Topic: All About Ambisonics

[Len Moskowitz (Core Sound)]

Ambisonics Is About Much More Than Reproducing Soundfields

When people think about using an ambisonic microphone, they usually think of the one classical application for ambisonic microphones that's been used since the 1970s. That's using one ambisonic microphone to record everything going on acoustically at its position. Then we decode the recording to a speaker array (or headphones) for playback, reproducing the soundfield.

In recent years, "ambisonic recording" has broadened to include three more applications.
---

First, let's state a simple definition of ambisonic recording:

- Ambisonic recording is the process of capturing everything going on acoustically at a point-in-space in a way that preserves the spatial information. The spatial information is stored in a specific format.
The format is called "B-format." B-format allows us to determine where sounds are coming from.

B-format can provide different amounts of spatial resolution. The more resolution the description has in capturing the spatial information, the higher its  "order." For example, second-order ambisonic recording has more information about where sounds are coming from than first-order - it has greater spatial resolution.

First-order has greater resolution than zero-order. ("Zero-order" is an omnidirectional description, with no spatial information). Zero-order B-format uses one channel to store its spatial information. First-order B-format uses four channels. Second-order B-format uses nine channels. Higher orders use even more channels

Now back to our original topic: the three other applications of ambisonic recordings.
---

The second application is to use a single ambisonic microphone as a single "virtual microphone."

What's a "virtual microphone"?

Since an ambisonic recording ideally captures everything going on acoustically at a point-in-space, we can use that information to model -  in a computer - how any imaginary microphone would record that acoustic event. That imaginary microphone could have any conceivable pickup pattern. That imaginary modelled microphone is called a "virtual microphone."

The commonly used mono microphones we're all familiar with have a limited range of pickup patterns: omnidirectional, sub-cardioid, cardioid, super-cardioid, hyper-cardioid and figure-eight. Those are called "first-order directivity patterns."

A first-order ambisonic microphone has the spatial resolution to model any or all of the first-order directivity patterns. So a first-order ambisonic microphone can function as a virtual cardioid microphone. Or a virtual super-cardioid microphone. Or a virtual figure-eight microphone. All at the same time, from a single recording.

A second- (or higher-) order ambisonic microphone can model those common mono microphone pickup patterns even better than the best mono microphone. That's because they have much finer spatial resolution.

If you wanted even more spatial resolution than common mono microphones, and if your ambisonic microphone could capture that level of spatial detail, you could create virtual microphones with much more spatial resolution than the mono microphones we're so familiar with. These "second-order" (or "higher-order") ambisonic microphones could have tighter directivity and almost no backlobes. They would pick up sounds pretty much only in the directions we want, and almost nothing in the directions we don't want. They could be asymmetrical if we wanted. And they could potentially reject sounds coming from any desired angle.

So higher-order ambisonics allow us to synthesize types of microphones that we've never had before. It's very difficult to build them physically, so you'll never be able to buy a well-behaved second-order mono microphone. But it's pretty much trivial to make one (or as many as you want, each simultaneously pointed in different directions) - truly excellent ones - using higher-order ambisonics.

Using an ambisonic microphone as a virtual microphone has another very interesting application. Since an ambisonic microphone can be used as a single virtual mono microphone, we can use a bunch of them in spaced arrays. (Examples of common spaced arrays  are ORTF, ORTF-3D, Decca Tree, ESMA, Hamasaki square and cube, Fukada tree.) Since a well-designed-and-calibrated higher-order ambisonic microphone can have more stable directivity patterns and more extended frequency responses than traditional mono microphones, the spaced arrays perform better. And with higher-order ambisonic microphones placed around a space, you can interpolate between their output for 6-degrees of freedom walkarounds in Virtual Reality.
---

The third application of "ambisonic recording" is an elaboration of the second one: we can use  a single ambisonic microphone as a coincident array of virtual microphones. It's not necessarily intended to recreate a complete soundfield.

Just as mono microphones can be used in coincident arrays for stereo recording (e.g., Blumlein, X-Y), so can ambisonic virtual microphones. But where each of the mono microphones in a coincident array takes up real space and so can't create a truly coincident array (because the microphones are displaced in space from the array's theoretical center), ambisonic virtual microphones are truly coincident. For an ambisonic virtual microphone, the displacement of its capsules from the central point-in-space is mathematically compensated, so we end up with a truly coincident array. This results in much better performance than an array made with mono microphones.

