Michael R. Kesti wrote:
Second, a 16-bit system provides 16-bits of resolution regardless of
the level to which it is driven. Resolution is the ability of a
system's ability to accurately measure its input signal. A 16-bit
system can measure its input single with an accuracy of one part in
65,536 or about 0.01038 dB. A 16-bit converter therefore resolves the
difference between -11.98962 dBfs and -12.00000 dBfs as well as it
measures the difference between -0.01038 dBfs and 0.00000 dBfs.

if I limit my 16-bit ADC to half its bit capacity, (IE use only 8-bits,)
how am I still getting 16-bits of resolution? couldn't I get equivalent
results by using an 8-bit ADC and scaling my input signal appropriately?

One needs no more dynamic range in a digital system than that of the
signals one is recording. A 16-bit ADC provides slightly more than 96
dB of dynamic range. If one allocates that range for a nominal level
of -20 dBfs, the remaining 76 dB to between the noise floor and
nominal level probably exceeds that of even the finest studio
environments and their analog signal chains. Any additional resolution
is simply unused.

aren't you forgetting quantization noise?

--
Aaron J. Grier | "Not your ordinary poofy goof." |
"someday the industry will have throbbing frontal lobes and will be able
to write provably correct software. also, I want a pony." -- Zach Brown

"Aaron J. Grier" wrote in message
...
Michael R. Kesti wrote:

if I limit my 16-bit ADC to half its bit capacity, (IE use only 8-bits,)
how am I still getting 16-bits of resolution? couldn't I get equivalent
results by using an 8-bit ADC and scaling my input signal appropriately?

Yep, the same could be said for a 64 bit system using only the last 8 bits.
So what?

TonyP.

Scott Dorsey wrote:
Yes, but pretty much every DAC today _is_ a sigma-delta system.

in which case added noise on the output is necessary to avoid a
cumulative DC offset, correct?

--
Aaron J. Grier | "Not your ordinary poofy goof." |
"someday the industry will have throbbing frontal lobes and will be able
to write provably correct software. also, I want a pony." -- Zach Brown

TonyP wrote:
"Aaron J. Grier" wrote in message
...
Michael R. Kesti wrote:

if I limit my 16-bit ADC to half its bit capacity, (IE use only 8-bits,)
how am I still getting 16-bits of resolution? couldn't I get equivalent
results by using an 8-bit ADC and scaling my input signal appropriately?

Yep, the same could be said for a 64 bit system using only the last 8
bits.
So what?

exactly.

Michael was claiming that a 16 bit system utilizing only 14 bits was
still a 16 bit system. technically it may be, but you're losing
potential resolution by doing so.

to tie this back into the 24 vs 16 discussion, I posit that it's easier
to get 16 bits of resolution with a 24 bit system than it is with a 16
bit system.

--
Aaron J. Grier | "Not your ordinary poofy goof." |
"someday the industry will have throbbing frontal lobes and will be able
to write provably correct software. also, I want a pony." -- Zach Brown

"Aaron J. Grier" wrote:

Michael R. Kesti wrote:
Second, a 16-bit system provides 16-bits of resolution regardless of
the level to which it is driven. Resolution is the ability of a
system's ability to accurately measure its input signal. A 16-bit
system can measure its input single with an accuracy of one part in
65,536 or about 0.01038 dB. A 16-bit converter therefore resolves the
difference between -11.98962 dBfs and -12.00000 dBfs as well as it
measures the difference between -0.01038 dBfs and 0.00000 dBfs.

if I limit my 16-bit ADC to half its bit capacity, (IE use only 8-bits,)
how am I still getting 16-bits of resolution?

Because resolution is the measure of the ability to resolve changes in
amplitude that are small with respect to the greatest amplitude that can
be represented rather than with respect to the greatest amplitude present.

Try it this way: We get 6 dB of dynamic range per bit of resolution.
Dynamic range is the difference between the noise floor and that greatest
amplitude that can be represented. A 16-bit system results in a noise floor
96 dB below full scale and a 1-bit system's noise floor is only 6 dB below
full scale. If it is true that recording peaks no greater than -12 dBfs
results in a 14-bits of resolution then it must also be true that recording
random noise at -95.98962 dBfs results in a 1-bit of resolution. When I
play that recording back, however, the resulting noise is certainly far
less than 6 dB below the full scale. This can only be because there is
still more than 1 bit of resolution in the recording.

couldn't I get equivalent
results by using an 8-bit ADC and scaling my input signal appropriately?

You can get equivalent results only if the signal to noise ratio of that
input signal is less than 48 dB. Below some amount of resolution, the
dynamic range of the digital medium becomes less than that of the analog
signal being represented and the results are no longer equivalent.

One needs no more dynamic range in a digital system than that of the
signals one is recording. A 16-bit ADC provides slightly more than 96
dB of dynamic range. If one allocates that range for a nominal level
of -20 dBfs, the remaining 76 dB to between the noise floor and
nominal level probably exceeds that of even the finest studio
environments and their analog signal chains. Any additional resolution
is simply unused.

aren't you forgetting quantization noise?

I don't think so. How do you feel that I am?

