[castleofargh]

On a more serious note, I'm curious about how to find info about the noise floor, as it's the biggest thing that is bothering me right now with my DAC, and nobody seems to bother to mention it in reviews, as if the specs sheet tells it so there's no point in even mentioning it.

The specs sheet for my DAC says this:

"Noise Floor 7 µV RMS (A-weight), max gain"

Can anyone explain what that means, and as a rule of thumb, how does it stand? Is it very below average, very good, just a marketing term used to confuse people and don't tell anything without more info on the reference used, just a made up pseudoscientific term, etc.?

Just to add a bit, on max gain, the DAC I use can output "2.9 V rms / +11.5 dBu." Can someone kindly explain how to use this in calculations to find the noise floor?

Thanks. ^_^

it's typically impractical or most of the time you find yourself with values not relevant to your use. noise being almost the only part of fidelity I still care about, I'm always very concerned that some sources will be noisy on my sensitive IEMs.
you can try to crunch numbers all day long, with SNR and max voltage you end up like in your example with a voltage for noise(but here, A weighted... ok), even then it doesn't mean you'll have that once loaded and with a different setting. on some devices lowering the volume level and/gain will also lower the noise, on other gears it won't change a thing. IMO until you get the opportunity to check for yourself, the best information is to find a few noise obsessed people with really sensitive IEMs who tried the gear, and ask them. it's horrible that in sound science I'm suggesting to trust subjective opinions, but in practice that's what worked the best for me. 
of course you must only care for those specific people, because for each noise paranoid dude with sensitive gears and ears, you'll get 50 guys swearing on their mother's pudding that there is no audible noise. so subjective, but stay away from vox populi.

 

[U-3C]

it's typically impractical or most of the time you find yourself with values not relevant to your use. noise being almost the only part of fidelity I still care about, I'm always very concerned that some sources will be noisy on my sensitive IEMs.
you can try to crunch numbers all day long, with SNR and max voltage you end up like in your example with a voltage for noise(but here, A weighted... ok), even then it doesn't mean you'll have that once loaded and with a different setting. on some devices lowering the volume level and/gain will also lower the noise, on other gears it won't change a thing. IMO until you get the opportunity to check for yourself, the best information is to find a few noise obsessed people with really sensitive IEMs who tried the gear, and ask them. it's horrible that in sound science I'm suggesting to trust subjective opinions, but in practice that's what worked the best for me. 
of course you must only care for those specific people, because for each noise paranoid dude with sensitive gears and ears, you'll get 50 guys swearing on their mother's pudding that there is no audible noise. so subjective, but stay away from vox populi.

I will officially provide this service for other people now with all the DACs I own~ ^o^

Not scientific by any means, but I'll record the noise from my sensitive iems so if I ever have more than one DAC, I'll be able to create a very small library of videos that can be compared relative to each other. Unless someone pays me 300k to create a somewhat proper measuring rig and school me for the years on audio engineering theory.

 

[pinnahertz]

It means that with the volume turned all the way up and no input there is a 7uV RMS output. But that measurement is not just off a meter... they've weighted it to make it look better than it is*. You can figure out how loud that actually is (dB SPL) if you know the output impedance of the DAC/amp, sensitivity of your IEMs/headphones, and impedance of your IEMs/headphones.

*A-weighting is dumb. As far as I can tell it's only purpose to hide things the manufacturer is afraid of and to allow them to report higher than measured specs so they can charge more money. The whole thing about how the ear works is a bunch of garbage.

"But that measurement is not just off a meter... they've weighted it to make it look better than it is*.  *A-weighting is dumb. As far as I can tell it's only purpose to hide things the manufacturer is afraid of and to allow them to report higher than measured specs so they can charge more money. The whole thing about how the ear works is a bunch of garbage."
 