So, for example, the world's finest Blumlein array uses a single higher-order ambisonic microphone (e.g., a Core Sound OctoMic). Its two orthogonal figure-eight patterns stay consistent across the frequency band, and the nulls stay deep and true.
---

And finally the fourth application is not about live recording at all, but rather the creation of spatial recordings in the studio using audio editing software, to be played back over speaker arrays or headphones. The original audio sources for the recording might be live recordings made with ambisonic microphones. Or they could use live recordings made with mono microphones and inserted into an ambisonic editing framework. Or they could be sounds created artificially - not live recordings.

What makes this fourth application "ambisonic" is that all of the sound sources end up in B-format, and can be decoded for spatial playback using the ambisonic process. The result when played back over speakers is potentially an accurate soundfield in space. And when played back over headphones, we hear that soundfield presented binaurally.
---

Summarizing, there are currently four applications of ambisonic recording:

1. Recording with a single ambisonic microphone and playing back over speakers or headphones to re-establish the recorded soundfield
2. Using an ambisonic microphone as a virtual microphone - alone or in spaced arrays
3. Using an ambisonic microphone as a coincident array of virtual microphones, often for stereo or binaural playback
4. Creating spatial recordings in the studio using B-format

---

Disclaimer: I own Core Sound, a manufacturer of ambisonic and binaural microphones.

[Len Moskowitz (Core Sound)]

Ambisonics For Live Concert Recording

A single ambisonic microphone can function as any coincident array of traditional mono microphones you can imagine. Place the single microphone where the sound is good. Hit "record".

Back in the studio, decide how many microphones you want to define, what their polar patterns should be, and in what directions they should point. Decode the original recording using those settings. The SPARTA Beamformer plugin is a good tool for this - it's a free download.

You can do this as many times as you want, each time defining a different mix of microphones, polar patterns and pointing angles. It takes only a very few minutes to define and make a new decode.

For example, you can decode to a Blumlein array (two crossed figure-8s rotated 45 degrees from the 0 degree angle). Or XY (two cardioids, or hypercardioids, spread to any angle you want). Or XCY, or XCY with auxiliary mics. Or all of these - a new decode takes just a few minutes.

You can also define playback for 5.1, 7.1, 7.1.4 (Atmos) or any speaker playback layout you want.

Or you can decode that same recording to binaural for playback over headphones, using a generic HRTF (e.g., a Neumann KU100 dummy head) or your own personal HRTF. The binaural recording will respond to a headtracker's outputs, allowing you to turn and tilt your head, and to hear correctly what you'd hear with that head rotation and tilt. So that works great with VR headsets.

What recorder will you need?

For a single first-order tetrahedral ambisonic microphone, you'll need a four channel recorder whose channels track in gain to 0.1 dB. A Zoom F8n Pro or a Sound Devices Mix Pre 10-II will handle two of those microphones. For a single second-order ambisonic microphone, those same recorders will do the job.

You'll also need a shock mount (like a Rycote INV-7) and a stand. And if the venue is windy, a windscreen.

If you plan on decoding to a spaced array like ORTF, you'll need two microphones.

For two first-order microphones, you'll need only eight channels, so the same two recorders will do the job.

For two second-order microphones, you'll need 16-channels. A Sound Devices Scorpio will do the job, or you can use two Zoom F8n Pros linked with a time-code BNC coaxial cable and synced from their menus.

All of these recording setups will easily fit in the usual bags.

---
Disclaimer: I own Core Sound, a manufacturer of ambisonic and binaural microphones.

[Gutbucket]

Ambisionics! Its advanced multi-dimensional Mid/Side.

I look forward to discussion in this thread (upon my return from a trip next week). I'm a Tetramic user and relatively well versed in Ambisonics.

