• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Audibility thresholds of amp and DAC measurements

Hearing a signal below a noise floor is different thing. Apples and oranges.

Now I am very confused and I think you are too.

When someone talks about SINAD thresholds, they are talking about noise and harmonics. Harmonics are non-random spurious signal related to the signal and noise is typically more random and unrelated to the signal (but AC mains noise is not random…)

If you take your music signal, play it back, and subtract your music signal, what’s left is the unintended signal.

That’s where the signal below noise floor calculations are directly correlated.

That is, if your music is playing at 130 dB and your SNR is 130 dB, then that chart shows you what you might hear.


GXAlan, I wasn't trying to declare that I have a better number or anything like that. I was just hoping someone could point me to some published science that support statements about strict limits in the first post of the thread.

At this point it still seems a lot like an opinion.

There is opinion, and then educated conclusion/forecast based upon available data.
So it is an opinion, but unlike asking people for their favorite food it’s more like asking anesthesiologists for the safest inhaled anesthetic. (The mechanism of general anesthesia is not fully understood — it is just safely reproduced).

It goes more than opinion, but I think if you are trying to say that this number is too relaxed, then you will find more support. If you think this number isn’t tight enough, you probably will find support going up to 130 or even 144 dB.

Some of the 120 dB range came from the estimate that you gave 24-bit data and the best technology still giving up 4 bits of noise.



On that, I have some concerns. Statistics seems to have gotten more complicated in recent years: https://www.edge.org/response-detail/11715 No longer do bell curves have to be normal.

See this is where we have some boundary conditions which don’t exist in the author’s examples. There is a physical limit to the how loud you can achieve and this gets to my point about what your actual question/goal is.

Take for example, your skin. Will a military bullet fired from a pierce it at 10 ft? Yes. That’s because your skin is made with cells which are made from specific arrangements of proteins, lipids, etc.

If you look at something like your eardrum bursting, you can see that while there is conflict over the “50% threshold” depending on which study you look at, the 100% threshold was fairly consistent.

30 psi is 200 dB.

THEN, we have our final question.

Suppose you (or someone else) has the magical X-men style mutation that allows him/her to hear to -200 dB. We can then use blind testing or ABX testing to show that the person can distinguish between the two…

In a way, because the best electronics are still only at -120 dB it’s not easy to design that study. That said, even now, it will be hard to distinguish two DACs that are level and frequency response matched.

If you want to argue for arguments sake, you can have an infinite set of what if’s. Quantum physics says you can drive a car through a concrete wall and pass through with a non-zero probability. That said, possible and probable are two different things. It is possible that -120 dB is not accurate but it is probably that it’s pretty close, if not generous.

If you want to argue that two amplifiers sound different, then that’s different. 5W 1 kHz SINAD is a good predictor of overall performance but real world speaker loads can result in frequency response differences that are audible…

So yes, -120 dB isn’t precise. But it’s not pulled out of thin air.
 
In the first post of this thread the term "strict" is used. Where does that come from?
That's a different usage. The poster means "strict" in the sense of "can never be audible, ever". For the wrong signal in the wrong equipment that can be quite high, but in order to even be able to calculate such a number, we must talk, frankly, about a particular set of hardware in a particular listening environment.
"Strict threshold" the way the cite used it is not actually a 'thing' in the same sense as "absolute threshold" and the like.

You're struggling because the term isn't actually defined that well. It's not you.

I will say with consumer equipment, the 120dB number is probably pretty good.
 
You're struggling because the term isn't actually defined that well. It's not you.
That’s actually a great explanation.

I agree wholeheartedly with this comment.
 
Suppose you (or someone else) has the magical X-men style mutation that allows him/her to hear to -200 dB. We can then use blind testing or ABX testing to show that the person can distinguish between the two…

Yeah, that's a problem. The white noise level of the atmosphere, AT THE EARDRUM, is approximately +6 to +8.5 dB SPL.

