Re: Native Device Formats
Re: Native Device Formats
- Subject: Re: Native Device Formats
- From: "Mikael Hakman" <email@hidden>
- Date: Tue, 10 Jun 2008 14:07:28 +0200
- Organization: Datakonsulten AB
That was only an example to illustrate how you easily get a sequence of
digital values, none of which coming even close to FS when your input signal
is at FS. You can make the sequence longer by using sine wave frequency
closer to SR/2. However, your comment shows that we (you and me) are talking
about the same phenomena, which is good.
The article is interesting but hardly new. You sample a signal that in
itself is entirely at FS then you may very well get digital values that all
are well under FS. You then scale up the digital signal so that the digital
values get up to FS. Then you feed this scaled digital signal into a DAC.
This will overflow the DAC because you are feeding it with digital
representation of a signal that is well above FS, even if none of the
digital values is above FS. When you scaled up your digital signal, you
also, conceptually, scaled up your original signal by the same factor.
Therefore your original signal, as perceived by the DAC, is now above FS and
therefore will be clipped. This is what the article talks about. It also
discuses the current mastering habit of using compressors/limiters in order
to push the perceived loudness as high as possible. One example of such
recording where this has been pushed to the very limit is the Bigger Bang by
Rolling Stones - on some tracks the envelope stays at FS all the time. The
article then rightfully points out that you cannot use compressors/limiters,
however good they may be, without introducing distortion.
Your original problem isn't in ADC/DAC, nor in your DAW, but in your scaling
up your digital signal based on minimum/maximum values of digital, sampled
values instead of based on minimum/maximum values of your original,
un-sampled signal. Using high SR and appropriate headroom will make this
problem less pronounced.
It should be possible to write software that analyses a digital signal and
computes the maximum values that would be obtained after DAC - a kind of DAC
simulator. Then you could use these values as a basis for final scaling so
that your final digital signal uses as much of dynamic range as possible but
doesn't overflow a real DAC.
/Mikael
On Monday, June 09, 2008 3:45 PM, Dave wrote:
I agree with your statement that that's how it can come over, though in
that example so long as the signal persists for sufficient time (more than
a millisecond or so since 24-23 = 1kHz) sufficient information will have
been gathered to meet the Sampling Theory criteria. However, this actually
doesn't affect my point - the precise sub multiple example I used was
purely there to make it simple to visualise the process, and it is true in
the more general (non-locked) case, too. See, for instance,
http://www.tcelectronic.com/media/nielsen_lund_2003_overload.pdf for more
details.
Dave
On Jun 9 2008, Mikael Hakman wrote:
If you start sampling e.g. 23 kHz sine wave with 48 kHz SR at an
unfortunate time, you will have a large number of very small positive and
negative values slowly increasing. Only after many cycles the sample time
will coincide with actual peak or low. If this signal ceases before the
sample values reach any significant "height", and you then scale up the
digital values (because the signal is so low), then no headroom in the
world will help you. Therefore you should attenuate/amplify your signal
before ADC in such a way that the actual (not the sampled) peaks and lows
are below FS (and some headroom). Then, whether your ADC gets samples
exactly at these peaks and lows, or not, will be immaterial. Providing
this sampled signal to a DAC will then reproduce the original wave.
On Monday, June 09, 2008 10:56 AM, Dave Malham wrote:
Then, of course, there's the whole hairy question of intersample
"digital overs" where a signal re-constructed from a stream of digital
words can have positive peaks higher (or negative peaks lower) than the
steady state levels the digital words can represent (or the DAC they are
sent to can reproduce). If this seems odd, think of a max level sine
wave at an odd sub-multiple of the sample rate. Under these
circumstances it is possible for the sample points to fall either side
of the peaks of the sine wave, rather than on them, so the peak of the
sine wave will be implicitly, rather than explicitly, represented and
can, in fact, actually be over 0dBFs. This doesn't matter if all you are
doing is storing the signal**, but as soon as you do any processing in
the time/frequency domain it can pop up and bite you.
Dave
**It shouldn't really cause you any problems going to analog, so long as
the designer has done the proper, professional thing and allowed some
headroom in the analog part of the circuit - but that doesn't always
happen. :-(
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Coreaudio-api mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden