Mic2496 input noise (was: Mic level A/D Converters?)

[alberto]

On 3 Giu 2003, 23:20, Bob Cain > wrote:

> To try and put that in terms more familiar I constructed an example of two
> tracks, one whitenoiseand one signal, whose FFT spectral levels differ by 122
> dB, as Len's graph does, and I find that the RMS level of thenoiseis -85.6 dB
> relative to the signal. Given that the signal in the test is 50 mV that would
> give about 2.62 uV RMS inputnoisefor theMic2496. Anybody who feels that is
> an invalid procedure for estimating total inputnoisefrom such a spectral plot,
> fire away.
>
> Bob
> --
>
> "Things should be described as simply as possible, but no simpler."
>
> A. Einstein

Hello,

I'm answering to this rather old post

http://groups.google.it/group/rec.audio.pro/msg/d2925d686b0c5eb7

because I recently got interested in the purchase of Mic2496 and this
post seemed to be the only place with an estimate of its equivalent
input noise.

Bob, it is an invalid procedure unless you use the same FFT window
size and function that Len used to do the plot.

Assuming the noise is white and is at a level L in the spectrum plot,
the total RMS of the noise expressed in dBFS is L+10log(N/2B), where N
is the FFT size and B is the equivalent noise bandwidth of the window
function (see "Window function" in the wikipedia). I have derived the
formula myself based on the definition of equivalent noise bandwidth
and verified it against white Gaussian noise artificially generated in
CoolEdit Pro. If anyone knows better, I'd appreciate a comment.

Unfortunately Len doesn't disclose either B or N used for his plot,
but we can do an educated guess.

Since the plot is stair-like it is possible, at the lower side of the
frequencies, to single out the frequency bins and estimate their width
Df, hence the FFT size N=Fs/Df where Fs=96000 is the sampling
frequency. As N is always a power of two, after a look at the plot N
must be 16385.

B usually ranges between 1.5 and 2, making 10log(N/2B) probably
between 36 and 37 decibel, so the uncertainty on B doesn't make a big
difference after all. Let's take B=37dB. If the noise we can read on
the plot can be taken as white noise at -137dB, then the total RMS is
-137+37=-100dBFS. Since a full scale sine wave is 575mV per specs,
that means (rounding up a little) an input noise of 5.8uV or -105dB re
1V. If one is interested in an A-weighted value, then take away 2 to
3dB: about 4.6uV or -107dBA re 1V.

All that said, I eventually bought a Mic2496 and could evaluate its
input noise myself, and I found an EIN ranging between 1.3 and 6.0 uV
A-wtd, depending on the sampling frequency and the gain setting. The
procedure I followed and the results I found are online at

http://www.regoloarmonico.com/audio/mic2496/

Ciao
Alberto

[Mark]

On Dec 7, 12:08 pm, alberto > wrote:
> On 3 Giu 2003, 23:20, Bob Cain > wrote:
>
> > To try and put that in terms more familiar I constructed an example of two
> > tracks, one whitenoiseand one signal, whose FFT spectral levels differ by 122
> > dB, as Len's graph does, and I find that the RMS level of thenoiseis -85.6 dB
> > relative to the signal. Given that the signal in the test is 50 mV that would
> > give about 2.62 uV RMS inputnoisefor theMic2496. Anybody who feels that is
> > an invalid procedure for estimating total inputnoisefrom such a spectral plot,
> > fire away.
>
> > Bob
> > --
>
> > "Things should be described as simply as possible, but no simpler."
>
> > A. Einstein
>
> Hello,
>
> I'm answering to this rather old post
>
> http://groups.google.it/group/rec.audio.pro/msg/d2925d686b0c5eb7
>
> because I recently got interested in the purchase of Mic2496 and this
> post seemed to be the only place with an estimate of its equivalent
> input noise.
>
> Bob, it is an invalid procedure unless you use the same FFT window
> size and function that Len used to do the plot.
>
> Assuming the noise is white and is at a level L in the spectrum plot,
> the total RMS of the noise expressed in dBFS is L+10log(N/2B), where N
> is the FFT size and B is the equivalent noise bandwidth of the window
> function (see "Window function" in the wikipedia). I have derived the
> formula myself based on the definition of equivalent noise bandwidth
> and verified it against white Gaussian noise artificially generated in
> CoolEdit Pro. If anyone knows better, I'd appreciate a comment.
>
> Unfortunately Len doesn't disclose either B or N used for his plot,
> but we can do an educated guess.
>
> Since the plot is stair-like it is possible, at the lower side of the
> frequencies, to single out the frequency bins and estimate their width
> Df, hence the FFT size N=Fs/Df where Fs=96000 is the sampling
> frequency. As N is always a power of two, after a look at the plot N
> must be 16385.
>
> B usually ranges between 1.5 and 2, making 10log(N/2B) probably
> between 36 and 37 decibel, so the uncertainty on B doesn't make a big
> difference after all. Let's take B=37dB. If the noise we can read on
> the plot can be taken as white noise at -137dB, then the total RMS is
> -137+37=-100dBFS. Since a full scale sine wave is 575mV per specs,
> that means (rounding up a little) an input noise of 5.8uV or -105dB re
> 1V. If one is interested in an A-weighted value, then take away 2 to
> 3dB: about 4.6uV or -107dBA re 1V.
>
> All that said, I eventually bought a Mic2496 and could evaluate its
> input noise myself, and I found an EIN ranging between 1.3 and 6.0 uV
> A-wtd, depending on the sampling frequency and the gain setting. The
> procedure I followed and the results I found are online at
>
> http://www.regoloarmonico.com/audio/mic2496/
>
> Ciao
> Alberto

copy from link>>>>>
Since I own no fancy instrumentation, all that I did is to hit record
with no microphones attached, process the digital track with some
mathematics, and use the provided values of full-scale sine wave (575
mV RMS) and gain range (40 dB between maximum and minimum). I hope
there are no flaws in this procedure. Any comments are welcome: please
write me an email at or answer to my post. Thank
you.<<<<<

I have two comments...
Instead of doing this test with "no microphones attached" you should
probably connect a "dummy load" to simulate the source impedance of a
typical mic. Generally the noise level of a circuit will be effected
by the impedance it sees at it's input.

