In the mpeg familly, we basically have 2 kind of joint stereo:
*mid/side stereo: 1 channel is (L+R), the other is (L-R)
*intensity stereo: 1 channel is similar to (aL + (1-a)R), other channel is just a simple position indicator per sfb

What I am wondering is why is there no generalized joint stereo mode that would be intensity/side:
*1 channel intensity
*1 channel side channel according to the intensity position
*signalisation of the position per sfb.

It seems to me that this would allow current modes:
*m/s: position signalled to be centered
*is: side channel quantized to zero

...but would also allow to use joint stereo more often, as with instruments that are not centered. As this could gracefully degrade down to current IS, it would also be suitable for low bitrates.

The most similar thing seems to be the "square polar mapping" of Vorbis, which is 1 channel with intensity and the other channel with position, position beeing able to be defined up to each frequency coeff in the extreme case, down to coarse representation similar to IS.

So, why is there no Intensity/Side? Is is just because one mode is coming from AT&T and the other one from FhG, or is there a real technical reason?

Are you suggesting that -- instead of a 45° rotation M/S does (actually it's a 45° rotation followed by inverting one channel's sign) -- you want arbitraty angles to maximize decorrelation per scalefactor band ?
If so, you need to transmit the angle as side information too which might not be worth the hassle in contrast to scalefactor band selective M/S (like in AAC).

Otherweise, I'm not following you.

As for maximizing decorrelation in Vorbis, this can actually done via the floor curves. In MP3 you do M/S mapping BEFORE rescaling the spectralvaues via scalefactors (encoder). In Vorbis the floor curves are able to normalize both channels (subband-wise) PRIOR further decorrelation attempts. In case of a simply panned signal (no phase differences between L and R) Vorbis may scale both channels via certain floor curves to produce (approx) equal residue vectors. Vorbis then may use channel-interleaved VQ coding with a code book that assigns short codes to vectors containing similar components to exploit that there is actually no or little difference beweet the two residue vectors.

(I use the word "may" cause it's up to the encoder what to do. This strategy is certainly within the bounds of the specification)

Square Polar Mapping without channel interleaved VQ isn't of any use unless you want to do intensity stereo for the full spectrum (and you don't want to do that, I guess). (see this thread)

SebastianG

Yes, I would need to transmit the angle as side info (as with current IS). This would increase the side information, but with differential encoding I think that it should be reasonable, and I think that it would be worth the coding gain.

What seems strange to me is that it seems that Intensity/Side could be defined as simple generalisation over the two usual joint stereo modes, giving more freedom in the encoding process, but instead we have two different joint stereo tools defined.

You could even go one step further and try to tramsmit ITD cues as well. ;-)

But yeah, I think this should definitely be one thing (among others) to think about when desiging another lossy format. -- If only I was paid to do so ...

S