--
================================================== ======================
Michael Kesti | "And like, one and one don't make
| two, one and one make one."
| - The Who, Bargain

(Aaron J. Grier) wrote in message ...
TonyP wrote:
"Aaron J. Grier" wrote in message
...
Michael R. Kesti wrote:

if I limit my 16-bit ADC to half its bit capacity, (IE use only 8-bits,)
how am I still getting 16-bits of resolution? couldn't I get equivalent
results by using an 8-bit ADC and scaling my input signal appropriately?

Yep, the same could be said for a 64 bit system using only the last 8
bits.
So what?

exactly.

Michael was claiming that a 16 bit system utilizing only 14 bits was
still a 16 bit system. technically it may be, but you're losing
potential resolution by doing so.

to tie this back into the 24 vs 16 discussion, I posit that it's easier
to get 16 bits of resolution with a 24 bit system than it is with a 16
bit system.

Those are good comments. The real story is that in a 24 bit real data,
there are some real bits trhat carry the music, and also some
"marketing bits" that contribute nothing but noise. It is convinient
to use 2 bytes for 16 bits, and 3 bytes for 24 bits (a byte is 8 bits)
but that is the only value in 24 bits. The few lower bits are noise!

So how many bits are real? Look at a non weighted dynamic range spec,
or noise floor spec. My AD122 MKII yields 126dB dynamic range, which
is the best by a long margin! That is about 21 real bits! The last 3
bits are garbage.

Now, there are a lot of 24 bit machines that will bearly yield 17
bits, and they have about 7 bits of garbage. Most of the 24 bit AD's
are speced with A weighting, which is not a one to one comparison...

BR
Dan Lavry
Lavry Engineering

dan lavry wrote:

[16 bit vs "24 bit" converters]

So how many bits are real? Look at a non weighted dynamic range spec,
or noise floor spec. My AD122 MKII yields 126dB dynamic range, which
is the best by a long margin! That is about 21 real bits! The last 3
bits are garbage.

126 dB is impressive. I was just wondering - why is it (apparently) so
hard to get less noise than this? Is it because of the thermal noise of
the circuit itself or are there no better converters available yet?

Now, there are a lot of 24 bit machines that will bearly yield 17
bits, and they have about 7 bits of garbage. Most of the 24 bit AD's
are speced with A weighting, which is not a one to one comparison...

Assuming that the noise is evenly distributed, what would the "real" SNR
be, given 102dB(A) SNR? Around 96 maybe?

Johann
--
np: Abigor - Nachthymnen (From The Twilight Kingdom) - 2 - Scars In The
Landscape Of God

"Aaron J. Grier" wrote:

Michael was claiming that a 16 bit system utilizing only 14 bits was
still a 16 bit system.

A 16-bit system is, of course a 16-bit system regardless of the level you
record, but this is not what I claimed. Instead, I stated "a 16-bit system
provides 16-bits of resolution regardless of the level to which it is
driven."

technically it may be, but you're losing
potential resolution by doing so.

No, you are not. You are not using all of the system's available dynamic
range, but its resolution, potential or otherwise, is not lost. Resolution
limits available dynamic range but failing to use all of the dynamic range
does not limit resolution.

Suppose, using a 16-bit system, I make a one hour recording that never
exceeds -12 dBfs and contains several moments of "silence" that are
actually the -82 dBfs residual noise of my studio, mic, and mic preamp.
(Yeah, I know, I have a great room and some very good gear!) What is the
resolution, in bits, of this recording?

Now suppose that this recording includes one peak that drives the system
to full scale for one sample. Does the resolution of this recording differ
from the first? Is it in any way a better recording?

to tie this back into the 24 vs 16 discussion, I posit that it's easier
to get 16 bits of resolution with a 24 bit system than it is with a 16
bit system.

This may be, but primarily because system designers put more effort into
the design of 24-bit systems, especially their analog sections.

--
================================================== ======================
Michael Kesti | "And like, one and one don't make
| two, one and one make one."
| - The Who, Bargain

Johann Burkard wrote:

126 dB is impressive. I was just wondering - why is it (apparently) so
hard to get less noise than this? Is it because of the thermal noise of
the circuit itself or are there no better converters available yet?

Both, perhaps, but thermal noise issue is the limiting issue. If we scale
an ADC's input so that +24 dBu (about 21.8 volts) drives the ADC to full
scale, then 126 dB lower is a mere 6.2 microvolts. This is well into the
range of thermal noise and is currently quite difficult to achieve.

--
================================================== ======================
Michael Kesti | "And like, one and one don't make
| two, one and one make one."
| - The Who, Bargain

Michael R. Kesti wrote:

if I limit my 16-bit ADC to half its bit capacity, (IE use only
8-bits,) how am I still getting 16-bits of resolution?

Because resolution is the measure of the ability to resolve changes in
amplitude that are small with respect to the greatest amplitude that
can be represented

this is _potential_ resolution, which counts for nothing if it's never
used.

rather than with respect to the greatest amplitude present.

if the max amplitude is never reached, the increased resolution of the
system has no benefits.

by your reasoning I could digitize with an 8 bit ADC, zero pad the other
bits, and claim 16 bit resolution, but it ain't so.

Try it this way: We get 6 dB of dynamic range per bit of resolution.

sure.