Interesting opinion of the internationally standardized A-weighting curve ( IEC 61672:2003). As far as the whole thing about how the ear works being a "bunch of garbage",  well, science disagrees with you rather completely.  The sensitivity of the the human hearing system to sounds at low levels has been thoroughly researched, and well documented.  The A-weighting curve was intended to mimic hearing frequency response at low levels (40 phon), as applied to SPL measurements.  Its use in fully electronic noise measurements does create a bit of ambiguity (what SPL corresponds to either max output or the noise level?), but it can be assumed, especially in a modern digital audio system, that noise floors will be low enough for the A-weighting curve to be relevant.  For electronics noise measurements, A-weighted figures are meaningful, but would be more meaningful if unweighted figures were presented as well.  The curve rolls off low frequency response below 1kHz, de-emphasizing low frequency noise, and also rolls off above 10kHz.  Both sections of the curve make sense in low-level noise evaluation.  Comparing unweighted and A-weighted noise figures to each other give us an indication of the low frequency content.  Since low frequencies are far less audible at low levels, the curve gives us a better idea of how the noise level will sound subjectively.
 
Low frequency noise weighting is used commonly in evaluating noise in acoustic spaces.  ANSI standard Noise Curves (NC - ANSI S12.2-2008) are a family of curves that relate to specific SPL in octave bands.  The NC40 curve lines up fairly well with the low frequency portion of the A-weighting curve (which is a single curve, not a family).
 
@U-C3: ""Noise Floor 7 µV RMS (A-weight), max gain"

Can anyone explain what that means, and as a rule of thumb, how does it stand? Is it very below average, very good, just a marketing term used to confuse people and don't tell anything without more info on the reference used, just a made up pseudoscientific term, etc.?"
 
The formula for this calculation is: dB = 20 * Log (Vout / Vin),  -112dB=20*Log (2.9 / .000007)  The added weighting scales a non-weighted result for more agreement with percieved audibility of low level noise.
 
In the example, the noise is -112dB below maximum output with the gain all the way up.  It's likely that an unweighted measurement would place the noise 10dB higher, based on the typical spectral density of purely electronic noise in critical bands.  Even though 24 bit quantization theoretically has a range of 144dB, you won't find a DAC or ADC doing much better than 112dBA.  So typical, and good at the same time.   The figure is used in marketing, it is slightly confusing and incomplete, but is not a made-up pseudoscientific term.  What's missing is the test conditions.  We might assume it has a digital signal with no signal, not even LSB jitter.  But some DACs go into a "mute" condition when that happens, or when no bit stream is present.  So we don't know, but it's within the realm of typical 24 bit DAC performance.  What will that noise sound like with your high sensitivity IEMs plugged in?  Hard to say without knowing the IEM sensitivity, but you won't run the gain all the way up, and we don't know if the noise floor changes with gain or not.  But assuming it does, which would be expected to a point,  and you adjust that gain so average listening level at say 90dB SPL, that residual noise will be far below the threshold of hearing, and far, far below any residual noise found in any recording.  

 

[GRUMPYOLDGUY]

"But that measurement is not just off a meter... they've weighted it to make it look better than it is*.  *A-weighting is dumb. As far as I can tell it's only purpose to hide things the manufacturer is afraid of and to allow them to report higher than measured specs so they can charge more money. The whole thing about how the ear works is a bunch of garbage."

Interesting opinion of the internationally standardized A-weighting curve ([COLOR=252525] [/COLOR]IEC
[COLOR=252525] 61672:2003)[/COLOR]. As far as the whole thing about how the ear works being a "bunch of garbage",  well, science disagrees with you rather completely.  The sensitivity of the the human hearing system to sounds at low levels has been thoroughly researched, and well documented.  The A-weighting curve was intended to mimic hearing frequency response at low levels (40 phon), as applied to SPL measurements.  Its use in fully electronic noise measurements does create a bit of ambiguity (what SPL corresponds to either max output or the noise level?), but it can be assumed, especially in a modern digital audio system, that noise floors will be low enough for the A-weighting curve to be relevant.  For electronics noise measurements, A-weighted figures are meaningful, but would be more meaningful if unweighted figures were presented as well.  The curve rolls off low frequency response below 1kHz, de-emphasizing low frequency noise, and also rolls off above 10kHz.  Both sections of the curve make sense in low-level noise evaluation.  Comparing unweighted and A-weighted noise figures to each other give us an indication of the low frequency content.  Since low frequencies are far less audible at low levels, the curve gives us a better idea of how the noise level will sound subjectively.