You can't hear that, because your ear responds in a frequency-filtered domain, and at the most sensitive point of the hearing apparatus, that level is about 5 dB below the known thresholds of hearing for an unimpaired human.

If one could hear to -200dB, all they would hear was the hiss due to the molecular nature of air at that level.
 
Hmmm. Are you familiar with stochastic resonance? How about Suprathreshold stochastic resonance?

They are scientific facts in a general way, although one can argue about how they might apply to humans.

Are you aware of the noise level of the atmosphere due to its molecular nature on an area the size of the eardrum? It provides masking for anything below that level, and that level us surprisingly close to the ear's actual performance.

Stochastic resonances can not choose what they resonate, and will expose the noise of the atmosphere just as well as the sound of anything else. This sudden diversion makes me quite suspicious of where you seem to be going. This route has been traveled before.
 
Now I am very confused...
Noise can be at any SPL, unless someone defines a specific SPL. For some noise levels a human may be able to hear a signal to some degree under the noise floor.

Okay, that's for noise.

For single tones there is a minimal SPL to be able to hear a tone, for the average ear. The definition of an absolute threshold requires that no other tones be present. If there are other tones or maybe even if there is suitable noise, then the a tone below the threshold may be raised above the threshold some of time.

I believe that's what the scientific research says, if that make sense.

Then there is the question of trying to apply the above science to measurements of SINAD.

Okay. I will stop here for the moment. I think the above describes my viewpoint here. Mostly I am interested in the underlying science. That's what I was trying to ask about. The reason for summarizing is to try to explain why I don't feel confused regarding the above points.
 
In which case it's wrong, because around ear canal resonances, you can hear (if you're unimpaired) below 0dB SPL.

Agreed. But it is helpful to know that it’s not infinitely negative.

The reference level is 20 microPascals, which is about the threshold of human hearing. 6 dB is double the sound pressure.

So -6 dB means 10 micropascals. -12dB is 5 micropascals. -18 is 2.5 micropascals. And -24 is 1.25 micropascals.

So clearly 144 dB is more than enough fudge factor for dynamic range.

Yeah, that's a problem. The white noise level of the atmosphere, AT THE EARDRUM, is approximately +6 to +8.5 dB SPL.

You can't hear that, because your ear responds in a frequency-filtered domain, and at the most sensitive point of the hearing apparatus, that level is about 5 dB below the known thresholds of hearing for an unimpaired human.

If one could hear to -200dB, all they would hear was the hiss due to the molecular nature of air at that level.

+1000. I picked 200 dB as a convenient unrealistic example, but that is an additional useful data point.

For the record, I am perfectly happy listening to vinyl LPs with a SINAD in the 50’s at best.
 
Thread's subject refers to term "audibility", meaning that if we're using decent speakers running at about 85-90dB SPL then we should expect a THD+N of about 1% on the low-end and down to 0.1% on the rest of the audibile frequency band. With headphones these figures will improve a bit, but in the end it won't matter much if the DAC + AMP will have a SINAD of 120dB (THD+N of 0.0001%) or 100dB (THD+N of 0.001%), because the limiting factor is related to the speakers / headsets used.

@Wanman, there are several websites were everyone can test it's own hearing limits, as well as THD limits with different sinewaves or plain songs. Not sure this helps to partially answer to your question from two pages below, but hearing thresholds differ from one person to another and only SPL limits are properly documented because this is needed in working environment (airports, manufacturing processes etc.).
 
...or plain songs.
Thank you. However, songs are not usually single tones. How can absolute limits of hearing apply to a signal which violates the definition of the limit of hearing?

To say it another way, I don't think superposition can apply here since the system in question has a threshold which makes the system highly nonlinear. So how is anyone getting from single tones to music?
 
Thank you. However, songs are not usually single tones. How can absolute limits of hearing apply to a signal which violates the definition of the limit of hearing?

