How to measure EIN (Equivalent Input Noise) of an interface's Mic preamp

[Rja4000]

Weighting and non-flat noise densities can be done but it's a bit of a pain as you basically have to divide your frequency range of interest into a bunch of chunks, each with roughly constant noise density and weighting factor. (People did this for phonopre noise calculation in the '70s.) You not only have to pick the ranges but also enter the noise density as well, which quickly becomes tedious when doing it by hand. If you are happy with a flat noise density approximation, my RMS summing and unsumming calculators should do the job.

If you want A-weighting, I'd suggest using something like REW to determine the difference to flat 20-20k on your measurement spectrum (no need to reinvent the wheel if someone else has already done the work, and it keeps you from having to type a bazillion values). I have also used an A-weighting EQ preset for Audacity's filter curve effect before. (Note that A-weighting is only considered valid up to 20 kHz, and any substantial content above this may need to be disposed of using a sharp lowpass filter.) The wave file analysis of RMAA should also do the job, I think... Plus I reckon there's various metering plugins for your favorite DAW that would work.

It's actually quite easy to do.
If there is an interest, I can provide an Excel file I'm using to do that.

You export the FFT from REW.
You copy/paste the exported file values into the Excel sheet.
You set your low/high limits values.

The Excel will computes the RMS energy between your limits, with various weightings (un-weighted, A-weighted and CCIR-2k weighted) using the original weighting formulae.

The Excel includes flat energy values per default, so you may already try and get the difference of level for each weighting method for a flat noise.

 

[nanook]

If you are happy with a flat noise density approximation, my RMS summing and unsumming calculators should do the job.

I think flat noise densities would suffice.
They usually do not increase with frequency within the audio band and it needs a fairly high 1/f corner to get visible in the final result.

 

[Philbo King]

Well, with the short at the input, that's exactly what the EIN gives you.
The preamp's self noise.

This how I've always done it.

 

[Blumlein 88]

So just measured my Antelope Audio Zen Tour. I've measured it before. I come up with a surprising 129 db EIN at 65.8 db of gain. Max is spec'd at 65 db, but it is a bit more than that. This with 150 ohm resistor. It is 127 db by 20 db gain and above. Shorted I get 133.2 dbu which seems too high to fit with the EIN. The noise is very nearly flat. No upward tilt until 400 hz and below and not much of that. Max input with 0 gain is 9.4 dbu (spec is 8.2 dbu)

I did quickly work thru the 4 mic pres. This one is mic pre 2 which is the quietest. The others vary slightly, but in general measure 1 db less for EIN.

 

[Rja4000]

Max is spec'd at 65 db, but it is a bit more than that.

Gain doesn't mean anything for a digital interface, where you don't have access at the analog preamp's output.
What do you mean ?

 

[Rja4000]

Shorted I get 133.2 dbu which seems too high to fit with the EIN.

With -133.2 dBu for short, you should get -128.9 dBu for 150 ohm.
So it looks coherent.

 

[Blumlein 88]

Gain doesn't mean anything for a digital interface, where you don't have access at the analog preamp's output.
What do you mean ?

What I mean by that is the labeled gain vs actual gain change. Like if I go from 0 db to a labeled 65 db without changing input level does the device record 65 db higher signal or something different. In this case, 65 db raised the level recorded in dbFS by 65.8 db. I've been doing the same test on my Babyface for instance and it gives exactly 65 db change for an indicated 65 db gain. Both these interfaces let you control gain by even 1 db steps. Should that be labeled gain in dbFS?

 

[Blumlein 88]

Here are numbers for the RME Babyface Pro FS.
EIN is also 129 dbu for 65 db gain. Max input is 8.5 dbu (spec is 8 dbu). Shorted was -132 dbu.

It did have lesser EIN readings for gain settings below 40 db. The noise floor was even closer to being flat or white in character. No real tilt up until below 100 hz. Not much then.