Low frequency noise weighting is used commonly in evaluating noise in acoustic spaces.  ANSI standard Noise Curves (NC - [COLOR=252525]ANSI S12.2-2008[/COLOR]) are a family of curves that relate to specific SPL in octave bands.  The NC40 curve lines up fairly well with the low frequency portion of the A-weighting curve (which is a single curve, not a family).

@U-C3: ""Noise Floor 7 µV RMS (A-weight), max gain"


Can anyone explain what that means, and as a rule of thumb, how does it stand? Is it very below average, very good, just a marketing term used to confuse people and don't tell anything without more info on the reference used, just a made up pseudoscientific term, etc.?"

The formula for this calculation is: [COLOR=575757]dB = 20 * Log (Vout / Vin),  -112dB=20*Log (2.9 / .000007)[/COLOR] [COLOR=575757] The added weighting scales a non-weighted result for more agreement with percieved audibility of low level noise.[/COLOR]

In the example, the noise is -112dB below maximum output with the gain all the way up.  It's likely that an unweighted measurement would place the noise 10dB higher, based on the typical spectral density of purely electronic noise in critical bands.  Even though 24 bit quantization theoretically has a range of 144dB, you won't find a DAC or ADC doing much better than 112dBA.  So typical, and good at the same time.   The figure is used in marketing, it is slightly confusing and incomplete, but is not a made-up pseudoscientific term.  What's missing is the test conditions.  We might assume it has a digital signal with no signal, not even LSB jitter.  But some DACs go into a "mute" condition when that happens, or when no bit stream is present.  So we don't know, but it's within the realm of typical 24 bit DAC performance.  What will that noise sound like with your high sensitivity IEMs plugged in?  Hard to say without knowing the IEM sensitivity, but you won't run the gain all the way up, and we don't know if the noise floor changes with gain or not.  But assuming it does, which would be expected to a point,  and you adjust that gain so average listening level at say 90dB SPL, that residual noise will be far below the threshold of hearing, and far, far below any residual noise found in any recording.  

To be clear, the "it's a bunch of garbage" comment is in reference to using how the ear works as justification for changing the measurements of a device, not a dismissal of the very much understood knowledge of how the ear actually works.

I don't care that it's standardized, A weighting shouldn't even exist. When I look up the measurements of a device, I want the measurements of that device. Period.

If I were caught changing measurements by our customers, the company I work for would be blacklisted from ever doing business with them again. But in audio, it's the standard. Ridiculous.

 

[pinnahertz]

To be clear, the "it's a bunch of garbage" comment is in reference to using how the ear works as justification for changing the measurements of a device, not a dismissal of the very much understood knowledge of how the ear actually works.

I don't care that it's standardized, A weighting shouldn't even exist. When I look up the measurements of a device, I want the measurements of that device. Period.

If I were caught changing measurements by our customers, the company I work for would be blacklisted from ever doing business with them again. But in audio, it's the standard. Ridiculous.

It's not changing measurements in audio, it's making them more meaningful in relation to actual audibility.  It's as if your customers asked you how heavy your product is, and the weight you quoted included the shipping container.  They don't care about that, they're  not going to use the container.   They only want a measurement of what they're actually going to use.  A-weighting removes the inaudible noise, and measures only the noise the customer will actually hear.  Your "shipping weight" (product and shipping materials) figure is accurate too, just not applicable, or of interest to the customer's use.  
 
Unweighted measurements are also accurate, but they don't represent how a noise floor will sound.  I suppose you also object to band limiting a noise measurement to below 20kHz.  Yet, technically, that should also be done even while using an A-weighting filter because we can't hear ultrasonic noise.  
 
I agree, I'd like to see unweighted noise too, but not exclusively.  I need to see both.  Or better, a spectrum of that noise.  But if it has to be a single number, A-weighted is probably the most meaningful.
 
But I do appreciate that you have an opinion that is in direct opposition to they way audio has been measured and evaluated (not just writing specs) for the last 50+ years, and is in place because of some serious research.
 
