Mixed-order schemes
Chris Travis
(I have also posted this to the Ambisonics Association reflector)
For the Ambisonics Symposium, I have now finalized and submitted my
paper on mixed-order schemes. I have also tightened up its conclusions
and recommendations, and adjusted its abstract to match. The adjusted
abstract is viewable at
<http://ambisonics.iem.at/symposium2009/authors/a-new-mixed-order-scheme-for-ambisonic-signals>.
Here's a copy:
>>Traditionally the directional resolution of a 3D Ambisonic signal is
>>uniform over the sphere. It is determined by a single scaling
>>parameter, the periphonic order P. Recently there has been increasing
>>interest in mixed-order schemes that provide higher resolution in the
>>horizontal plane than at the poles. The most widely known is a
>>two-parameter scheme (#H#P) in which the signal is the union of a
>>higher-order horizontal-only component set and a lower-order
>>fully-periphonic component set. We present an alternative
>>two-parameter scheme (#H#V) which truncates the spherical harmonic
>>expansion in a different way. It gives resolution-versus-elevation
>>curves that are flatter, in and near the horizontal plane. The paper
>>includes simulation results for various mixed-order signals and
>>speaker layouts. On the basis of these results the author recommends
>>deprecating #H#P signals with P greater than 1.
Note the final sentence in particular.
If that recommendation is adopted, we will have the following
categories of Ambisonic signal:
- Fully periphonic (P+1)^2 components
- Mixed order (H+1)^2 - (H-V)^2 components
- Horizontal plus Z 2H+2 components
- Horizontal only 2H+1 components
(Plus of-course a "none-of-the-above" category, for custom combinations
of components and matrixed components, e.g. as supported by the
Ambisonics portion of MPEG-4 Part 11.)
For an interface spec a three-parameter approach as discussed here
previously would still be workable, in the sense that it would be more
general than the above list. But other approaches might now be more
appealing. For example, one could instead have two parameters and a
switch. The two parameters could be H and V, with V=H covering
fully-periphonic and V=0 covering the horizontal-only and
horizontal-plus-Z cases. The switch would distinguish between
horizontal-only and horizontal-plus-Z. In the terms of the
three-parameter approach, the switch could be regarded as a P-minus-V
field, constrained to take a value of zero or one.
Discussion of this can wait until the Symposium. But I thought a
"heads-up" might be appropriate now.
Chris Travis
Oliver Thuns
first thoughts:
Do I understand it correctly, that the HV scheme is the subset of the
HVP scheme where P=V?
I think the simplification from HVP to HV is a good thing, less
options are better.
I believe HV makes a good (non-psychoacoustic) lossy compression
scheme (= consumer format), but I cannot imagine it as a production
standard.
e deleflie
> categories of Ambisonic signal:
>
>
> - Fully periphonic (P+1)^2 components
>
> - Mixed order (H+1)^2 - (H-V)^2 components
>
> - Horizontal plus Z 2H+2 components
>
> - Horizontal only 2H+1 components
it might be good to break them up into 2 high level groups: fully
periphonic and other (since horizontal only may not be mixed, but it
is still a sub-group of real-world 3D space).
But again, after all these discussions of mixed order schemes, when
you consider the benefit to those working in ambisonics vs the
complexity it introduces (especially in software authoring
environments), I still believe discussions on mixed-order schemes is a
misplaced focus.
Rather than look at how mixed orders could be represented, I'm
thinking its perhaps more pertinent to look at what speaker arrays
should be represented.... and then see how those speaker arrays can be
served by mixed orders.
Its an interesting debate really, because one of Ambisonics' core
promises is speaker array agnosticism ... but when you extend that
promise to real-world file formats and software implementations, my
gut instinct is that catering for all potential permutations of
mixed-order schemes will damage Ambisonics' useability beyond its
benefit of speaker array agnosticism.
Etienne
e deleflie
> promises is speaker array agnosticism ... but when you extend that
> promise to real-world file formats and software implementations, my
> gut instinct is that catering for all potential permutations of
> mixed-order schemes will damage Ambisonics' useability beyond its
> benefit of speaker array agnosticism.
permutations, you are still speaker array agnostic ... but it just
means that you may not be using the most optimal mixed order combo for
that specific array (which will be an issue for
universities/researchers more than consumers).
Chris Travis
>[OT] Do I understand it correctly, that the HV scheme is the subset of
>the HVP scheme where P=V?
Yes.
>[OT] I believe HV makes a good (non-psychoacoustic) lossy compression
>scheme (= consumer format), but I cannot imagine it as a production
>standard.