Summarizing, there are currently four applications of ambisonic recording:

1. Recording with a single ambisonic microphone and playing back over speakers or headphones to re-establish the recorded soundfield
2. Using an ambisonic microphone as a virtual microphone - alone or in spaced arrays
3. Using an ambisonic microphone as a coincident array of virtual microphones, often for stereo or binaural playback
4. Creating spatial recordings in the studio using B-format

A few comments on the list above-
#3. Is what members here at TS will be interested in for the most part.  That is essentially an advanced version of a stereo microphone used for X/Y or mid/side. An ambisonic microphone is arguably the ultimate single point stereo microphone and the ultimate learning tool to really understand coincident stereo mic'ing. This is the use for which I have gotten the most usage out of the Tetramic and the basis upon which I can recommended the Tetramic to other tapers. 
#1. Has been done by no tapers that I'm aware of, or anyone else I know personally.  I've been quite interested in doing so via a speaker array for over a decade, yet have still not done so as setting that up is a burden to entry that is more imposing than standard surround playback, and standard surround playback has historically gotten almost zero traction here at TS.  I can envision and would encourage tapers to try this over over headphones, even better with head tracking.  I've not tried this either yet, but it has a much lower barrier to entry.  In any case this is likely to represent a much smaller subset of tapers.
#2. I've done somewhat in combination with other mono microphones, and would like to experiment with more.  I can see a small subset of tapers maybe doing this. I've also given a lot of thought to the use of two or more ambisonic microphones in an array over the years, however I don't envision any tapers doing that.
#4. Is not microphone related but more of a post processing / mixing stage type of thing and in that way has some potential taper applications.

In addition to any discussion on ambisonics in general, I'm happy to answer questions about taper use of tetramic in particular.

[Len Moskowitz (Core Sound)]

One of the nicest advantages of ambisonic recording is that a recording made with a single microphone can end up being played on many different playback configurations.

So a single recording made with a single ambisonic microphone can end up as a stereo track, or a 5.1, or 7.1, or a 7.1.4, or 24.2.8 for playback over speakers. Or it can be played back as fixed-head or headtracked binaural for headphones.

All you need is a single ambisonic microphone, a stand and a recorder.

For a first-order microphone, you'll need a four track recorder. For improved spatial precision and accuracy, you'd use a second-order microphone and an eight channel recorder (e.g., a Zoom F8n Pro or a Sound Devices MixPre-10 II).

[Gutbucket]

Accommodating multiple playback configurations has always been important in my multichannel microphone arrays.  And I do consider using your Octo-mic > Zoom F8, mostly as a small single point recording system that can make for a very portable second rig.

But I'm yet to be convinced that single point ambisonics of less than 3rd order is the best route to quality multichannel playback.  Convenient yes, but my experience points to really needing time-of-arrival cues (spacing between microphones) for multichannel playback to really work well.  I've not had any experience with a 2nd order microphone however. 

Unfortunately, surround recording has never gotten much traction at TS.  Too bad because it works so incredibly well for live music recording playback. More so than most other forms of music IMHO.

[wforwumbo]

How does a co-located microphone capture difference in time of arrival information for binaural reproduction? The error in phase between the point source of microphone location and where either ear would be located from a binaural head-related transfer function induces a time- and phase-based distortion that cannot be recovered. Even assuming the source is a perfect plane wave and along the same frontal plane as the two theoretical HRTFs, there's going to be a nonlinear phase distortion from solving the wave equation that cannot be compensated practically, and theoretically that filter is huge.

And if there is no phase compensation, then the phase distortion is multiplicative between the time/phase distortion of the ambisonic microphone, and that of the HRTF from not being in the same location. It pops out of the dot product projection of placing the ambisonic mic on the head-related transfer function filter...

I am skeptical that an ambisonic microphone can capture the information needed for binaural playback.

[Gutbucket]

^ I've not used it but Harpex is spoken of favorably for rendering B-format to various playback formats and mic'ing emulations, including binaural and spaced pairs. I'm unsure how exactly they do the binarual and spaced pair transcoding, both of which involve time/phase relationships not captured by a single ambisonic microphone.  Here's a link to a couple technical papers hosted at the Harpex website which may cover some of the concerns you raise above: https://harpex.net/documentation.html  I've not yet read either in depth, but the abstract and conclusions are copied below.