Dynamic range is the difference between the noise floor and [the]
greatest amplitude that can be represented.

also agreed.

A 16-bit system results in a noise floor 96 dB below full scale and a
1-bit system's noise floor is only 6 dB below full scale.

by the 6dB/bit rule, OK.

If it is true that recording peaks no greater than -12 dBfs results in
a 14-bits of resolution then it must also be true that recording
random noise at -95.98962 dBfs results in a 1-bit of resolution.

on our 16-bit ADC. right. I'm still with you at this point.

When I play that recording back, however, the resulting noise is
certainly far less than 6 dB below the full scale. This can only be
because there is still more than 1 bit of resolution in the recording.

even if the other 15 bits are zero? couldn't I take that noisy LSB and
play it back with a 1-bit DAC and attenuate the output down and get the
same result? or take the 14-bits worth of signal, and play them through
a 14-bit DAC, similarly attenuated down, to get the equivalent of the
unattenuated 16-bit signal?

what have those unused bits in my 16 bit system gained me from a lower
bit system with prescaling?

couldn't I get equivalent results by using an 8-bit ADC and scaling
my input signal appropriately?

You can get equivalent results only if the signal to noise ratio of
that input signal is less than 48 dB.

here I don't follow you.

it appears you are assuming that the dBFS on our 8- and 16-bit ADCs are
the same scale? I'm not. maybe that's where my confusion lies.

Below some amount of resolution, the dynamic range of the digital
medium becomes less than that of the analog signal being represented
and the results are no longer equivalent.

I can't parse this either. which resolution are you talking about?

nevermind the quantization noise; I'll come back to it once I get this
squared away.

--
Aaron J. Grier | "Not your ordinary poofy goof." |
"someday the industry will have throbbing frontal lobes and will be able
to write provably correct software. also, I want a pony." -- Zach Brown

Michael R. Kesti wrote:
"Aaron J. Grier" wrote:

Michael was claiming that a 16 bit system utilizing only 14 bits was
still a 16 bit system.

I stated "a 16-bit system provides 16-bits of resolution regardless of
the level to which it is driven."

if I take an 8-bit ADC and wire it to a 16-bit bus, ground the top 8
data lines, is this now a 16-bit system? it is always providing 16 bits
of data, irregardless of the input level. the linearity past the 8th
bit is horrible, but it does always produce 16 bits of data.

technically it may be, but you're losing potential resolution by
doing so.

No, you are not. You are not using all of the system's available
dynamic range, but its resolution, potential or otherwise, is not
lost.

so there would be no benefit by raising the gain of our
14-actual-bits-encoded-by-16-bit-ADC system by 6dB? or even 12dB?

(I'm think I'm starting to understand what you're getting at...)

Resolution limits available dynamic range but failing to use all of
the dynamic range does not limit resolution.

failing to use all of the dynamic range does not _necessarily_ limit
resolution, but it can, IF the noise floor of the signal being encoded
is below the LSB of the system. conversely, after some point in adding
bits, you're only digitizing noise.

Suppose, using a 16-bit system, I make a one hour recording that never
exceeds -12 dBfs and contains several moments of "silence" that are
actually the -82 dBfs residual noise of my studio, mic, and mic
preamp. (Yeah, I know, I have a great room and some very good gear!)
What is the resolution, in bits, of this recording?

it could be expressed without any information loss in 14-bits. (and
perhaps less...)

Now suppose that this recording includes one peak that drives the
system to full scale for one sample. Does the resolution of this
recording differ from the first?

if that one-bit sample is indeed a glitch and not part of the signal
you're trying to record, then you've still only got 14-bits of actual
data.

Is it in any way a better recording?

we have a:

16-bit system recording max level -12dBfs, noise floor of -82dBfs

could be recorded with the same results by a

14-bit system recording max level -0dBfs, noise floor of -70dBfs.

the dynamic range of the input signal has not changed between the two
ADCs. in fact we could encode such a signal onto 12 bits without losing
any information, since a 12 bit system has 72dB of dynamic range,
correct?

what am I missing?
--
Aaron J. Grier | "Not your ordinary poofy goof." |
"someday the industry will have throbbing frontal lobes and will be able
to write provably correct software. also, I want a pony." -- Zach Brown

"Les Cargill" wrote in message...

FWIW, N-Track will save mixes as 32 float files, and it's provided
for ( type 3 ) in the .WAV RIFF spec. Works fine in snack, too.

Just give me a darned tape machine. ;-)

DM

"Aaron J. Grier" wrote in message
...
to tie this back into the 24 vs 16 discussion, I posit that it's easier
to get 16 bits of resolution with a 24 bit system than it is with a 16
bit system.

If we have a real world converter/analog stage with 90dB DNR, how will the
resolution be increased by saving 24 bits of data instead of 16?
For a converter with 96 dB DNR, then you would be correct. There are quite
a few 24 bit systems with less than 96dB DNR though.

TonyP.

"Michael R. Kesti" wrote in message
...
to tie this back into the 24 vs 16 discussion, I posit that it's easier
to get 16 bits of resolution with a 24 bit system than it is with a 16
bit system.

This may be, but primarily because system designers put more effort into
the design of 24-bit systems, especially their analog sections.