I can't say it any better than this, a company that deals in measurement devices, not audio gear:
 

"The human ear responds more to frequencies between 500 Hz and 8 kHz and is less sensitive to very low-pitch or high-pitch noises. The 

frequency weightings

 used in sound level meters are often related to the response of the human ear, to ensure that the meter is measuring pretty much what you actually hear.

It is extremely important that sound level measurements are made using the correct frequency weighting - usually A-weighting. For example, measuring a tonal noise of around 31 Hz could result in a 40 dB error if using C-weighting instead of A-weighting."

So if you like the 40dB error, then go unweighted. 

(C-weighting is, of course, not unweighted, it does have a curve, but it's quite flat compared to A-weighting)

 

[GRUMPYOLDGUY]

It's not changing measurements in audio, it's making them more meaningful in relation to actual audibility.  It's as if your customers asked you how heavy your product is, and the weight you quoted included the shipping container.  They don't care about that, they're  not going to use the container.   They only want a measurement of what they're actually going to use.  A-weighting removes the inaudible noise, and measures only the noise the customer will actually hear.  Your "shipping weight" (product and shipping materials) figure is accurate too, just not applicable, or of interest to the customer's use.  

Unweighted measurements are also accurate, but they don't represent how a noise floor will sound.  I suppose you also object to band limiting a noise measurement to below 20kHz.  Yet, technically, that should also be done even while using an A-weighting filter because we can't hear ultrasonic noise.  

I agree, I'd like to see unweighted noise too, but not exclusively.  I need to see both.  Or better, a spectrum of that noise.  But if it has to be a single number, A-weighted is probably the most meaningful.

But I do appreciate that you have an opinion that is in direct opposition to they way audio has been measured and evaluated (not just writing specs) for the last 50+ years, and is in place because of some serious research.

I can't say it any better than this, a company that deals in measurement devices, not audio gear:

"The human ear responds more to frequencies between 500 Hz and 8 kHz and is less sensitive to very low-pitch or high-pitch noises. The frequency weightings used in sound level meters are often related to the response of the human ear, to ensure that the meter is measuring pretty much what you actually hear.

It is extremely important that sound level measurements are made using the correct frequency weighting - usually A-weighting. For example, measuring a tonal noise of around 31 Hz could result in a 40 dB error if using C-weighting instead of A-weighting."

So if you like the 40dB error, then go unweighted. 

(C-weighting is, of course, not unweighted, it does have a curve, but it's quite flat compared to A-weighting)

If they're getting 40dB error, then they need new test equipment.

The audibility of something is irrelevant when characterizing the hardware. The hardware is the hardware, period.

And I doubt if I comb through these forums I will find a single post that says "Oh boy, 10dB noise figure? Man I'm glad they used an A weighted measurement, that's so much more relatable to me now!! Praise the sun god!"

 

[headdict]

The audibility of something is irrelevant when characterizing the hardware. The hardware is the hardware, period.

 
What's the use of (audio related) hardware characteristics if they are (to a large extent) irrelevant to audibility?

 

[GRUMPYOLDGUY]

What's the use of (audio related) hardware characteristics if they are (to a large extent) irrelevant to audibility?

How is a measurement taken straight from the test equipment irrelevant?

I think I agree with pinnahertz, I would be okay with it if unweighted measurements were also provided. But I still argue that a weighted measurement is just an opportunity for manufacturers to hide poor performance.

 

[bfreedma]

Quote:Originally Posted by GRUMPYOLDGUY The audibility of something is irrelevant when characterizing the hardware. The hardware is the hardware, period. What's the use of (audio related) hardware characteristics if they are (to a large extent) irrelevant to audibility?

Marketing?.....

I think the importance of electronics that don't have audible issues and don't attempt to address Nyquist is because audibly flat response is relatively easily achieved there. Transducers are more variable, so trying to fix it in the hardware vs. via EQ has, IMO, two significant issues:

1. The EQ needed to address various headphones, recordings and individual hearing isn't static, so a hardware based solution isn't very flexible

2. It's going to get messy when you have hardware in the playback chain, each with it's own manufacturers implementation of a Nyquist (or other curve) based solution

 

[headdict]

How is a measurement taken straight from the test equipment irrelevant?

I think I agree with pinnahertz, I would be okay with it if unweighted measurements were also provided. But I still argue that a weighted measurement is just an opportunity for manufacturers to hide poor performance.

