How are S/N ratios determined and what do they really mean?

[[email protected]]

If a sound card claims a certain S/N ratio, say 100 db, I had it in my
head that this was supposed to mean that the sound card itself is
capable of a noise floor of -100 db as you might see on the meter on
Soundforge etc. I.e. that under ideal circumstances you could have 100
db worth of sound before you run into self-generated noise floor. Is
this not correct?

Using a Sound Blaster Audigy Platinum EX which is rated at 100db, I'm
wondering under what circumstances it's supposed to have a noise floor
that low. I can't seem to duplicate it unless the meters on Soundforge
and Cakewalk are grossly incorrect. Around -84 db seems to be the best
I can do, and that's with the mixer levels at 2%, the lowest they'll
go without being off altogether, certainly not a useful level for
recording purposes.

Thanks

[William Sommerwerck]

The signal-to-noise ratio for digital devices is usually references to peak
level -- all bits "on".

The number also depends on weighting. C weighting (if I recall correctly)
attenuates the lowest octaves to mimic the ear's loss of LF response at low
listening levels. Weighted S/N measurements will therefore always be better
than unweighted measurements, though you're unlikely to actually hear any
difference.

[Mike Rivers]

wrote:
> If a sound card claims a certain S/N ratio, say 100 db, I had it in my
> head that this was supposed to mean that the sound card itself is
> capable of a noise floor of -100 db as you might see on the meter on
> Soundforge etc. I.e. that under ideal circumstances you could have 100
> db worth of sound before you run into self-generated noise floor. Is
> this not correct?

Usually it's not correct. They come up with those numbers as
creatively as they can so that people like you who look at them will
see their brand as better than the competition. Most have figured out
that there's noise contributed by the analog circuitry and A/D and D/A
conversion and that you can't just say that at 16 bits the S/N ration
is 96 dB, or at 24 bits, 144 dB.

They do make actual measurements but rarely tell you the conditions
under which they were made. They always have some termination on the
input (most people doing such measurements at home forget this,
leaving the "antenna input" open), and almost always, the frequency
response of the noise measurement is "weighted." A-Weighting is
common, which cuts out most of the low frequencies (which gets rid of
any hum that creeps in through the power supply or weak ground paths,
and cuts the highs, reducing the contribution of hiss and noise from
the various clocks associated with sound card operation. Typically an
A-weighted noise measurement will be (when looking at a single number)
15 dB better than a broadband noise measurement.

> Using a Sound Blaster Audigy Platinum EX which is rated at 100db, I'm
> wondering under what circumstances it's supposed to have a noise floor
> that low. I can't seem to duplicate it

You can try putting shorting plugs in the input jacks, setting the
gain (if there's a control) to minimum, and recording some noise. Then
roll off the lows and highs, and measure the noise on playback. You
might come close. But that won't tell you how quiet it is under
"normal" circumstances.

On the other hand, how much do you need? Most pop music has a dynamic
range of 10 dB or less (you can always "fade to digital black" between
songs) and few listening environments can support a dynamic range
greater than about 50 dB, so
80 dB is good enough for anyone who can only afford an Audigy.

Best not to worry about it.

[Scott Dorsey]

> wrote:
>If a sound card claims a certain S/N ratio, say 100 db, I had it in my
>head that this was supposed to mean that the sound card itself is
>capable of a noise floor of -100 db as you might see on the meter on
>Soundforge etc. I.e. that under ideal circumstances you could have 100
>db worth of sound before you run into self-generated noise floor. Is
>this not correct?

It means that the difference between the _maximum_ operating level and
the lowest possible operating level is 100 dB.

Now, there is a lot of equipment out there whose coloration changes at
different levels. The older Mackie mixers are good examples.... you
really don't want to use the top 20 dB of operating range. So the
effective S/N that you get in the real world is 20 dB lower than the
measured system S/N.

>Using a Sound Blaster Audigy Platinum EX which is rated at 100db, I'm
>wondering under what circumstances it's supposed to have a noise floor
>that low. I can't seem to duplicate it unless the meters on Soundforge
>and Cakewalk are grossly incorrect. Around -84 db seems to be the best
>I can do, and that's with the mixer levels at 2%, the lowest they'll
>go without being off altogether, certainly not a useful level for
>recording purposes.

How high will it go? If you are reading levels here in dBFS, I would
indeed call that much less than 100 dB S/N. If you are reading levels
in something else, and there is a peak level higher than 0dBwhatever,
it could well be higher.

Note also that the numbers are probably rated for line inputs either
shorted or terminated with a 600 ohm source resistor. If you use any
other configuration, you will get poorer numbers.

Note that the the theoretical maximum range of a 16 bit system is 96 dB.
Which is really a whole lot.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Arny Krueger]

> wrote in message
ups.com

> If a sound card claims a certain S/N ratio, say 100 db, I
> had it in my head that this was supposed to mean that the
> sound card itself is capable of a noise floor of -100 db
> as you might see on the meter on Soundforge etc. I.e.
> that under ideal circumstances you could have 100 db
> worth of sound before you run into self-generated noise
> floor. Is this not correct?

Sort of.

> Using a Sound Blaster Audigy Platinum EX which is rated
> at 100db, I'm wondering under what circumstances it's
> supposed to have a noise floor that low.

Input loaded with a typical source impedance, gains set optimally, and noise
measurement taken over a reasonable bandwidth like 20-20 KHz @ -3 dB down, 6
or 12 dB/ocatave roll-offs. Or, A-weighting if so stipulated.

> I can't seem to
> duplicate it unless the meters on Soundforge and Cakewalk
> are grossly incorrect. Around -84 db seems to be the best
> I can do, and that's with the mixer levels at 2%, the
> lowest they'll go without being off altogether, certainly
> not a useful level for recording purposes.

Noise measurements are only comparable if a standard measurement bandwidth
is used. Sounds like you're just taking the natural bandwith of the sound
card as your measurement bandwith. That makes little sense, since it
penalizes cards with good bandwidth, and gives the advantage to the card
with the worst frequency response.

[[email protected]]

On Feb 3, 8:41 am, "Mike Rivers" > wrote:
> wrote:
> > If a sound card claims a certain S/N ratio, say 100 db, I had it in my
> > head that this was supposed to mean that the sound card itself is
> > capable of a noise floor of -100 db as you might see on the meter on
> > Soundforge etc. I.e. that under ideal circumstances you could have 100
> > db worth of sound before you run into self-generated noise floor. Is
> > this not correct?
>
> Usually it's not correct. They come up with those numbers as
> creatively as they can so that people like you who look at them will
> see their brand as better than the competition.
>
> almost always, the frequency
> response of the noise measurement is "weighted." A-Weighting is
> common, which cuts out most of the low frequencies (which gets rid of
> any hum that creeps in through the power supply or weak ground paths,
> and cuts the highs, reducing the contribution of hiss and noise from
> the various clocks associated with sound card operation. Typically an
> A-weighted noise measurement will be (when looking at a single number)
> 15 dB better than a broadband noise measurement.


Yup, it's indicated as being A-weighted, and the -85db or so would
seem to jive with what I'm observing - i.e. 100 minus 15. Of course, I
can get that only at levels that are way too low to record at.

By "broadband", do you mean essentially "as per what the Soundforge
record meter indicates"?

[Mike Rivers]

On Feb 6, 12:17 pm, wrote:

> By "broadband", do you mean essentially "as per what the Soundforge
> record meter indicates"?

I meant over the full frequency response range (20 Hz or lower to just
a tad under half the sample rate), which is essentially what the DAW
meter indicates.