I contend you are both wrong when measuring a Sound Blaster Audigy. :-)

TonyP.

"Aaron J. Grier" wrote:

snip

so there would be no benefit by raising the gain of our
14-actual-bits-encoded-by-16-bit-ADC system by 6dB? or even 12dB?

No, there would be no benefit because you would apply that gain to the
signal's noise component, too. All you will have done is shifted the
dynamic range toward the more significant bits. You will not have
increased the recorded signal's dynamic range.

(I'm think I'm starting to understand what you're getting at...)

And I think that I'm beginning to understand you, too. I see that the
resolution offered by digital recording is potential in nature. The
96 dB of dynamic range offered by 16-bit sampling usually exceeds the
dynamic range of the analog signal that is sampled and mapped to that
digital representation. This is a good thing as one wants the dynamic
range of a recording medium to exceed that of the signal being recorded
so that, on playback, the signal's noise dominates rather than the medium's.

What those who insist that increasing the gain to "use all of the bits"
and "achieve greater resolution" fail to see is that the gain is applied
to the noise as well as the peaks. As long as the analog signal's dynamic
range still fits in the digital medium's, every additional high level bit
used by adding gain is offset by using a low level bit to express noise.
Unless additional level is added without additional noise there can be no
additional resolution.

snip

--
================================================== ======================
Michael Kesti | "And like, one and one don't make
| two, one and one make one."
| - The Who, Bargain

Aaron J. Grier wrote:
Michael R. Kesti wrote:
Second, a 16-bit system provides 16-bits of resolution regardless of
the level to which it is driven. Resolution is the ability of a
system's ability to accurately measure its input signal. A 16-bit
system can measure its input single with an accuracy of one part in
65,536 or about 0.01038 dB. A 16-bit converter therefore resolves
the difference between -11.98962 dBfs and -12.00000 dBfs as well as
it measures the difference between -0.01038 dBfs and 0.00000 dBfs.

if I limit my 16-bit ADC to half its bit capacity, (IE use only
8-bits,) how am I still getting 16-bits of resolution?

Yes, unless you apply gain to the 8 bits worth of signal.

couldn't I
get equivalent results by using an 8-bit ADC and scaling my input
signal appropriately?

Yes, as long as the 8 bit component never get amplified above it's original
binary value.

geoff

Johann Burkard wrote:
dan lavry wrote:

[16 bit vs "24 bit" converters]

So how many bits are real? Look at a non weighted dynamic range spec,
or noise floor spec. My AD122 MKII yields 126dB dynamic range, which
is the best by a long margin! That is about 21 real bits! The last 3
bits are garbage.

126 dB is impressive. I was just wondering - why is it (apparently) so
hard to get less noise than this? Is it because of the thermal noise
of the circuit itself or are there no better converters available yet?

Ever heard a real room quieter than, say, -80dB ?

geoff

"Michael R. Kesti" wrote in message...

What those who insist that increasing the gain to "use all of the bits"
and "achieve greater resolution" fail to see is that the gain is applied
to the noise as well as the peaks. As long as the analog signal's dynamic
range still fits in the digital medium's, every additional high level bit
used by adding gain is offset by using a low level bit to express noise.
Unless additional level is added without additional noise there can be no
additional resolution.


Thank you, sir....


--
David Morgan (MAMS)
http://www.m-a-m-s DOT com
Morgan Audio Media Service
Dallas, Texas (214) 662-9901
_______________________________________
http://www.artisan-recordingstudio.com

"Geoff Wood" -nospam wrote in message
...
Ever heard a real room quieter than, say, -80dB ?

I've never heard a room that *produces* wide band noise, have you?

TonyP.

TonyP wrote:
"Geoff Wood" -nospam wrote in message
...
Ever heard a real room quieter than, say, -80dB ?

I've never heard a room that *produces* wide band noise, have you?

TonyP.

No, but crank the gain up enough and you'll probably get a slightly bassy
version of what you describe.

geoff

Geoff Wood wrote:
Johann Burkard wrote:
126 dB is impressive.

Ever heard a real room quieter than, say, -80dB ?

Ever heard the term "headroom"?

Johann
--
Immer diese beleidigenden Spinner die sich auch noch fuer den Nabel der
Welt halten. Ab in Deinen Sandkasten husch husch, Du kleiner Netzterrorist.
(*Tönnes zu mir in )

Johann Burkard wrote in message ...
dan lavry wrote:

[16 bit vs "24 bit" converters]

126 dB is impressive. I was just wondering - why is it (apparently) so
hard to get less noise than this? Is it because of the thermal noise of
the circuit itself or are there no better converters available yet?

Assuming that the noise is evenly distributed, what would the "real" SNR
be, given 102dB(A) SNR? Around 96 maybe?

Johann


This is a big subject. The noise is mostly evenly distributed, and
certainly resistors yield a flat with frequency. Semiconductors are
mostly flat, though there is a rise in the very low frequencies 0-30Hz
or 0-50Hz or to 100Hz, depends on the device. In fact there is more
than one mechanism for that low freq. noise rise. Semiconductor noise
comes in 2 flavors as well, there is noise voltage and there is noise
current, the later interacts with the resistors in the circuit to
increase the overall noise voltage.