I think we have very different opinions on what contributes to the meaningfulness or relevance of a measurement. Maybe it's because I'm not an engineer and thus not interested in any technical details that have no effect on my enjoyment of music.

 

[GRUMPYOLDGUY]

I think we have very different opinions on what contributes to the meaningfulness or relevance of a measurement. Maybe it's because I'm not an engineer and thus not interested in any technical details that have no effect on my enjoyment of music.

So the same measurement with weighting means something more to you? Why? See my example above... what does a 10dB noise figure mean to you? Now what does it mean to you if I tell you it's weighted?

 

[headdict]

So the same measurement with weighting means something more to you? Why? See my example above... what does a 10dB noise figure mean to you? Now what does it mean to you if I tell you it's weighted?

I suppose the weighting takes characteristics of human hearing into account. That's why it is more meaningful to me than the corresponding unweighted number, as I expect it to be more closely related to my music perception. The unweighted number might be interesting in some other respect, but why should I care? If it tells me something about a shortcoming of the hardware that is however not reflected in the weighted number, why is it relevant to me as a music listener?

 

[GRUMPYOLDGUY]

I suppose the weighting takes characteristics of human hearing into account. That's why it is more meaningful to me than the corresponding unweighted number, as I expect it to be more closely related to my music perception. The unweighted number might be interesting in some other respect, but why should I care? If it tells me something about a shortcoming of the hardware that is however not reflected in the weighted number, why is it relevant to me as a music listener?

Sorry, would you mind answering the question first before countering it with your own?

But to answer your new question... I'm almost positive the number doesn't matter to you either way.

 

[pinnahertz]

If they're getting 40dB error, then they need new test equipment.

The audibility of something is irrelevant when characterizing the hardware. The hardware is the hardware, period.

And I doubt if I comb through these forums I will find a single post that says "Oh boy, 10dB noise figure? Man I'm glad they used an A weighted measurement, that's so much more relatable to me now!! Praise the sun god!"

1. If they're getting 40dB error, then they need new test equipment.
 
2, The audibility of something is irrelevant when characterizing the hardware. The hardware is the hardware, period.
 
3. And I doubt if I comb through these forums I will find a single post that says "Oh boy, 10dB noise figure? Man I'm glad they used an A weighted measurement, that's so much more relatable to me now!! Praise the sun god!"
 
1. If your profession was audio engineering you'd know that every single audio performance measurement system from the multi-kilo$ AP right down to the free REW includes several standardized weighting filters.  Every SPL meter worth anything over $10 has weighting filters.  No new equipment needed.  It's a question of relevance.  And correctly applied weighting in noise measurements is very, very relevant. 
 
2. Not if it's audio hardware.  Sorry, that's completely naive.  In audio equipment the measurement "holy grail" is audible correlation of measured parameters.  In fact, it's about all that matters.
 
3. You won't.  Because there's no such thing as a 10dB noise figure.  All noise figures must reference something like full output, a specific sound pressure, even thermal noise in a stated resistor, there MUST be a reference, and the noise figure is a ratio.  
 
What you wouldn't understand unless you've done it is, you can measure two devices unweighted, they measure very close to, or exactly the same noise level, but one sounds much quieter than the other.  So, what do you do?  Publish unweighted and misleading noise specs?  That's what weighted measurements attempt to reconcile.  To understand why one device (or acoustic space) sounds noisier than another you must...absolutely MUST...take hearing response into account.  Every single audio engineer knows this, and applies it.  You can take hearing response into account with fixed weighting filters, or plotted curves and spectrum analysis, but you have to do it or you aren't engineering anything.  

 

[headdict]

Sorry, would you mind answering the question first before countering it with your own?

But to answer your new question... I'm almost positive the number doesn't matter to you either way.

The exact figure means nothing to me and I wasn't arguing about that. I'm just trying to understand whether (and why) I need to care about an unweighted figure, given that I'm not interested in the characteristics of the hardware itself, but just in the results I can get out of it.
 
Sorry if I'm not expressing myself clearly most of the time. Not only am I not an engineer, but English is not my first language either, as you probably have noticed.