Also another nit...RMS power is an incorrect term. RMS voltage is
correct and AVERAGE power is correct. RMS power is not correct.

Also, you can probably create a 20 Hz to 20kHz filter (or whatever
weighting filter you want to user) in the DAW and measure the signal
and noise power (or voltage) directly instead of having to integrate
the spectrum analyzer plots.

Mark

[alberto]

On 7 Dic, 18:48, Mark > wrote:
> I have two comments...
> Instead of doing this test with "no microphones attached" you should
> probably connect a "dummy load" to simulate the source impedance of a
> typical mic. Generally the noise level of a circuit will be effected
> by the impedance it sees at it's input.

I'm not familiar with the standard operations for evaluating the
intrinsic noise. Is there a standard load I should apply? To what pins
in the XLR connector?

> Also another nit...RMS power is an incorrect term. RMS voltage is
> correct and AVERAGE power is correct. RMS power is not correct.

Right, I've corrected it on the website.

> Also, you can probably create a 20 Hz to 20kHz filter (or whatever
> weighting filter you want to user) in the DAW and measure the signal
> and noise power (or voltage) directly instead of having to integrate
> the spectrum analyzer plots.

Right, a quicker alternative. (Oh, I loved my little Perl program
though...)

Meanwhile, I found Mike Rivers' review of Mic2496 on PAR, where a
noise figure of -73dBFS at maximum gain is stated (same as what I
found).

Thanks
Alberto

[Paul Stamler]

"alberto" > wrote in message
...
>
> I'm not familiar with the standard operations for evaluating the
> intrinsic noise. Is there a standard load I should apply? To what pins
> in the XLR connector?

150 ohms, 1% metal film resistor (or wirewound, but they're usually too big
to fit into an XLR). Connect between pins 2 and 3.

Peace,
Paul

[Mike Rivers]

On Dec 11, 11:19 am, alberto > wrote:

> I'm not familiar with the standard operations for evaluating the
> intrinsic noise. Is there a standard load I should apply? To what pins
> in the XLR connector?

If you want to make measurements like most people do, you'll want to
connect a 150 ohm metal film or wirewound resistor between pins 2 and
3, and put the metal shell back on the XLR connector to shield it as
best you can from noise floating around your room.

> > Also another nit...RMS power is an incorrect term. RMS voltage is
> > correct and AVERAGE power is correct. RMS power is not correct.

Someone seems to have that programmed into a function key or
something. It's true, but for the sake of making comparative
measurements, it's not important to worry about this.

> Meanwhile, I found Mike Rivers' review of Mic2496 on PAR, where a
> noise figure of -73dBFS at maximum gain is stated (same as what I
> found).

Well, we must have done about the same thing. I wouldn't argue with
anyone who came up with a figure a few dB off that, however. It's an
eyeball average of the fast responding and peak reading meter in
whatever DAW program I happened to bring up, probably Sound Forge
because it has a big meter.

[alberto]

On 11 Dic, 17:48, "Paul Stamler" > wrote:

> 150 ohms, 1% metal film resistor

A sloppier resistor (say 5%) isn't going to make any difference, is
it? I mean, assuming the Johnson noise from the resistor (0.2 uV) is
negligible compared to the amp's. Just wondering, as the closest store
doesn't carry 150 +/- 1%...

Alberto

[Mike Rivers]

On Dec 13, 4:11 am, alberto > wrote:

> A sloppier resistor (say 5%) isn't going to make any difference, is
> it?

No, it's just a matter of standardization. You can use a short
circuit, too, but expect that your noise measurement will be about 6
dB lower (or maybe it's 3 dB lower) than with a 150 ohm resistor. The
input noise is a function of input current of the first stage. With no
resistance to drop voltage across, there's no input noise, so your
measurement will be of noise after the input stage.

A 150 ohm resistor is "typical" of many microphones, but microphones
aren't 1% tolerance either.

[Arny Krueger]

"Mike Rivers" > wrote in message

> On Dec 13, 4:11 am, alberto > wrote:
>
>> A sloppier resistor (say 5%) isn't going to make any
>> difference, is it?
>
> No, it's just a matter of standardization. You can use a
> short circuit, too, but expect that your noise
> measurement will be about 6 dB lower (or maybe it's 3 dB
> lower) than with a 150 ohm resistor. The input noise is a
> function of input current of the first stage. With no
> resistance to drop voltage across, there's no input
> noise, so your measurement will be of noise after the
> input stage.
>
> A 150 ohm resistor is "typical" of many microphones, but
> microphones aren't 1% tolerance either.

The very common 5% carbon film resistors make such a difference over
open-circuit that further refinements tend to be moot.

I remember when carbon film resistors were over a buck, and rare. Now every
Radio Shack probably still has them for 5 for less than a buck.