I don't think Ambisonics has been served well by the idea that the
ultimate associated speaker layout is spherical. Dedicating equal
resource to polar sounds as to horizontal-plane sounds, e.g. seeking to
render them with the same directional fidelity, is in-fact a bad
move. This is true from multiple different perspectives, including
those of live-music recordists, content constructors/composers,
psychoacousticians and end-users. [It is not true from the point of
view of "big-project academics" or niche architects, but they can look
after themselves in other ways.]
It would be good to wean people off of the idea that 3D Ambisonics is
about regular-polyhedral speaker arrays. Such arrays have some
distinctly bad points. If you render a 1P signal conventionally to a
regular 3D array you get rV=0.58 (quite a lot of blur). But if you
render it to a horizontal array you get rV0.71 (noticably less blur)
for all important source directions. So in that sense, going from 2D
to 3D makes things worse! This is really just an illustration that
regular 3D arrays are *not* the ultimate speaker layout. This comment
applies to Ambisonics, but also more broadly. We would do well to look
at the 10.2 and 22.2 proposals, and ask why they differ so much from
e.g. the experimental spherical and hemispherical systems at various
academic institutions dotted around the world. It is because they come
from people unencumbered by a knowledge of the maths! They come from
people who have been placing much more reliance on their real-world
experience of what works and what doesn't, in practice.
The mixed-order thinking relates very-much to these points. Yes,
mixed-order schemes have a bit-rate-reduction aspect, but I don't see
that as being particularly important. The exercise more to do with
changing what Ambisonics is perceived to be about (and as a result,
changing what Ambisonics ends up being about).
---
Taking a different tack.. My background is in broadcasting. My
experience is that the need for streamlined practices and resource
efficiency can be pretty strong on the production side. If a
broadcaster wants to explore an HOA approach to implementing NHK 22.2,
we could try to interest them in e.g. 4H1V (16 channels) or 4P (25
channels). They'd want some reassurance about the achievable
horizontal resolution across the front stage, so might want to look at
5H1V (20 channels) and 5P (36 channels). My experience tells me that
the 4P and 5P options would never fly. Furthermore, you tend to get
one chance only with these things. If you went in emphasizing 4P and
5P, you'd have blown your one chance.
Chris Travis
Chris Travis
>[ED] after all these discussions of mixed order schemes, when you
>consider the benefit to those working in ambisonics vs the complexity
>it introduces (especially in software authoring environments), I still
>believe discussions on mixed-order schemes is a misplaced focus.
Sorry that I have not responded before to the details of your
'Universal Ambisonic' assay. With one thing and another I have been
rather short of time recently. But here are some quick comments now..
As I understand it, the scheme requires that modules have a way of
knowing how many active channels have been connected to them. As far
as I can tell at the moment, this is the single biggest problem with
the proposal. It limits its scope to a small minority of DAWs, and
hence means that it fails to meet its aims. I have stated this in very
black-and-white terms so that you can counter it with similar
forthrightness if it is wrong! :-)
I think the better way forward is to accept that modules will need
configuring, even if just once on installation. It would be grand if
the configuration could be via some kind of global(s), so that a single
central change is seen by multiple modules at the same time.
At the very simplest, the configuration could be a single binary switch
selecting fully periphonic or horizontal only. But when putting the
mechanism in place, it would be sensible to make the field larger than
1 bit wide, even if 'Universal Ambisonic' places implementation
requirements on only two of the 2^n codes. In the case of a global, it
might also be a good idea to make the mechanism extendable to cover
e.g. two globals.
So how might this way of doing things mesh with the mixed-order
thinking? Well, the one global could be V. V=0 gives us the
horizontal-only case. If the global is e.g. a four-bit unsigned
integer, V=15 give us the fully-periphonic case. (Actually, if the
total number of channels the system is capable of is e.g. 64, V being
any value greater than 6 gives us the fully-periphonic case. This
simply falls out of the maths. No need to include catches for special
values.)
The H parameter possibly does not need to be explicitly
communicated. This is because it doesn't affect the
component-to-channel mapping. It only determines how many channels are
active, within that mapping. On the other hand, one can imagine
situations in which it would be valuable to have the H value globally
available. For example, it could be used (with V) to automatically set
the width of inter-module connections.
In summary, I think 'Universal Ambisonic' is broken and that fixing it
necessarily involves having some kind of configuration mechanism. (Or
supporting e.g. just fully-periphonic signals and processing!) Along
the way, some effort should be put into establishing a framework for
one-or-two Ambisonic globals. Implementation mechanisms for these
could be ad-hoc or more general, on a case-by-case basis. While fixing
Universal Ambisonic, one could quite-easily include the hooks for
mixed-order signals. One wouldn't have to make a big deal this at
present. Just get the hooks in, so that we are better equipped for
what the future will bring.
Chris Travis
Oliver Thuns
B-Format and 8 channel HOA mixed-order. What is better 2H1V or 3H1P?