 

[Blumlein 88]

Here are results for a Focusrite 18i20 1st generation. Two different sets. There are 2 mic inputs on the front and 6 on the rear.

The front:
Max input of 7.9 dbu. Gain of 50 db. EIN 127.5 dbu shorted 128.3 dbu. This measurement might be a bit corrupted. This device has low -145 dbu idle tones every 1 khz all the way to 21 khz. I've seen this before.

The rear:
Max input of 7.9 dbu. Gain of 50 db. EIN of 124 dbu shorted 124.9 dbu. Same 1 khz spaced idle tones only 8 or so db higher in level.

 

[Rja4000]

Should that be labeled gain in dbFS?

Thanks for the explanation.
I call that "Gain range", personally

 

[Blumlein 88]

Well here is an interesting one, and I agree about gain range. A Universal Audio Arrow. Same product is now called a Solo, but no changes.

EIN is -130 dbu. Shorted -134 dbu. Max input is only 2 dbu. The software shows minimum gain of 10 db with a gain range of 55 db. So +10 db to +65 db. Yet like I suspected it actually is only 55 db which is an example of how when you connect to an ADC it is somewhat arbitrary.

But of the 4 devices I checked out today which one is better for say a Shure SM7b? That mic puts out about -56 dbu @ 94 db SPL. Assuming you wanted peaks up to 120 db SPL then those peaks would be -30 dbu. So the Arrow with its lower max input (or inversely the more sensitive input) even with lower gain range of 55 db would manage to hit 0 dbFS with the least gain and along with its good EIN the least noise.

 

[Rja4000]

Well here is an interesting one, and I agree about gain range. A Universal Audio Arrow. Same product is now called a Solo, but no changes.

EIN is -130 dbu. Shorted -134 dbu. Max input is only 2 dbu. The software shows minimum gain of 10 db with a gain range of 55 db. So +10 db to +65 db. Yet like I suspected it actually is only 55 db which is an example of how when you connect to an ADC it is somewhat arbitrary.

But of the 4 devices I checked out today which one is better for say a Shure SM7b? That mic puts out about -56 dbu @ 94 db SPL. Assuming you wanted peaks up to 120 db SPL then those peaks would be -30 dbu. So the Arrow with its lower max input (or inversely the more sensitive input) even with lower gain range of 55 db would manage to hit 0 dbFS with the least gain and along with its good EIN the least noise.

That's why I plead for using sensitivity (another word for max input level or Full scale level) instead of gain for digital interfaces.

You should compare EIN for those interfaces at the "gain" where sensitivity is around -24dBu or so @0 dBFS in your use case.
The interface with best EIN at that "gain" will be the one giving the less noise.

Actually, that's close to what I called "reference gain" in my example of EIN vs gain in the top post here.
I choose reference gain as the gain that gives 0dBFS for 100mV (-18dBu), so you may look at the EIN for 6dB higher gain than that.

This is with 150 ohm load.
If we were ploting the shorted measurement, we shouldn't see this horizontal asymptote, which is due to the resistor's own noise.

 

[Blumlein 88]

I don't know if you have looked at Julian Krause's idea of what he calls "line up" gain. Another way to compare.
As to which would be best at your levels, the Antelope and UA Arrow would be that. Neither changes much at all until you get to quite low levels of gain range, and both have the better EIN as well. The Arrow would squeak out a slight win there though the Antelope would be a touch better at somewhat lower gain levels. Close to a tie in effect.

I was thinking of the Shure SM7b which falls in those levels you mention. OTOH, a Shure KSM44a would come close to clipping the Arrow at the same spl levels that work with the SM7b. The Arrow does have a 20 db pad.

While interesting, I think even the Scarlett 18i20 would work well enough ambient noise is sure to render it of no difference. We are blessed to have gear this good.

