[sdiy] Information Reduction in Digital media

[Jay Schwichtenberg]

My experience (4 sound cards) with converters says that unless you put them
on good layed out PCB, have high quality support circuitry and a good power
supply they will work like crap and you will hear the difference. These are
extremly critical.

But in reality can you hear the difference if these things are done right
with good converters?

Most 24 bit Crystal & AKM type converters are really only about 21 bits of
real resolution and are called 24 bit because of marketing. Going from 16
bits to 24 bits you can hear a big difference. Going from 44.1 kHz to 96 kHz
sampling there is very little difference, but it is noticable.

One thing that I never got a good answer for when I was doing this was why
the multi-bit converters sounded so much better than the single bits did. My
theory was the internal integration done with multiple bits has less error
than done with a single bit. Multi-bits would then be more tolerable to
jitter and internal voltage errors than single bits.

There are a number of issues working in the digital domain that effect the
sound considerably. Here are a few suggestions when working with your run of
the mill interfaces (M-Audio, Event, Aardvark....) and typical software
(CakeWalk, VST...).
* Most software that converts sound from 96 kHz to 44.1 kHz is pretty bad.
Your better off just working at 44.1 kHz.
* When you mix inside of most computers or digital boxs you are losing bits
and thus dynamic range. It's better to mix in the analog domain if you can.
* Always record as high as possible into a digital box (-1 to -3 dbFS, db
Full Scale). You get better results attenuating a signal than amplifying it
digitally. For example if you record at -10 dbFS and then amplify it by
let's say +6 dbFS you are just shifting the data up one bit. The bottom bit
of resolution is 0 and is lost. The more amplification done the more bits
lost. Now attenuating from -1 dbFS to -4 dbFS you should end up with real
data in the bottom bit.
* If you have to compress something to get a good full scale reading do it
in the analog domain.

Jay


> -----Original Message-----
> From: owner-synth-diy at dropmix.xs4all.nl
> [mailto:owner-synth-diy at dropmix.xs4all.nl]On Behalf Of Grant Richter
> Sent: Friday, October 04, 2002 9:28 AM
> To: Brock Russell; synth-diy
> Subject: Re: [sdiy] Information Reduction in Digital media
>
>
>
> > Grant wrote:
> >> My experiments with 24 bit converters and Spectra-Lab software
> have shown
> >> that even a fairly crappy analog design running from +/- 15 volts
> >> (breadboard), has an unmeasurable signal to noise ratio at 24 bits.
> >>
> >> In other words, even simple analog really does run at a much better
> >> information content than even a good 24 bit A/D converter can resolve.
> >
> > Wow! I want your crappy power supply and breadboard that has a noise
> > floor of better than -144dB.
>
> Believe me, no one was more surprised than myself.
>
> To clarify, using a real time spectrum analyzer with an M-Audio Delta 66
> card, the general noise floor is at -120 dB relative to +4 dB.
> This is with
> a shorted input.
>
> The broad spectrum noise floor appear as a fluctuating horizontal line.
>
> Using this method, you can differentiate between noise signals
> appearing at
> certain frequencies and broadband noise. So while there is an
> obvious spike
> at 60 Hz and it harmonics, the general noise floor remains at the
> A/D level.
>
> To me, this indicates that he white noise content is equal to or less than
> the self noise of the audio card.
>
>