mixing 1st and 3rd
e deleflie
four 1st order to channels to whatever higher order scheme is chosen,
makes the most sense.
Assuming we have a fixed higher order scheme ... 3rd order... 16 channels.
What makes the most sense... for a highly simplified, limited ambisonics scheme?
1) having 3 possible channel counts: 4 (1st order), 16 (3rd order) and
20 (3rd then 1st)
2) having only 1 possible channel count: 20 (1st order followed by 3rd
order) . Empty channels are allowed in the first 4 and the last 16 but
not both (does this work for decoders?)
3) having only 1 possible channel count: 20 ... 16 then 4 (i.e. should
the first order channels come before or after the 16 chn 3rd order
ones?)
4) having 2 possible channel counts: 4 and 20 (i.e. 1st order and
combined 1st and 3rd).
As usual, the motivation is the simplicity of software implementation
and interface, routing signals etc.
Etienne
e deleflie
I think aloud.
I'm wondering what VST instruments would look like if you were mixing
a third order and 1st order piece together.
Would you have one stream of 4 channels (the recorded) ... then an
other stream of 16 channels (the synthesised) .... then some super
funky VST that adds them up into 20 channels and spits them out?
or would you have every VST be 20 channels but the recorded stuff just
uses the first 4 and the synthesised stuff just uses the last 16?
Etienne
Paul Doornbusch
I think that for what you are proposing you will need to define your output and speaker structure, and set up your mix buss' accordingly. I suppose you could try and standardise that for mixed order work, or at least you could set up your own standard mixed order DAW templates to suit your own situation.
Sorry if that's not much help,
Paul
Richard Lee
Ambi, I think we are moving even further away from this objective.
Good, simple and easy to understand formats, help this. 1st order Ambi does
this. FMH, for all its sins, re extensibility, sorta does this.
[rant]
Having separate channels for 1st & 3rd order obfuscates matters for everyone
and seriously turns (Ambi naive) Yamaha away.
Possible Ambi converts will realise this isn't much (if at all) better than
zillion.1, just as wasteful of resources and MUCH more complicated.
I WANT to do a good General Media player with the best Ambi technology based on
VideoLAN. Ambi tech has advantages even for zillion.1 and 5.1 playback IN THE
HOME.
The choice of SN3D in all its forms fills me with dread cos the magic nos. As
Aaron has shown in his experimental codec, this is NOT easy to program. You
need to use extra special functions in the GSL library to avoid overload from
the zillion factorials involved. Its bad enough having to do this for N3D
(which you MUST do internally anyway) without having to generate a 2nd, even
more complicated set for conversion.
Having to deal with extra sets of channels just makes me want to go away and
concentrate on FMH.
________________
Once again ..
A RECORDING format should define what it sounds like; the soundfield.
Shelf filters, psychoacoustic decoding, NFC etc are decoding issues; the
PLAYBACK problem. None of these should influence the recording format.
Let the decoder designers work it out. They use the recorded definition and do
the best they can. Fons has already made a start.
Do we seriously think we have heard the last word on high order decoders? We
still haven't got enough good 1st order decoders! (Only 3 software ones)
That's what BLaH3 is about.
We've hardly touched on what is necessary for good 1st order periphony let
alone HOA playback.
How about some serious work, including listening tests, on HOA decoders
instead of
more pontificating leading only to
even more seriously complicated formats.
[/rant]
Lets keep our format as simple and efficient as possible.
e deleflie
What I am trying to do is remove the complexity of mixed orders. Mixed
orders introduce complexity on a number of levels:
1) what orders and mixed orders should developers support? (why cant
it just be 1 order!) where do they stop?
2) ditto decoders ... and hardware decoders.... where do you draw the
line at what you support? ... which order do you stop at, and which
mixed orders do you support?
3) how can one manage, in routing use cases, all the different
possible permutations of channel sequences?
4) how does the end user (my mother in law) choose which mixed-order,
let alone higher order to use?
UO spec V0.94 attempts to simplify all this, but still falls short (i think).
Further to that, one of Ambisonic's promises is speaker array
agnosticism ... but mixed orders are designed to cater for specific
arrays ... so using mixed orders is like watering down the
speaker-array agnostic promise.
There are 2 user groups who use ambisonics ... those who record
soundfields and those who synthesise soundfields. A combination of 1st
and some higher order represents those two groups well.
An other of Ambisonic's big promises is its capacity to mix recorded
sounds with synthesised sounds. Diffusion cant really do that (and
maintain spatial accuracy), vbap cant do that, I dont know if WFS can
do that .... Soundfield recordings are 1st order, and will remain so
for a while yet. So it is really important to be able to deliver the
ability to mix higher order ambisonics with first order.
We need to set a ceiling for higher orders ... something that is a
default for 'how far to go when implementing higher orders. 3rd order
is good. 8 speakers horizontal is becoming a bit of a default
standard. That's 3rd order. 3rd order then serves 12 (icosahedron) and
16 speakers layouts quite well.
3rd order is also a good default for feeding 22.1 (if I remember
discussions well)
I cant see the Yamaha engineers trying to decide exactly which mixed
orders to support, let alone understand why they have to support a
whole bunch of different permutations of harmonic combinations from
different orders ... when the head engineer is saying "but I thought
ambisonics was supposed to be speaker array agnostic! ... why do we
need all sorts of different channel combos to server different speaker
arrays? ... I dont get it. On top of that if we support all these
mixed order combos, we have more channel schemes than all the
zillion.1s combined!"
mixed orders has the same complexity as zillion.1.....
5.1 / 7.1 is second order horizontal ... is 6 channels of ambisonics
1V 2P 4H (or whatever)
7.1 is third order horizontal ... 8 channels V this P that H somethign else.
12.1 is some H, a bit of V and a bit of P ... is 10 channels of ambisonics
22.1 is third order horizontal plus some Vs and some Ps .... is 17
channels of ambisonics
etc.
(BTW, the above info is all wrong .... but you get my point)
Ideally, it would just be 16 channels ... end of story .... but that
doesn't allow us the ability to serve and cater for mixing synthesised
with recorded soundfields...
Etienne
Sampo Syreeni
> Having separate channels for 1st & 3rd order obfuscates matters for
> everyone and seriously turns (Ambi naive) Yamaha away.
Quite so. What we really have is a very good recording format that is
integral. In the sense that adding together multiple orders yields a
perfectly good representation of mixed accuracy representations of the
soundfield as a whole. Sure, the lower orders do spatially average more
than the higher ones, but in the end the physical description of the
soundfield that results from simple addition is sound.
As such I would once again say that since the psychoacoustical
optimization comes from the decoding end, let's just leave it there.
Let's just blithely mix across orders, and wait for the point in time
where decoders are intelligent enough to separate lower order signal
sets from the higher ones, even in sum. Because at least in theory, that
is at least partially achievable. Let's not litter our standardised
transmission formats with shortcuts to the end of proper psychoacoustic
decoding, over mixed orders. Especially since mixed orders in the
general sense would lead to a signal set plagued by combinatorial
explosion of sidechannels, used to facilitate pantophony, periphony,
multiple orders, and all of the combinations.
Let's instead build a new, perhaps active, theory of decoders, that can
handle sum signals originating with different orders, or in plain words,
signals that sometimes happen to have components where the higher order
coefficients are constantly set nil.
> Possible Ambi converts will realise this isn't much (if at all) better than
> zillion.1, just as wasteful of resources and MUCH more complicated.
Quite so. We want to keep the architecture clear and simple. Even at the
cost of increasingly complicated decoders. Because we would have gone
there anyways, even given first order signals, and certainly the
processing power to do that upto third order is already there, and
increasing exponentially.
> The choice of SN3D in all its forms fills me with dread cos the magic
> nos.
I don't like that either. It should be all N3D because that's what we
operate on.
> Having to deal with extra sets of channels just makes me want to go
> away and concentrate on FMH.
To amplify, I just want to run away from the whole thing if I have to
concentrate on extra channels besides the basic, physically
well-founded, spherical harmonic encoded signal set.
> A RECORDING format should define what it sounds like; the soundfield.
> [...]
I'm right there with you. Proper modularization and a certain separation
of concerns is what has made, and should in the future make, ambisonic
as elegant as it is/could be.
> Do we seriously think we have heard the last word on high order
> decoders? We still haven't got enough good 1st order decoders!
Nor do we have a single active decoder for any order. Eventhough the
original BBC paper on Matrix H *did* suggest an active design could make
a significant positive difference in the case of underdetermined signal
sets (in that case single dominant sources, but the idea can be fully
generalized).