2 quick points to make:

First, it was very difficult to make 126-127dB dynamic range AD (21
bits). It would be almost impossible to approach 24 real bits because
3 more bits is 1/8 the noise I have!

Second, I would not try for more than 21 bits AD until such time that
we can feed it a signal that is noise free beyond 21 bits. Fact is,
the signal from the best mic pres driven by real mics (even the lowest
drive resistance mics) will not yield much better than about 125dBu of
noise. The mic pre needs to amplify the signal to say 24dBu peak to
peak (for pro gear) and in doing so it also amplifies the noise. Say
your gain is 30dB, than you end up with 125dBu(mic pre input noise) –
30dB (Gain) + 24dBu (max signal) = 119dB dynamic range. That is less
than 20 bits. Say the mic signal is so strong that you only need 20dB
micpre gain, now you are near 21 bits. That is 21 bits when screaming
directly into a high efficiency mic…

I made the MKII to accommodate the most extreme cases and 21 bits is
good enough. I too have marketing bits in my 24 bit machine. In my
case it is 3 marketing bits. It least it is not 7…

When I first broke the 20 bit barrier (un weighted noise floor) in
1995, I called my AD the AD122 (my dynamic range was 122dB). I proudly
called it a 20 bit converter. Next year in the AES I was the only one
with a 20 bit machine. Everyone else had a 24 bit machine for sale,
most of them around 17 bits performance if that. In 1997 I had the
software revised to pass on the last 3 bits of noise. I was thinking
of adding 100 bits of least significant bits of noise and have the
only 124 bit machine around 

BR
Dan Lavry
Lavry Engineering

Johann Burkard wrote:
Geoff Wood wrote:
Johann Burkard wrote:
126 dB is impressive.

Ever heard a real room quieter than, say, -80dB ?

Ever heard the term "headroom"?

sure. but this discussion has spurred thinking about this the other way
around.

let's say you have a room with a noise floor of 30dBSPL. to accurately
capture any signal up to 150dBSPL, you'd need a range of 120dB. 120dB
of range can be expressed in 20 bits.

it would seem that instead of setting 0dBFS to the (expected) peak of
the signal, one could adjust the preamp so that the noise floor of the
signal being recorded tickles the LSB of the ADC.

--
Aaron J. Grier | "Not your ordinary poofy goof." |
"someday the industry will have throbbing frontal lobes and will be able
to write provably correct software. also, I want a pony." -- Zach Brown

"Aaron J. Grier" wrote in message

Johann Burkard wrote:
Geoff Wood wrote:
Johann Burkard wrote:
126 dB is impressive.

Ever heard a real room quieter than, say, -80dB ?

Ever heard the term "headroom"?

sure. but this discussion has spurred thinking about this the other
way around.

let's say you have a room with a noise floor of 30dBSPL. to
accurately capture any signal up to 150dBSPL, you'd need a range of
120dB. 120dB of range can be expressed in 20 bits.

it would seem that instead of setting 0dBFS to the (expected) peak of
the signal, one could adjust the preamp so that the noise floor of the
signal being recorded tickles the LSB of the ADC.

That might make some sense were it not for the fact that the noise floor of
the signal being recorded probably has one kind of spectral shape and the
converter has a different one.

I believe the reason for say 24 bits on a dac is to get the intermediate
values you need for the interpolation filter used for oversampleing.
Oversampeling is used to raise the effective sampeling frequency so that it
is much easier to filter out the noise from the DAC. On the other hand in
recording it allows you to do more more processing in the digital domain
without introducing too much error into your final 16 bit product.

Bob


"Aaron J. Grier" wrote in message
...
Johann Burkard wrote:
Geoff Wood wrote:
Johann Burkard wrote:
126 dB is impressive.

Ever heard a real room quieter than, say, -80dB ?

Ever heard the term "headroom"?

sure. but this discussion has spurred thinking about this the other way
around.

let's say you have a room with a noise floor of 30dBSPL. to accurately
capture any signal up to 150dBSPL, you'd need a range of 120dB. 120dB
of range can be expressed in 20 bits.

it would seem that instead of setting 0dBFS to the (expected) peak of
the signal, one could adjust the preamp so that the noise floor of the
signal being recorded tickles the LSB of the ADC.

--
Aaron J. Grier | "Not your ordinary poofy goof." |
"someday the industry will have throbbing frontal lobes and will be able
to write provably correct software. also, I want a pony." -- Zach Brown

16 and 24 bit refer, of course, to the old system of measurement in considering
payment of CD's. In other words, a 16 bit Cd cost $2.00 and a 24 bit Cd cost
$3.00, seeing that two bits is a quarter (25 cents). And if you don't take my
word for it, ask Dr. Science, he knows more than you do.

Wow! More than 16 hours later and nobody has responded to this one. I
guess I'll take a stab at it. At least I'll inspire responses if I get
it wrong!

"Aaron J. Grier" wrote:

Johann Burkard wrote:

Ever heard the term "headroom"?

sure. but this discussion has spurred thinking about this the other way
around.

let's say you have a room with a noise floor of 30dBSPL. to accurately
capture any signal up to 150dBSPL, you'd need a range of 120dB. 120dB
of range can be expressed in 20 bits.

it would seem that instead of setting 0dBFS to the (expected) peak of
the signal, one could adjust the preamp so that the noise floor of the
signal being recorded tickles the LSB of the ADC.

