## Topic: What do you mean by Channel coupling? (Read 12431 times)previous topic - next topic

0 Members and 1 Guest are viewing this topic.
• SebastianG
• Developer
What do you mean by Channel coupling?
##### Reply #25 – 07 August, 2006, 02:42:25 PM
Quote
(...) I thought that at -q6 and above the left and right channels are still interleaved and VQ-coded which accounts for inter-channel redundancy(...)

I am pretty sure interleaving is still done. (...)

So, what's your definition of "channel coupling" then? You said at -q6 there is no channel coupling.

EDIT: I checked the HA wiki on this. I think it needs some improvement regarding Vorbis (Maybe I'll do tomorrow). Quoting: "Lossless coupling is equivalent to independent encoding of the two channels ('dual stereo' in MP3), but with the benefit of additional space-saving. It does polar mapping/channel interleaving using the residue vectors." WTF? How can it be independent coding of two channels if SPM + channel interleaved VQ is in use? BTW: Square Polar Mapping doing any good is a lie that's surprisingly still around.

• HotshotGG
What do you mean by Channel coupling?
##### Reply #26 – 07 August, 2006, 02:57:16 PM
Quote
So, what's your definition of "channel coupling" then? You said at -q6 there is no channel coupling.

EDIT: I checked the HA wiki on this. I think it needs some improvement regarding Vorbis (Maybe I'll do tomorrow). "lossless coupling = independant coding of L/R like dual stereo for MP3?  What's with all the "coupling" then ?  BTW: Square Polar Mapping doing any good is a lie that's surprisingly still around.

Yeah your right I do sound confusing there. I didn't rewrite the section someone else did. We can just call it "lossless". I thought interleaving was still a form of coupling to reduce redundancy? I meant coupling as in phase/point stereo coupling. There are no internal mappings for point/phase couplings for the five channels. You will get some artifacts if you attempt to encode 5 channels now, with what's already in there. Coupling is always done on the channels. Yes, square polar mapping doesn't do much good either. I guess maybe it has something to do with the VQ? you probably know more about the math, behind it then I do.  I know how VQ books work, I have never actual played around with them though. If you looked at it you would probably be able to figure out what sort of mappings routines need to be written to couple 5.1 channels using point/phase routines. It requires an extensive 5.1 setup though. I don't know how one would go about performing blind tests either.

Quote
It does polar mapping/channel interleaving using the residue vectors." WTF? How can it be independent coding of two channels if SPM + channel interleaved VQ is in use?

Interleaving is independent of both channels is that what you are getting at? I thought interleaving was a form of coupling. I guess we need to work on this.
budding I.T professional

• SebastianG
• Developer
What do you mean by Channel coupling?
##### Reply #27 – 07 August, 2006, 03:05:34 PM
Yeah your right I do sound confusing there. I meant coupling as in phase/point stereo coupling.

Ok, the way I see it is: coupling = joint coding (ie exploit redundancy (possibly by interleaving+VQ or M/S) + maybe a little loss like Point Stereo)

Yes, square polar mapping doesn't do much good either. I guess maybe it has something to do with the VQ? you probably know more about the math, behind it then I do.  I know how VQ books work, I have never actual played around with them though

Since the current codebooks are made of vectors with INTEGER samples the difference between having SPM enabled and not is -- from a code book point of view -- only this one: The symbols are rearranged. No gain. No loss. You could merge [SPM+CodeBook] into another code book without having the need to turn on SPM. SPM -- initially(*) designed to decorrelate -- failed doing so (decorrelating) because of |angle| being bounded by a multiple of |magnitude|. Example: mag=0 => angle=0. (mag=0 & ang!=0 is impossible). Exploiting this relationship is done via VQ. So, decorrelating is actually done via channel interleaved VQ

edit: (*) just guessing. It may has been designed to allow certain stereo tricks. But since SPM produces two non-inpenendant outputs you better do channel interleaved VQ anyway which sort of makes SPM dispensable

• SebastianG
• Developer
What do you mean by Channel coupling?
##### Reply #28 – 07 August, 2006, 03:20:36 PM
Interleaving is independent of both channels is that what you are getting at?

It is, but if you then VQ code the interleaved residue vector then, of course, the channels aren't coded independently anymore.

I thought interleaving was a form of coupling.

I like to think so, yes. As it allows to exploit redundancy via VQ coding.

What do you mean by Channel coupling?
##### Reply #29 – 07 August, 2006, 03:39:57 PM
Wow! Everything was just bouncer for me..
But I got HotshotGG's point for not going for 6ch vorbis .. because "There are no internal mappings for point/phase couplings for the five channels" and if you want to convert at -q6 or above , it'll be lossless encoding. So you've only one option, going for -q6 multi channel encoding which is same as encoding 6 mono channels and remuxing. SO for 6 channel audio someone has to work on point/phase couplings for the five channels.

Thankyou once again! And I think Channel coupling has been setup in AAC 6channel encoder.

:-)
The virtue of love isnt finding the perfect person, but by loving the imperfect person perfectly.

• HotshotGG
What do you mean by Channel coupling?
##### Reply #30 – 07 August, 2006, 04:17:54 PM
Quote
It is, but if you then VQ code the interleaved residue vector then, of course, the channels aren't coded independently anymore.

After looking at the source code your explanation makes perfect sense to me now.
budding I.T professional