> How about some serious work, including listening tests, on HOA decoders
Quite so. And also new decoder designs, which depart from the
traditional dogma.
> Lets keep our format as simple and efficient as possible.
Hail to that. Whatever went into the mix, the basic S3N spherical
harmonic decomposition should in my mind stand as the ultimate format.
Things like near field coding should be added after the fact, via
scaling coefficients or the like.
--
Sampo Syreeni, aka decoy - de...@iki.fi, http://decoy.iki.fi/front
+358-50-5756111, 025E D175 ABE5 027C 9494 EEB0 E090 8BA9 0509 85C2
Richard Lee
it just be 1 order!)
That's a very sensible decision. They can stop anywhere they want. You don't
have to support 192kB sampling just cos some deaf .. I mean Golden Pinnae
believe they are bats.
> 2) decoders. Ditto. This is an AmbDecLib issue and dependent on the no. of
speakers. A 5.1 AV receiver doesn't have to support 7.1 ... zillion.1
> 3) ... all the different possible ... Ditto
> 4) how does the end user (my mother in law) choose which mixed-order, let
alone higher order to use?
It is transparent to the end user. The recording producer decides if they want
5.1, 7.1 ... zillion.1 or HOA. The end user's Ambisonic Surround Decoder (see
my Ambisonia page on ASD) is matched to his speaker rig (automatically) and
does the best it can with what its presented.
> but mixed orders are designed to cater for specific arrays ... 2 user groups
...
Mixed orders are OPTIMISED for specific arrays but a mixed order signal is
still decoded AS GOOD AS POSSIBLE by the ASD.
> We need to set a ceiling for higher orders ...
That's good. UA has this, as has FMH.
________________
This is what we NEED. (from discussions circa 2007)
1) Good compressed Ambi
2) AmbDecLib
3) A good way to extend to HOA
NOTHING ELSE is REQUIRED (or now IMHO, even good to have.)
1) & 2) can be based on FMH. Only the researchers really like 3)
We should have had 1) a long time ago but the people who did good work on this,
Aaron & Sebastien Olter, were ignored as usual.
(http://www.ambisonia.com/tests/streamingAAC if you want to see what can be
done easily with little fuss)
AmbiDecLib should be the most important contribution by the Ambi community.
Apart from some small stuff by Fons, nothing has been done.
I include myself among those who should be de-pinnaed but I plead computer
problems.
AmbiDecLib which is also the ASD answers all your questions. If you look at
the block diagram, you will see it is also the guts of a HOA ENCODER as well as
a DECODER.
We need to do Listening Tests on HOA decoders to at least the level of the BLaH
tests.
Because AmbiDecLib is the "Answer to Life, the Universe and Everything", I look
at all these proposals and ask ...
Will they make AmbDecLib development easier or more difficult / obfuscating?
Yamaha AND the end user are similarly affected.
I don't think its worth discussing this further. I'll opt out and go back to
compiling VideoLAN with AmbDecLib for FMH on my Sinclair Spectrum.
Expect results come the next millenium.
e deleflie
then please elaborate ... even if its is just a mathematical theory
... put it on the table!
I'm sorry it appears that there's constant circular arguments. I think
the golden rule has already been said ... if you think something is a
good idea, implement it, put it out there ... and if it really is a
good idea, it will take.
I'm looking for a way to allow people to submit higher orders to (the
new) ambisonia ... I want to serve the 8 channel horizontal speaker
scene (and be able to include recordings in that). I dont want people
submitting 312 different permutations of different mixed orders ... I
dont want to write the software that analyses what mixed order the
uploader has done, work out whether there is a good or possible 8
channel horizontal decode, then beg someone to write a speaker decoder
that can optimise speaker feeds *for that specific mixed order*. Its
too complicated! (and I already have some understanding of orders).
... its my real-world use case and I dont think it would be too far
from other (commercial) real-world use cases.
Etienne
Richard Furse
first option is to optimise the 5.1 decode for that material.
The simplest way to deal with mixed order seriously is to re-encode it to
periphonic and ignore the issue. As RL mentioned, it's a (bandwidth)
optimisation.
--Richard
e deleflie
> periphonic and ignore the issue.
order but missing vertical components' or mixed orders as in 'mixing
1st and 3rd both fully periphonic'
Etienne