If by "preamp" you intend "microphone preamp", then no, this is not what
you want to do. Instead, you want to adjust your analog signal chain the
same way you would if you weren't going to convert to digital, for the same
reasons you always have. You need to account for the mic's sensitivity and
source's SPL to achieve an acceptable balance between noise and overload.

Once your analog chain is optimally set, however, you could adjust the
ADC's sensitivity so that, when the source is "silent", the digital side's
LSB is just being "tickled." This would ensure that no source information
is lost below the medium's noise floor and provide maximum headroom for
peaks. I might chose to tickle a couple or three LSBs, though, to ensure
that source noise dominates. (Legroom anybody?)

There is, however, a practical problem to this approach. There are, to
my knowledge, no digital audio meters that allow one to see when only the
last few LSB's are being tickled. Even the extended range products from
Durrough bottom out at -60 dBfs. Any ideas?

--
================================================== ======================
Michael Kesti | "And like, one and one don't make
| two, one and one make one."
| - The Who, Bargain

"Michael R. Kesti" wrote in message ...
Johann Burkard wrote:

126 dB is impressive. I was just wondering - why is it (apparently) so
hard to get less noise than this? Is it because of the thermal noise of
the circuit itself or are there no better converters available yet?

Both, perhaps, but thermal noise issue is the limiting issue. If we scale
an ADC's input so that +24 dBu (about 21.8 volts) drives the ADC to full
scale, then 126 dB lower is a mere 6.2 microvolts. This is well into the
range of thermal noise and is currently quite difficult to achieve.

I hate to be a pain, but 24dBu is 12.27V rms (34.72V peak to peak) and
for 24 bits you are allowed .731uV.
But you are correct about the figure 6.2uV.

BR
Dan Lavry

"Michael R. Kesti" wrote in message ...
Johann Burkard wrote:

126 dB is impressive. I was just wondering - why is it (apparently) so
hard to get less noise than this? Is it because of the thermal noise of
the circuit itself or are there no better converters available yet?

Both, perhaps, but thermal noise issue is the limiting issue. If we scale
an ADC's input so that +24 dBu (about 21.8 volts) drives the ADC to full
scale, then 126 dB lower is a mere 6.2 microvolts. This is well into the
range of thermal noise and is currently quite difficult to achieve.

I hate to be a pain, but 24dBu is 12.27V rms (34.72V peak to peak) and
for 24 bits you are allowed .731uV.
But you are correct about the figure 6.2uV.

BR
Dan Lavry

dan lavry wrote:

"Michael R. Kesti" wrote in message ...

Both, perhaps, but thermal noise issue is the limiting issue. If we scale
an ADC's input so that +24 dBu (about 21.8 volts) drives the ADC to full
scale, then 126 dB lower is a mere 6.2 microvolts. This is well into the
range of thermal noise and is currently quite difficult to achieve.

I hate to be a pain, but 24dBu is 12.27V rms (34.72V peak to peak) and
for 24 bits you are allowed .731uV.
But you are correct about the figure 6.2uV.

Oops. I use http://www.sengpielaudio.com/calculator-db-volt.htm for this
sort of thing. Given one of dBu, dBV, or voltage it computes and displays
the other two. It computes that +24 dBu is equal to +21.8 dBV and 12.28
volts. I simply read the wrong field!

Thanks for the correction, Dan.

--
================================================== ======================
Michael Kesti | "And like, one and one don't make
| two, one and one make one."
| - The Who, Bargain

dan lavry wrote:

"Michael R. Kesti" wrote in message ...

Both, perhaps, but thermal noise issue is the limiting issue. If we scale
an ADC's input so that +24 dBu (about 21.8 volts) drives the ADC to full
scale, then 126 dB lower is a mere 6.2 microvolts. This is well into the
range of thermal noise and is currently quite difficult to achieve.

I hate to be a pain, but 24dBu is 12.27V rms (34.72V peak to peak) and
for 24 bits you are allowed .731uV.
But you are correct about the figure 6.2uV.

Oops. I use http://www.sengpielaudio.com/calculator-db-volt.htm for this
sort of thing. Given one of dBu, dBV, or voltage it computes and displays
the other two. It computes that +24 dBu is equal to +21.8 dBV and 12.28
volts. I simply read the wrong field!

Thanks for the correction, Dan.

--
================================================== ======================
Michael Kesti | "And like, one and one don't make
| two, one and one make one."
| - The Who, Bargain

"sycochkn" wrote in message ink.net...
I believe the reason for say 24 bits on a dac is to get the intermediate
values you need for the interpolation filter used for oversampleing.
Oversampeling is used to raise the effective sampeling frequency so that it
is much easier to filter out the noise from the DAC. On the other hand in
recording it allows you to do more more processing in the digital domain
without introducing too much error into your final 16 bit product.

Bob

Of course it is so in an ideal world, where the data comming in to the
DAC was made to 24 bits accuracy, and the DAC can make a signal to 24
bits accuracy.