Still a bit of a conundrum as to what test level makes the most overall sense for mic preamps and noise levels. Seems most reputable manufacturers have us well taken care of in effect. I do think lesser gear tends to fall off more at lower levels of the gain range. So maybe 10 mv and 100 mv is a better test.

 

[AnalogSteph]

*obligatory reminder to please state measurement bandwidth (and/or weighting) alongside results - there's almost a dB of difference between 20 kHz flat and a typical 48 kHz bandwidth*

I don't know if you have looked at Julian Krause's idea of what he calls "line up" gain. Another way to compare.

dBFS out / dBu or dBV in is about the only good way of specifying system gain in a black-box device such as an audio interface.

I choose reference gain as the gain that gives 0dBFS for 100mV (-18dBu), so you may look at the EIN for 6dB higher gain than that.

This is with 150 ohm load.

That Yamaha AD8HR must have a very funky gain structure for its curve to look like that. It almost looks like input amplifier gain is switched in ~12 dB steps (or perhaps it's actually an input attenuator? Almost makes more sense), followed by another stage with much finer ones. Quirky. The theory of it using an input attenuator would be supported by the fact that the (mic) input can accept levels up to +30 dBu when mic input amplifiers tend to have a minimum gain of 0 to +10 dB (I very much doubt that the unit would be running internal levels of +30 to +40 dBu). The noise penalty of this approach is quite obvious.

 

[Blumlein 88]

44.1 khz sampling rate. I used 20-20,000 hz for the measurements. Sorry, meant to include that earlier. No weighting. A-weighting gave about 2.5 db better numbers in each case.

 

[Rja4000]

*obligatory reminder to please state measurement bandwidth (and/or weighting) alongside results - there's almost a dB of difference between 20 kHz flat and a typical 48 kHz bandwidth*

Agree
And if not specified (which shouldn't be the case), we should refer to 20Hz-20kHz

dBFS out / dBu or dBV in is about the only good way of specifying system gain in a black-box device such as an audio interface.

I would refer to "dBu @FS" or "Vrms @FS" (which, obviously, is 1/(FS/Vrms) )

That Yamaha AD8HR must have a very funky gain structure for its curve to look like that. It almost looks like input amplifier gain is switched in ~12 dB steps (or perhaps it's actually an input attenuator? Almost makes more sense), followed by another stage with much finer ones. Quirky. The theory of it using an input attenuator would be supported by the fact that the (mic) input can accept levels up to +30 dBu when mic input amplifiers tend to have a minimum gain of 0 to +10 dB (I very much doubt that the unit would be running internal levels of +30 to +40 dBu). The noise penalty of this approach is quite obvious.

I think this is 2 attenuator stages.
Your hear the relay switching at the 2 abrupt changes.

 

[Blumlein 88]

Here is a video where Mr. Krause explains his idea of line up gain. It is not in disagreement with what anyone here is saying, he is just trying to make it easier for someone who doesn't understand how this stuff plays out when an ADC is involve. The main parts explaining his idea of "line up gain" for comparing interfaces with differing parameters is found between the 4 minute and 7 minute mark.

 

[Blumlein 88]

dBFS out / dBu or dBV in is about the only good way of specifying system gain in a black-box device such as an audio interface.

So just using the 4 interfaces I measured yesterday.
Antelope audio Zen Tour.
dbu for 0 dbFS @ 0 gain is 9.4 dbu.
dbu for 0 dbFS @ 65 db (which is actually 65.8) gain is - 56.4 dbu.

The BabyFace Pro FS.
dbu for 0 dbFS @ 0 gain is 8.5 dbu.
dbu for 0 dbFS @ 65 db gain is -56.5 dbu.

Focusrite 18i20.
dbu for 0 dbFS @ 0 gain is 7.9 dbu.
dbu for 0 dbFS @ 50 db gain is -42.5 dbu.

UA Arrow.
dbu for 0 dbFS @ 0 gain is 2.0 dbu.
dbu for 0 dbFS @ 65 db (which is really 55 db) gain is -53.0 dbu.

