Core Sound Mic2496 is sold as a quiet preamp and A/D converter. However, specifications on the website do not include any figure representing its equivalent input noise. Len Moskowitz doesn't like the EIN specification very much, as he has explained for example on rec.audio.pro here and here, and also to me in a private communication. I won't enter into this argument here, but I believe that a published EIN figure can't do any harm. Since I have just purchased a Mic2496 from Core Sound I have done some measurements and will now share my results.

As you will see, at 44 and 48 kHz sampling frequency and at minimum
gain I have found an EIN of **-113 dBV A-wtd** i.e. *2.1 µV
A-wtd*. At higher sampling frequencies it seems to get up to 3.5 dB
noisier in the left channel, 9 dB in the right channel. Tuning up the
gain, the SNR seems to get remarkably better, and at maximum gain I
found an EIN of **-118 dBV A-wtd** i.e. *1.3 µV A-wtd*
at any sampling frequency.

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 aren't flaws in this procedure. However,
it seems to be confirmed by the fact that the results I've found agree
with Mike Rivers' review on Pro Audio Review, where an EIN of -73dBFS
at maximum gain is stated. Any comments are welcome: please write me
an email at alblonghi@yahoo.com
or answer to my
post. Thank you.

- I have connected my Mic2496 to an M-Audio Microtrack 24/96 through the S/PDIF interface with a coax cable. Both Mic2496 and Microtrack are powered by batteries. No microphone is attached to the Mic2496. See picture below Please note that my Microtrack is the first release, not the newer Microtrack II. Firmware version is 1.4.6. Please note that this measurement is not standard as I didn't put any load on the mic inputs. (From what I gather on the usenets, standard data might be some 6dB better that I show here, but I take no responsibility of this estimate. I show what I measured and I tell you how I did it - use this data as you see fit.)
- In such a configuration I have recorded at 24 bits for 4+ minutes at all sampling frequencies: 44, 48, 88, 96, and at both minimum and maximum gain, for a total of eight takes. All following steps are repeated for each such take.
- In CoolEdit I have thrown away the first minute and trimmed the end to leave 3 minutes exact.
- In CoolEdit I have highlighted the whole take, done the
*Statistics*analysis, and taken note of the*Total RMS Power*expressed in dB relative to FS (full-scale) sine wave. - In CoolEdit I have highlighted the whole take, done the
*Frequency analysis*with the following parameters:- FFT size: 32768

window: Blackman-Harris*Scan*to average over the whole length of the take. Then I did*Copy to clipboard*and pasted the data into a text file. - Finally, I have processed the spectrum with my Perl script
`spectrum.pl`. This script:- Calculates the RMS of the track expressed in dB relative to the RMS of a full-scale sine wave.
- Calculates the A-weighted level dBA (also relative to the RMS of a full-scale sine wave). For frequencies f>20kHz the A curve is taken equal to –∞ (minus infinity).
- Rearranges the spectrum so to have data points uniform in log f: this is convenient when plotting the spectrum in logscale, so the graphic is not burdened with overdense points in the high end of the spectrum. When the new binning is sparser the data is power averaged and the plot is smoother; where the new binning is denser, a stepping shows up.

All following results are in dB relative to a FS (full-scale) sine wave.

Sampl./Gain RMS L RMS R 2444/min -105.89 -106.00 2444/max -70.62 -69.85 2448/min -105.74 -105.67 2448/max -70.30 -69.61 2488/min -100.08 -94.77 2488/max -67.24 -66.57 2496/min -99.71 -94.46 2496/max -66.95 -66.23

Note that the RMS levels calculated bySampl./Gain RMS L RMS R | dBA L dBA R 2444/min -106.01 -106.09 | -108.69 -108.77 2444/max -70.63 -69.86 | -73.26 -72.44 2448/min -105.85 -105.80 | -108.84 -108.85 2448/max -70.32 -69.63 | -73.28 -72.54 2488/min -100.11 -94.78 | -105.27 -99.87 2488/max -67.26 -66.58 | -72.99 -72.18 2496/min -99.74 -94.47 | -105.27 -99.91 2496/max -66.96 -66.24 | -73.12 -72.21

To convert these figures to mV or dBV you can use the value specified as the FS sine wave at minimum gain in the specs of Mic2496 on the website: 575 mV RMS (that is -4.8 dB re. 1V).

For example, taking the value of -108.7 dBFS A-wtd as a typical level at minimum gain, then the EIN can be evaluated of -4.8 − 108.7 = -113.5 dBV A-wtd, that is 2.1 µV A-wtd. At the higher sampling rates the noise level seems to raise considerably: about 3.5 dB in the left channel and 9 dB in the right channel.

However, it also appears that the noise floor doesn't scale with the gain: tuning the Mic2496's gain from min to max, the signal gets 40 dB (per specs, also roughly verified on my own device within about 1 dB), while the noise raises only 35 dB or less, so the SNR is better with higher gains. At maximum gain, the noise floor is about -73 dBFS A-wtd for all sampling frequencies.

In the light of that, and assuming the FS sine wave at maximum gain is 40 dB lower than that at minimum gain, the EIN value at maximum gain is -4.8 − 40 − 73 = -118 dBV A-wtd (with some rounding), i.e. 1.3 µV A-wtd.

spectrum-2444-max_gain-blackmannharris32768.txt

spectrum-2444-min_gain-blackmannharris32768.txt

spectrum-2448-max_gain-blackmannharris32768.txt

spectrum-2448-min_gain-blackmannharris32768.txt

spectrum-2488-max_gain-blackmannharris32768.txt

spectrum-2488-min_gain-blackmannharris32768.txt

spectrum-2496-max_gain-blackmannharris32768.txt

spectrum-2496-min_gain-blackmannharris32768.txt

For a PDF version of the plots click here.