Say your AD is good to 16 bits. Say each individual sample is only
good to 16 bits. Now you want to upsample - add the values in between
samples. You are using data that is limited to 16 bits as referance
points to generate new data. Do you think the new data (the upsamled
values) will ever be 24 bits accurate? It will not. If you are running
a geographical survey and your markers are only accurate to say 1
foot, all the points on the map will be limited by that 1 foot
tolerance...

You are not totaly wrong. You are correct in theory, and to some
degree in practice. To what degree? To the degree determind by the
accuracy of the AD and DA...

There has been a lot of misleading hype about 24 bit upsampling dac.
The big point to remember is that a 24bit upsamplin dac does not yield
24 bits performance when fed say a 16 bit signal. It could not do such
a miracle. The statment should be:

If one had a 24 bit DAC,than one could up sample the the data to be a
gopod 16 bits at faster rate (and that upsampling is done for easier
job of analog filtering). If the DAC is less than 24 bits acurate, the
generated data adds less than apropriate values. But there is no way
to ever get 24 bits performance with less than 24 bit data and 24 bits
conversion!

BR
Dan Lavry

"sycochkn" wrote in message ink.net...
I believe the reason for say 24 bits on a dac is to get the intermediate
values you need for the interpolation filter used for oversampleing.
Oversampeling is used to raise the effective sampeling frequency so that it
is much easier to filter out the noise from the DAC. On the other hand in
recording it allows you to do more more processing in the digital domain
without introducing too much error into your final 16 bit product.

Bob

Of course it is so in an ideal world, where the data comming in to the
DAC was made to 24 bits accuracy, and the DAC can make a signal to 24
bits accuracy.

Say your AD is good to 16 bits. Say each individual sample is only
good to 16 bits. Now you want to upsample - add the values in between
samples. You are using data that is limited to 16 bits as referance
points to generate new data. Do you think the new data (the upsamled
values) will ever be 24 bits accurate? It will not. If you are running
a geographical survey and your markers are only accurate to say 1
foot, all the points on the map will be limited by that 1 foot
tolerance...

You are not totaly wrong. You are correct in theory, and to some
degree in practice. To what degree? To the degree determind by the
accuracy of the AD and DA...

There has been a lot of misleading hype about 24 bit upsampling dac.
The big point to remember is that a 24bit upsamplin dac does not yield
24 bits performance when fed say a 16 bit signal. It could not do such
a miracle. The statment should be:

If one had a 24 bit DAC,than one could up sample the the data to be a
gopod 16 bits at faster rate (and that upsampling is done for easier
job of analog filtering). If the DAC is less than 24 bits acurate, the
generated data adds less than apropriate values. But there is no way
to ever get 24 bits performance with less than 24 bit data and 24 bits
conversion!

BR
Dan Lavry

In article writes:

There is, however, a practical problem to this approach. There are, to
my knowledge, no digital audio meters that allow one to see when only the
last few LSB's are being tickled. Even the extended range products from
Durrough bottom out at -60 dBfs. Any ideas?

Yeah. Don't worry about what the meters are telling you down at that
end of the scale. Set the gain so you have the amount of headroom you
want, and let the bottom end lie where it falls. You can't do anything
about it anyway. Why worry.

I've always said that meters are a guide, not a rule.

--
I'm really Mike Rivers )
However, until the spam goes away or Hell freezes over,
lots of IP addresses are blocked from this system. If
you e-mail me and it bounces, use your secret decoder ring
and reach me he double-m-eleven-double-zero at yahoo

In article writes:

There is, however, a practical problem to this approach. There are, to
my knowledge, no digital audio meters that allow one to see when only the
last few LSB's are being tickled. Even the extended range products from
Durrough bottom out at -60 dBfs. Any ideas?

Yeah. Don't worry about what the meters are telling you down at that
end of the scale. Set the gain so you have the amount of headroom you
want, and let the bottom end lie where it falls. You can't do anything
about it anyway. Why worry.

I've always said that meters are a guide, not a rule.

--
I'm really Mike Rivers )
However, until the spam goes away or Hell freezes over,
lots of IP addresses are blocked from this system. If
you e-mail me and it bounces, use your secret decoder ring
and reach me he double-m-eleven-double-zero at yahoo

"Michael R. Kesti" wrote in message ...
There is, however, a practical problem to this approach. There are, to
my knowledge, no digital audio meters that allow one to see when only the
last few LSB's are being tickled. Even the extended range products from
Durrough bottom out at -60 dBfs. Any ideas?

My AD122 and my Model 3000S digital optimiser do offer such feature.
The normal display mode (very large and clear bar graph), and give you
an LED for each 1dB from 0 to -30dBFS, than 5dB steps to -55dB.
One can go and set an offset of -30dB, -60dB or -90dB. So say you set
the offset to -90dB. Than your top bar graph lamp idicates -90dB. The
next lamp is -91dB.... all the way to -120dB, and a few lamps for
-125, -130...

Also, the model 3000S shows you what is coded in the data stream, and
it has a mode where the display turns into a bit display (up to 24
lamps flickering, and with 16 active bits only 16 flicker...)

