Moteurs audio, 32 bits / 64 bits, distinguer le vrai du faux .

Hors sujet : Les miennes ont été un peu repoussées par vos contributions. Encore merci.

Citation :
Absolument, je me place comme toi du point de vue du résultat final, soit en 24bits virgule fixe.

Citation :

Concernant les programmes que tu as écrit, bravo, mais je suis une tanche dans ce domaine, 'n'vais pas pouvoir suivre sur ce terrain. Faut savoir admettre ses limites

Citation : Oui mais les calculs, eux, sont pas fait en 24 bits

Citation :
L'apport du DSD (ou plutôt dirais-je, ce qu'il a de moins) par rapport au PCM est très évident tant sur le plan technique que sur celui du résultat à l'écoute, mais hélas, on ne prend pas cette direction ...

Citation :
Oui Gabou, mais je me répète, au final dans mon fichier 24bits, un son qui s'il était seul serait modulé -80dBfs, ne bénéficie plus que d'une résolution réelle de 14bits. Ce que je dis n'est pas plus compliqué que ça, faut pas chercher plus loin.

Citation :
> As far as I understood DSD is pretty much what comes out of a sigma
> delta converter (ADC)

Yes, but...

> So unlike a high-speed-sampled 1 bit signal which indicates the
> current level (0 or 1) at the sampling instance the DSD signal are
> more like an "upupupdownupdowndownup..." sort of signal which tells
> whether the level at this sampling instance is higher/lower than was
> already accumulated.

True. But think again about 1 bit sampling. If you do it directly, it'll
be horribly nonlinear. If you add enough dither, it'll become linear in
average and work the way you describe, but then the utility signal will
be completely swamped in noise. The big insight of noise
shaping/sigma-delta is that it is possible to shape the noise floor so
that some minute, low-end fraction of the total bandwidth of the 1-bit
signal forms a linear, low noise channel, and the quantization noise is
moved upwards in frequency, while keeping the signal 1-bit. Nothing else
changes, so the utility signal is *still* some time average of the
bitstream. It's just that you don't need as steep a lowpass filter for a
given level of accuracy in reconstruction because the noise is already
concentrated in the high end.

Intuitively it's a bit difficult to see why the sigma-delta modulator
works this way. It's usually analysed statistically and in the frequency
domain, by inserting a dither source inside the quantization loop, then
treating the step quantizer as an independent source of noise, and
finally showing both that the input-output transfer function tends
towards unity in the utility band and that the closed loop response from
the dither input to the output tends to a highpass characteristic
related to the modulator order determining lowpass filter (in your
words, the "accumulator") embedded in the feedback loop.

I'll try to outline one way of getting it in the time domain. If you
look at the modulator, the loop tries to keep the output of the embedded
lowpass/accumulator as close to the input as possible. It always outputs
a correcting bit and assumes that the output will follow in the same
direction, reducing the error. This is a form of negative feedback
similar to that used in analog amplifiers; wrapped around a
nonlinearity, it counteracts it. From another perspective, the
accumulator is the ideal synthesis filter and the loop around it is a
naive optimization process which attempts to synthesize a binary string
which would drive it along the input waveform. The arrangement is stable
as long as the monotonicity assumption holds, and this happens in the
low frequency utility band, so there the process tends towards
transparency with rising sampling frequencies. Above the utility band no
energy was present before and now is, so the entire process tends
towards simple addition of high frequency noise which is designed to be
filtered out by the reconstruction filter.

Now it is relatively easy to see why the loop behaves as it does,
because the delta-sigma modulator can be broken in independent halves.
The lowpass filter inside the loop is basically the optimal, linear
reconstruction filter we're optimizing the bitstream for, and the loop
is the optimizer. Reconstruction is done by lowpass filtering because
that's what the optimizer fitted the bitstream for; the order of the
lowpass filter/accumulator only changes the *kind* of averaging that is
optimal, and not the overarching principle. In simple delta coding the
accumulator and the reconstruction filter are of first order and the
signal is multibit, in sigma-delta they have order upto six and the
signal is 1-bit, but both operators are still just different kinds of
moving averages, capable of emulating one another to a high degree.

Since LTI processing does not shift energy from one frequency band to
another, LTI filters can always be applied directly to DSD as though it
was 1-bit PCM. Any high frequency, shaped noise will stay up there and
be cut out upon reconstruction. Of course any filters will cause the
delicate cancellation between the utility signal and the added noise to
fail, so that the intermediate signals will no longer be 1-bit but will
require normal PCM bit widths. Any nonlinear processing can also cause
interference between the utility band and the noise component, and
corrupt the utility signal. That's why processing PCM is so much easier.
But it's easy to see that going into PCM requires nothing more than
filtering out the noise and (possibly) decimating into the utility
bandwidth. (Sony's solution is 8-bit intermediates without decimation
and reshaping back to DSD for delivery.)


BTW, the last way of looking at delta-sigma modulation is more general
than it seems and quite powerful. It allows us to formulate exactly the
conditions the system relies on, delineate its boundaries, and consider
alternatives. For example, we see that the optimizer is a greedy,
heuristic one and relies on a highly simplified assumption (a monotonic
input-output relationship). This is where the instability of high order
(i.e. more efficient) delta-sigma modulators comes from, so going with a
high order filter and replacing the optimizer with something a bit more
sophisticated can in theory boost the efficiency of bitstream coding and
probably sidestep the Lipschitz-Vanderkooy critique of DSD because a
dither linearized feedback loop is no longer present. It's also evident
from the structure that in principle delta-sigma isn't limited to
lowpass reconstruction -- just replace the embedded filter with
something else and use an optimizer which can cope. Et cetera.

> So in other words it has a 'memory' in it. From what I understood
> there is a specific bitstream that indicates 0.

Sort of. As long as we reconstruct by LTI lowpass filtering and the
transmitted signal is 1-bit, the bitstream will need to alternate
symmetrically about the mean and be as high in frequency as possible.
Depending on which metric you use, the best approximation to zero output
will come from an alternating series of zeroes or ones (least mean
square error: energy is concentrated as high in frequency as it can go
where the reconstruction filter has cut off maximally) or from a
top-heavy noise (supremum norm in the frequency domain: the energy
distribution will have to inversely follow the reconstruction filter
response).

> Given that (unlike PCM) DSD doesn't have the concept of zero there has
> to be a way of telling "ok we're going to start at 0 again". So I
> guess one would have to detect this bitstream to put your PCM signal
> to 0 and then create the signal based on the up/downs and then
> downsample based on that.

Both PCM and DSD have a concept of zero in the same sense: the mean
between the highest and lowest values directly representable. In PCM
systems that will mostly lie between the two middle sampling steps, so
neither PCM nor DSD can realize a zero exactly but only as a time
average of a quantized signal. In practice both systems have finite
memory, so decoding can start anywhere and the resulting average will
converge to the right value. Nowadays high quality PCM will be in-band
noise shaped before delivery as well, so the reasoning about residual
noise levels and zero patterns is almost identical for the two systems.

PCM and DSD really aren't that different if you push them hard.

j'ai une question :

es ce que sa derange la qualiter d'enregistrement si mon

mac pro est en 32 bits ( snow leopard )

le logic studio en 64 bits

enregistrement en 24 bits

?????????

merci !