HIGH ANGULAR RESOLUTION PLANE WAVE EXPANSION - Svein Berge, Natasha Barrett, 2nd International Symposium on Ambisonics and Spherical Acoustics, May 6-7, 2010, Paris

Abstract:
First-order ambisonics suffers from low angular resolution and a small sweet spot, and decoding to many loudspeakers does not help. Parametric decoding methods solve this problem ,at the risk of introducing audible artifacts. A method for high angular resolution plane wave expansion (HARPEX) is proposed, which combines the spatial sharpness of parametric methods with the physical correctness of linear decoding without introducing audible artifacts. In a formal listening test, a decoder using this method to decode first-order B-format signals scores much higher than max rE decoding of the same signals, and similarly to max rE decoding of third-order versions of the same signals.

6. Conclusion:
The proposed method provides a means for playing back first order material over large loudspeaker setups with improved spatial definition and a much larger sweet spot than is possible with the other method that was tested. Surprisingly good results were achieved over a 5.0 setup compared an eight-loudspeaker setup. The artifacts that are audible in a straight forward decoder using HARPEX can be suppressed to safely inaudible levels without giving up noticeable amounts of sharpness in the auditory scene.

A NEW METHOD FOR B-FORMAT TO BINAURAL TRANSCODING - Svein Berge∗, Natasha Barrett, AES 40th international conference, Tokyo, October 8–10, 2010

Abstract:
A frequency-domain parametric method for transcoding first-order B-format signals to a binaural format is introduced. The method provides better spatial sharpness than linear methods allow. A high angular resolution plane wave decomposition of the B-format establishes two independent direction estimates per time/frequency bin. This alleviates the requirement that the sound sources in a mix are W-disjoint orthogonal, implicit in previous nonlinear methods. The characteristics and causes of audible artifacts are discussed. Methods are introduced that suppress the different types of artifacts. A listening test is presented that ranks the sound quality of the method between third-order and fifth-order linear ambisonics systems.

6 Other observations:
The following observations are based on the authors’ own listening. The material used were the 208 sound files currently available at the Ambisonia website[2], Angelo Farina’s simultaneous binaural and B-format recordings[7] and our own field recordings and synthesized sound scenes. In informal blind testing, using the few available instances of simultaneous B-format and binaural recording techniques, the proposed method seems to provide binaural audio of similar quality to dummy head recordings. Without smoothing or suppression of phase noise, two classes of artifacts become apparent. Most prominent are the artifacts related to filter dispersion. Sharp transients are transformed into noise bursts and intermittent ticking occurs at the frame period, as the contents of some frames wraparound the frame boundaries. Somewhat less noticeable are artifacts related to the time variation in filter coefficients. In combination with overlap-add processing, this gives rise to flutter.

7 Conclusions:
It is known that nonlinear transcoding from B-format to a binaural format provides better sharpness than first order linear transcoding. We have shown how the artifacts that arise from such processing can be suppressed and shown that the resulting sound quality ranks between third order and fifth order linear transcoding. When used in combination with well-known B-format recording techniques, this method provides an alternative to dummy-head recording, with the added advantage of Bformat transformations and tailor-made HRTFs.

[Gutbucket]

Found this in the introduction of the first paper cited above, which doesn't go into what Harpex is doing yet explains how some phase information might be extracted.. presumably using the imperfect capsule coincidence of real world ambisonic microphones to advantage (my interpretation), by way of the positional offset between the directional channels, positioned on the surface of a sphere defined by the diameter of the capsule array, verses the W channel which is a sum of all directional channels and thus positioned virtually at the center of the sphere.

Previous work on sharpening the spatial image [3,4,5] has split the signal into frequency bands and used the short-time correlation between the W channel and each of the directional channels to calculate an estimate of the direction of arrival and “diffuseness” of the sound in each frequency band, which in turn has been used to steer part of the signal to the loudspeakers closest to that direction.

[Len Moskowitz (Core Sound)]

How does a co-located microphone capture difference in time of arrival information for binaural reproduction? ...

I am skeptical that an ambisonic microphone can capture the information needed for binaural playback.

When using a single ambisonic microphone, the latest binaural decoders use a variety of techniques. A good reference is the SPARTA AmbiBin plugin source code.

You can read about AmbiBin here: https://leomccormack.github.io/sparta-site/docs/plugins/sparta-suite/#ambibin
You can see the source code here: https://github.com/leomccormack/SPARTA
The latest research points to Magnitude Least Square (Mag LS) as the most effective process. I can attest that it's very good, though not perfect.

