professor Choueiri's Pure Stereo BACCH (Band-Assembled Crosstalk Cancellation Hierarchy) filter algorithm prevents your left ear from hearing the right channel speaker's sound, and your right ear from hearing the left channel speaker's sound. Professor Choueiri is hardly the first to develop such a crosstalk filter, but he may be the first to make one that doesn't adversely affect sound quality in the process. The Pure Stereo 3D filter is said to "purify" (eliminate crosstalk) between the two speakers and allow the listener to more accurately hear spatial cues in stereo recordings. In addition, the Pure Stereo 3D filter can provide room correction, to reduce acoustic problems in real world spaces.

Once the crosstalk is eliminated stereo recordings take on a 3D quality; the soundstage is wider, deeper and significantly more solid and realistic sounding. Professor Choueiri has developed three types of BACCH filters: the u-BACCH, a universal filter that can be used with any hi-fi system, without any adjustments; a t-BACCH (theoretical) filter that only requires the end user to input the distances between the two stereo speakers and the listening position in the room; and a c-BACCH filter custom-designed using the audiophile's own head to work with the user's ears, hi-fi system and listening room. The c-BACCH filter provides more exacting spatial cues than the other filters, but I only heard the u-BACCH, and it sounded great to me.

As it stands right now you can't buy a BACCH filter directly, but Professor Choueiri told me that Princeton University has just licensed a local custom installer company (Audtech Associates) to design Pure Stereo c-BACCH filters for audiophiles. Audtech Associates will integrate the filters in their customers' stereo systems. All of this is just now starting to take shape, and the Professor also told me Pure Stereo will also be used in some form with 3D video displays in the near future.

Professor Choueiri's Web site is the best place to find the latest information about Pure Stereo. While you're there, take a few minutes and watch Mike Wood's video, "Introduction to 3D Audio with Professor Choueiri" that demonstrates how Pure Stereo works.

[philthien]

Edgar’s problem is, you can’t put back what was never there.

Look at how 90% of discs are recorded and mastered (take anything from Adele for example) and tell me how a woman in a vocals both should image in relation to the synthesizer player that may or may not have been on-site when the vocals were recorded.

For the well recorded classical/jazz stuff, well, I’’ll keep what I have.

[wogg]

Edgar’s problem is, you can’t put back what was never there.

Look at how 90% of discs are recorded and mastered (take anything from Adele for example) and tell me how a woman in a vocals both should image in relation to the synthesizer player that may or may not have been on-site when the vocals were recorded.

For the well recorded classical/jazz stuff, well, I’’ll keep what I have.

+1

The recording / mixing / mastering engineers and producers are creating an entirely artificial soundstage for you (most of the time, live recordings excepted). Your system's job is to reproduce that. The mix is likely made with a 2 channel stereo monitoring system, and that's what it should be played on. Exceptions: recordings mixed and mastered for surround i.e. movies and music DVD etc.

[DDF]

Classical crosstalk cancellation uses HRTF and it creates tonal distortion. Dr. Choueiri claims to provide the cancellation without the tonal aberrations. Curious to see how (if?) that's accomplished, since our head naturally causes individualized head shadowing of the crosstalk.