Which nevertheless can confuse people. If you were using an insensitive ribbon or similar you might think the two devices giving 0 dbFS with the lowest dbu rating would be the best. Yet it wouldn't. The Arrow would be because while it gives max output with a slightly higher signal level it does so with less gain and the noise would be less apparent.

Yet that still isn't complete either until you give consideration for how EIN and gain figure into this. The number might make you think the top three are nearly equivalent, but the inferior EIN of the Focusrite will mean it won't function as well in regards to noise level.

 

[AnalogSteph]

Which nevertheless can confuse people. If you were using an insensitive ribbon or similar you might think the two devices giving 0 dbFS with the lowest dbu rating would be the best. Yet it wouldn't. The Arrow would be because while it gives max output with a slightly higher signal level it does so with less gain and the noise would be less apparent.

It's funny you should say that this is confusing, as I don't think I follow...

Given the same EIN, the interface with highest max system gain is obviously going to have the highest noise floor - but also the loudest signal. Now since EIN is not exactly constant in real life, the interesting part is how they all stack up when set to the exact same system gain(s)... there may be some surprises to be had. Probably not any real big ones though. (This is what @Rja4000's EIN vs. relative gain graphs are all about, they're just offset on the X axis. Or offset + mirrored if sensitivity is your metric of choice instead of gain. Either metric works as one is merely the inverse of the other, translating to the negative when expressed in dB, so really it's more a matter of convention than anything else.)

As for gain by itself, let's say you're a livestreamer with an SM7B (-59 dBV/Pa) and you want to be hitting -6 dBFS peak, a fairly common scenario these days. Spoken word from 10-20 cm (4-8") barely tend to get into the low 90s of dB SPL in my experience. Slightly misusing my DIY Acoustic Level Calibration calculator and entering data for the 18i20 with a -6 dBFS reference level gives you an input SPL of 102.7 dB being required, so clearly that one would be a miss. You'd want roughly 10 dB more gain to be comfortable, so any of the other three would be preferred in this application.

The traditional music recordist is going to be much less bothered about hitting absolute levels (aside from staying well out of clipping, with common target peak levels ranging from -12 to -18 dBFS) - things can easily be made louder in post if need be. Under these circumstances you only need enough maximum gain for analog noise to still swamp ADC noise by at least 10, ideally 20 dB. If your sensitivity at max gain is -42.5 dBu, then a -127 dBu EIN ends up at -84.5 dBFS, far above the 18i20's ADC noise floor (the 2nd gen sports like 109 dB(A) worth of dynamic range, so even if that's 106 unweighted we're still talking 20+ dB lower). This explains why the Focusrite's sensitivity was deemed entirely adequate by its designers.

Going back to my livestreaming example, a sensitivity of -53 dBu with an EIN of -130 dBu means a noise floor 77 dB below FS (or a 77 dB dynamic range). If this is all you commonly need, you can get away with a very modest ADC. Not coincidentally, there are a number of inexpensive USB interfaces targeted at podcasting that are using geriatric 16-bit codecs whose ADCs make it to 87 dB(A) worth of dynamic range on a good day. Their digital filters aren't that great either but it's not terribly important on spoken word anyway. These are quite arguably better-suited to the job than a unit with fancy converters but a noisy preamp.

 

[Rja4000]

Here is a video where Mr. Krause explains his idea of line up gain. It is not in disagreement with what anyone here is saying, he is just trying to make it easier for someone who doesn't understand how this stuff plays out when an ADC is involve. The main parts explaining his idea of "line up gain" for comparing interfaces with differing parameters is found between the 4 minute and 7 minute mark.

Actually, he's explaining the same I'm trying to explain.
And his bargraphs are similar to my sensitivity graphs, except he's just looking at the max gain value, and doesn't say anything about overall max value, which I also include.

This

Is the same than this

but without the left value (which I think is also interesting)