Chris Travis
>[OT] Let's settle on 4 channel 1st order B-Format and 8 channel HOA
On balance, 2H1V.
Chris Travis
e deleflie
There's essentially 2 things I'd like to say
1) You are right that the "auto-detect how many channels are coming
in" and "adjust outgoing number of channels dynamically" are a
problem. They are a problem because different audio platforms will be
able to satisfy them in different ways. I dont know if you have
expereince with PD or MaxMSP, Supercollider, audiomulch etc.... but
these will be capable of doing this. VST plugins may not. VST 3 may,
but that is undetermined.
My gut instinct is that this should just be removed from the spec...
(or clearly stated as optional)... until further research.
2) Globals cant easily be implemented across different audio
platforms. Take the example of Jack ... which I use to route audio
signals from SuperCollider to Fon's AmbDec. Jack is a very popular
audio routing app. The only way to do globals there is to have some
kind of text file on disk... and then each app has to have the
capacity to point to it/read it/parse it. Even just considering PD,
MaxMSP etc ... doing globals is messy, and different for each system.
I think the only reasonable work around is that each plugin just has
to be manually configured. This detracts from the usability ... but...
Etienne
e deleflie
> B-Format and 8 channel HOA mixed-order.
> What is better 2H1V or 3H1P?
..... put the question thus: what is the speaker array most likely to
host higher order content?
2 rows of 8? 2 rows of 10? ... 1 row of 6 then 1 row of 10 then 1 row
of 6? ...etc.
Etienne
Dave Malham
the more I think that manual configuration is the only foolproof
solution - or at least the only one whereby software developers can
avoid being blamed for problems. I've just just spent several days
frustrating days struggling to play a simple 5.1 DVD properly through
Videolan - and not succeeding because it insists on sending it through
channels 15 to 19 of our dual Motu HD192 system (which are connected to
some of the vertical speakers in our studio rig, causing _very_ strange
imaging). This has convinced me that, unless it is a completely closed
system that only ever processes material that it has generated (i.e
"Ambisonics Inside" systems), there must be full configurability (even
if it is buried well away from the casual user) to make problems
avoidable (which is not the same as avoiding problems).
Dave
These are my own views and may or may not be shared by my employer
/*********************************************************************/
/* Dave Malham http://music.york.ac.uk/staff/research/dave_malham/ */
/* Music Research Centre */
/* Department of Music "http://music.york.ac.uk/" */
/* The University of York Phone 01904 432448 */
/* Heslington Fax 01904 432450 */
/* York YO10 5DD */
/* UK 'Ambisonics - Component Imaging for Audio' */
/* "http://www.york.ac.uk/inst/mustech/3d_audio/" */
/*********************************************************************/
Oliver Thuns
I made up yet another Ambisonics "standard" by staring at 3D
renderings of the components for a while. Working title "Ambisonics
for the Living Room":
4 ch: 1H1V (1P)
8 ch: 2H1V
12 ch: 3H1V
16 ch: 4H1V
20 ch: 5H1V
24 ch: 6H1V
and just realized that Chris put exactly this group of mappings in the
following document under the name "H1V family"
http://ambisonics.googlegroups.com/web/Reworked+mapping+table+(two+versions)+V3.PDF
My point is that I would like to see a (consumer) standard that is as
simple as the H1V subset and doesn't waste channels for underfloor
loudspeakers.
e deleflie
thinking about things in terms of speakers .... you might have a point
about excluding full-periphony families.
Putting speakers underfloor is going to be unrealistic for many arrays
(... except perhaps universities and other institutions who will have
custom software anyway).
There's always going to be the speaker overhead though ... but from a
psycho-acoustic perspective (if I understand right) ... its not that
crucial.
I really feel like a producing a list of the 20 most likely//useful
arrays would be good to guide discussion. Tried raising that on
sursound a while back but didn't get much input (probably because
there aren't that many people doing HOA arrays out there). Maybe we
can revisit this speaker array stuff.
Etienne
Chris Travis
>Chris put exactly this group of mappings in the following document
>under the name "H1V family"
>http://ambisonics.googlegroups.com/web/Reworked+mapping+table+(two+versions)+V3.PDF
Here is a more-recent mapping table.
<http://ambisonics.googlegroups.com/web/Mappings%20for%20all%20%23H%23V%23P%20combinations%20with%20up%20to%2016%20components%20.PDF?>
Actually, this is a figure from my forthcoming paper. The paper
includes simulation results for 2H1V, 3H1V and 4H1V signals played
over dual-ring speaker layouts as illustrated for-example in Eric
Benjamin's recent paper on "Ambisonic Loudspeaker Arrays" (AES
Convention Paper 7605, October 2008). It also includes results for
3H1P and 7H1P signals over dual-ring layouts, and for 3P signals
over a 20-speaker icosahedral layout.
Chris Travis