That said, I've found that Bilateral Ambisonics, using two ambisonic microphones spaced by ITD (around 14 cm) to provide superior binaural decoding, including headtracking.
You can read about that technique here:
  https://research.facebook.com/publications/binaural-reproduction-using-bilateral-ambisonics/
  https://sites.google.com/view/acousticslab/audio-material-for-bilateral-ambisonics-reproduction
 This is also a relevant and important paper:
  https://acta-acustica.edpsciences.org/articles/aacus/pdf/2022/01/aacus210029.pdf

Recording using Bilateral Ambisonics, and decoding with a personal HRTF and headtracking allowed me to experience, for the first time, a playback experience that fooled me into thinking it was real, for all 3-degrees-of-freedom rotation and tilt conditions. In addition to the Bilateral Ambisonics recording it requires a personal HRTF, a headtracker and calibrated/corrected headphones.

Bilateral ambisonics allows for true binaural realism with second-order (and higher-order) microphones.


 

[Len Moskowitz (Core Sound)]

^ I've not used it but Harpex is spoken of favorably for rendering B-format to various playback formats and mic'ing emulations, including binaural and spaced pairs.

Harpex uses a parametric (i.e., non-linear) process to increase spatial precision and accuracy. When it works, it allows you to upmix lower order ambisonics to higher orders.

It breaks down when there are more than just a small number of simultaneous audio sources in the soundfield.

My experience using Harpex for creating an ORTF array (or other spaced array) from a single HOA microphone using plane wave decomposition and its parametric algorithm is that the results are not convincing; I got much better results using two spaced ambisonic microphones decoded to cardioids.

[wforwumbo]

How does a co-located microphone capture difference in time of arrival information for binaural reproduction? ...

I am skeptical that an ambisonic microphone can capture the information needed for binaural playback.

When using a single ambisonic microphone, the latest binaural decoders use a variety of techniques. A good reference is the SPARTA AmbiBin plugin source code.

You can read about AmbiBin here: https://leomccormack.github.io/sparta-site/docs/plugins/sparta-suite/#ambibin
You can see the source code here: https://github.com/leomccormack/SPARTA
The latest research points to Magnitude Least Square (Mag LS) as the most effective process. I can attest that it's very good, though not perfect.

That said, I've found that Bilateral Ambisonics, using two ambisonic microphones spaced by ITD (around 14 cm) to provide superior binaural decoding, including headtracking.
You can read about that technique here:
  https://research.facebook.com/publications/binaural-reproduction-using-bilateral-ambisonics/
  https://sites.google.com/view/acousticslab/audio-material-for-bilateral-ambisonics-reproduction
 This is also a relevant and important paper:
  https://acta-acustica.edpsciences.org/articles/aacus/pdf/2022/01/aacus210029.pdf

Recording using Bilateral Ambisonics, and decoding with a personal HRTF and headtracking allowed me to experience, for the first time, a playback experience that fooled me into thinking it was real, for all 3-degrees-of-freedom rotation and tilt conditions. In addition to the Bilateral Ambisonics recording it requires a personal HRTF, a headtracker and calibrated/corrected headphones.

Bilateral ambisonics allows for true binaural realism with second-order (and higher-order) microphones.


 

NOW you are cooking with gas! Bilateral ambisonics... this is the sort of thing I'd be VERY interested in hearing, it's the obvious solution to adding time cue encoding to ambisonics, but I've never had access to more than one ambisonic mic, nor have I thought about having a deck with enough inputs to record that. Not to mention, there would be quite a bit of post processing... but it's a method I think has potential.

Out of curiosity, why did you pick 14 cm? That's a narrower than any between-the-ears distance I've ever seen cited in literature or in practice.

I've heard SPARTA, Harpex, and MLS approaches - all leave a lot to be desired. The spatial resamplers just compound the phase error induced by a single ambisonic mic, and those are aggressively bad to my ear.

[Len Moskowitz (Core Sound)]

NOW you are cooking with gas! Bilateral ambisonics... this is the sort of thing I'd be VERY interested in hearing, it's the obvious solution to adding time cue encoding to ambisonics, but I've never had access to more than one ambisonic mic, nor have I thought about having a deck with enough inputs to record that. Not to mention, there would be quite a bit of post processing... but it's a method I think has potential.