Very few people use those modes, but I have been offering it for nealy
10 years. I also have a short tech note on how to use it for noise
measurnments on my site www.lavryengineering.com I belive it is a
short single page at the end of the AD122 MKII product manuel. I did
not include that page with the 3000S manuel, but the extanded range
feature is certainly a part of the many features I included (such as
test tones, THD+N and more) that are not used often. Folks buy the
Model 3000S mostly for sample rate conversion and dither with noise
shaping, but the feature you mentioned is there.

BR
Dan Lavry
Lavry Engineeing

BR
Dan Lavry

"Michael R. Kesti" wrote in message ...
There is, however, a practical problem to this approach. There are, to
my knowledge, no digital audio meters that allow one to see when only the
last few LSB's are being tickled. Even the extended range products from
Durrough bottom out at -60 dBfs. Any ideas?

My AD122 and my Model 3000S digital optimiser do offer such feature.
The normal display mode (very large and clear bar graph), and give you
an LED for each 1dB from 0 to -30dBFS, than 5dB steps to -55dB.
One can go and set an offset of -30dB, -60dB or -90dB. So say you set
the offset to -90dB. Than your top bar graph lamp idicates -90dB. The
next lamp is -91dB.... all the way to -120dB, and a few lamps for
-125, -130...

Also, the model 3000S shows you what is coded in the data stream, and
it has a mode where the display turns into a bit display (up to 24
lamps flickering, and with 16 active bits only 16 flicker...)

Very few people use those modes, but I have been offering it for nealy
10 years. I also have a short tech note on how to use it for noise
measurnments on my site www.lavryengineering.com I belive it is a
short single page at the end of the AD122 MKII product manuel. I did
not include that page with the 3000S manuel, but the extanded range
feature is certainly a part of the many features I included (such as
test tones, THD+N and more) that are not used often. Folks buy the
Model 3000S mostly for sample rate conversion and dither with noise
shaping, but the feature you mentioned is there.

BR
Dan Lavry
Lavry Engineeing

BR
Dan Lavry

dan lavry wrote:

"Michael R. Kesti" wrote in message ...
There is, however, a practical problem to this approach. There are, to
my knowledge, no digital audio meters that allow one to see when only the
last few LSB's are being tickled. Even the extended range products from
Durrough bottom out at -60 dBfs. Any ideas?

My AD122 and my Model 3000S digital optimiser do offer such feature.
The normal display mode (very large and clear bar graph), and give you
an LED for each 1dB from 0 to -30dBFS, than 5dB steps to -55dB.
One can go and set an offset of -30dB, -60dB or -90dB. So say you set
the offset to -90dB. Than your top bar graph lamp idicates -90dB. The
next lamp is -91dB.... all the way to -120dB, and a few lamps for
-125, -130...

Also, the model 3000S shows you what is coded in the data stream, and
it has a mode where the display turns into a bit display (up to 24
lamps flickering, and with 16 active bits only 16 flicker...)

Very few people use those modes, but I have been offering it for nealy
10 years. I also have a short tech note on how to use it for noise
measurnments on my site www.lavryengineering.com I belive it is a
short single page at the end of the AD122 MKII product manuel. I did
not include that page with the 3000S manuel, but the extanded range
feature is certainly a part of the many features I included (such as
test tones, THD+N and more) that are not used often. Folks buy the
Model 3000S mostly for sample rate conversion and dither with noise
shaping, but the feature you mentioned is there.

That's excellent stuff, Dan. Have you any thoughts on setting levels
using the noise floor as has been discussed in this thread?

--
================================================== ======================
Michael Kesti | "And like, one and one don't make
| two, one and one make one."
| - The Who, Bargain

dan lavry wrote:

"Michael R. Kesti" wrote in message ...
There is, however, a practical problem to this approach. There are, to
my knowledge, no digital audio meters that allow one to see when only the
last few LSB's are being tickled. Even the extended range products from
Durrough bottom out at -60 dBfs. Any ideas?

My AD122 and my Model 3000S digital optimiser do offer such feature.
The normal display mode (very large and clear bar graph), and give you
an LED for each 1dB from 0 to -30dBFS, than 5dB steps to -55dB.
One can go and set an offset of -30dB, -60dB or -90dB. So say you set
the offset to -90dB. Than your top bar graph lamp idicates -90dB. The
next lamp is -91dB.... all the way to -120dB, and a few lamps for
-125, -130...

Also, the model 3000S shows you what is coded in the data stream, and
it has a mode where the display turns into a bit display (up to 24
lamps flickering, and with 16 active bits only 16 flicker...)

Very few people use those modes, but I have been offering it for nealy
10 years. I also have a short tech note on how to use it for noise
measurnments on my site www.lavryengineering.com I belive it is a
short single page at the end of the AD122 MKII product manuel. I did
not include that page with the 3000S manuel, but the extanded range
feature is certainly a part of the many features I included (such as
test tones, THD+N and more) that are not used often. Folks buy the
Model 3000S mostly for sample rate conversion and dither with noise
shaping, but the feature you mentioned is there.

That's excellent stuff, Dan. Have you any thoughts on setting levels
using the noise floor as has been discussed in this thread?

--
================================================== ======================
Michael Kesti | "And like, one and one don't make
| two, one and one make one."
| - The Who, Bargain