Out of curiosity, why did you pick 14 cm? That's a narrower than any between-the-ears distance I've ever seen cited in literature or in practice.

...

Here's a link to a Facebook post about Bilateral Ambisonics, with links to the raw 18-track B-format recording and two-channel decodes for Bilateral Ambisonics and ORTF with 14 cm separation.
https://www.facebook.com/permalink.php?story_fbid=pfbid0283Ttvwge2KsGSxr95owdbmkg6yLeCFVNa9RkjhqyZt2mUujZAfFzLEFtRVrWp85il&id=100057229368329

From that post, here's the link to the fixed-head Bilateral Ambisonics track:
https://u.pcloud.link/publink/show?code=XZM1POVZNRiIgx17vu5ykvPuDhhR7F78j2fy
It uses AmbiBin's default HRTF data set.

If you'd like to compare it to a binaural decode using a single microphone, let me know and I'll decode one.

Here's the ORTF (14 cm) link:
https://u.pcloud.link/publink/show?code=XZpnPOVZOAF4upvDWmFsYKVOQdQjOF1SHHl7&fbclid=IwAR2e-mNjtZc8G9wf9KKGfL-nB5po2zFDODmX6bgo8mgUmLh2cdxQlnXrEKI

I chose 14 cm because that's my ITD (the distance between my ears), and also because it's reasonably close to ORTF's 17 cm separation.

The post-processing is very simple and fast. I can describe it if you want, or you can see it in the Facebook post.

[Gutbucket]

Thanks Len.  In my surround recording and playback endeavors of live music, I've found time cues most desirable along the L/R axis (in combination with level cues), whereas along the the Front/Back axis level difference becomes of primary importance in order to maximally differentiate front and rear surround content and allow for sufficient surround level without the front image getting pulled around into rear.  A bit of time cue helps that on the Front/Back axis as well, but less is needed and if necessary a touch of delay can be applied without causing problems.  This was recording using non-ambisonic multichannel microphone arrays and using speakers for playback, not rendering to binaural, however the bilateral ambisonics approach using two ambisonic microphones spaced along the L/R axis would seem to fit with this methodology.

Recording using Bilateral Ambisonics, and decoding with a personal HRTF and headtracking allowed me to experience, for the first time, a playback experience that fooled me into thinking it was real, for all 3-degrees-of-freedom rotation and tilt conditions. In addition to the Bilateral Ambisonics recording it requires a personal HRTF, a headtracker and calibrated/corrected headphones.

Bilateral ambisonics allows for true binaural realism with second-order (and higher-order) microphones.

That's encouraging to hear.  Was the source material recorded using two ambisonic microphones or more than two?  If two spaced laterally, how well was head-rotation handled at more extreme rotation angles away from the median plane? Did it become perceptually less robust at larger rotation angles as the lateral time cues approach zero?

Thanks for the links.  Looking forward to listening and reading further.

[Len Moskowitz (Core Sound)]

... Was the source material recorded using two ambisonic microphones or more than two?  If two spaced laterally, how well was head-rotation handled at more extreme rotation angles away from the median plane? Did it become perceptually less robust at larger rotation angles as the lateral time cues approach zero?

These files were recorded with two second-order ambisonic microphones spaced at 14 cm. We have other recordings made with a four microphone array that does ORTF-3D more accurately than the ORTF-3D array from another manufacturer. It's also considerably less expensive.

There are at least two approaches to doing headtracking rotations with Bilateral Ambisonics. The simpler one (which I used) was to rotate the soundfields of the two microphones in-place, and not rotate the complete head. Essentially for this approach, the ears rotate in-place. That preserves the ITD even for extreme rotations. My perception is that the approach is a good one.

In addition to decoding to binaural with headtracking, as you'd expect, this recording can also be decoded to ORTF and every coincident array that you can imagine, including Blumlein, XY, XCY, XCY with spot mics, 5.1, 7.1, 7.1.4 and many others. Each new decode takes just a few minutes.

[Len Moskowitz (Core Sound)]

Streaming Ambisonics For Headtracked Binaural Playback

Did you know that one manufacturer offers an A- to B-format encoder that has a very, very low (2- to 3- millisecond) latency?

That makes it possible to stream real-time headtracked binaural to your listeners.