To say it another way, I don't think superposition can apply here since the system in question has a threshold which makes the system highly nonlinear. So how is anyone getting from single tones to music?
Why don’t you make your point and stop dancing around while asking questions you already know the answer (or better answer) to?
 
In which case it's wrong, because around ear canal resonances, you can hear (if you're unimpaired) below 0dB SPL.

where did I say the lower bound was at 0db SPL?

( picture copied :) )

1705566671115.jpeg
 
Noise can be at any SPL, unless someone defines a specific SPL. For some noise levels a human may be able to hear a signal to some degree under the noise floor.

Okay, that's for noise.

For single tones there is a minimal SPL to be able to hear a tone, for the average ear. The definition of an absolute threshold requires that no other tones be present. If there are other tones or maybe even if there is suitable noise, then the a tone below the threshold may be raised above the threshold some of time.

I believe that's what the scientific research says, if that make sense.

Please go here: https://www.aes-media.org/sections/pnw/pnwrecaps/2019/apr2019/

This will help clear up your massive conclusion between SPL, tones, and noise.

White noise at 6dB SPL is an exact specification, and it is unquestionably the noise floor, due to the air in the normal atmosphere, present at your eardrum. This is not speculation. It is basic physics. The fact that you simply waffled and evaded that issue in a post above does not promise a good understanding on your part.
Thank you. However, songs are not usually single tones. How can absolute limits of hearing apply to a signal which violates the definition of the limit of hearing?

To say it another way, I don't think superposition can apply here since the system in question has a threshold which makes the system highly nonlinear. So how is anyone getting from single tones to music

Superposition is not an issue here. What matters is the total energy in each 1/3 ERB of the spectrum. Noise, tones, whatever. This has nothing to do with superposition.

We are getting from "single tones" to "music" based on well-known understanding of the human cochlea and how it filters signals with a rather remarkable, very wet mechanical filter, and THEN detects signals inside of each 1/3 ERB and to some extent each ERB. Until you understand the basics of this frequency analysis instrument that you have two of in your head, you're just going to be confused. It's that simple.

And, yes an absolute threshold set by a tone and an absolute threshold set by a narrowband noise source are very similar. Everything else in that frequency space is somewhere between pure noise and pure tone.

So, no, you can't hear miraculously through something because of other signal, unless it also excites the same small part of the cochlea, in which case you simply calculate the noise present inside that cochlear place.
 
where did I say the lower bound was at 0db SPL?

( picture copied :) )

View attachment 342777

You referred to someone else's 120dB derivation, and if they are doing threshold of pain to 0dB SPL, they are wrong. Not you, whoever said that.

If one looks at the noise floor of the atmosphere, and adds 96dB, one gets about 101dB SPL, again, without looking at spectrum, which of course matters greatly. If you measure a very very good listening room, you've got something remarkable if you have a room quieter than about 25dBA spl. (A-weight is appropriate near absolute threshold, and nowhere else) you have a very, very quiet room.

Now add the 96dB redbook CD range, and you've got 25+96=121dB spl peaks. There aren't many speakers that can do that. Note, yes, I am saying that almost all listening rooms make the noise level of the atmosphere irrelevant.

On the other hand, a full examination of signal spectrum and noise spectrum gives a much more complicated result, which applies ONLY to the signal and listening room under test. Still, a situation where more than 16 bits, properly dithered, fails to be enough for playback (note, PLAYBACK!) is very, very unusual in the real world.
 
You referred to someone else's 120dB derivation, and if they are doing threshold of pain to 0dB SPL, they are wrong. Not you, whoever said that.

If one looks at the noise floor of the atmosphere, and adds 96dB, one gets about 101dB SPL, again, without looking at spectrum, which of course matters greatly. If you measure a very very good listening room, you've got something remarkable if you have a room quieter than about 25dBA spl. (A-weight is appropriate near absolute threshold, and nowhere else) you have a very, very quiet room.

Now add the 96dB redbook CD range, and you've got 25+96=121dB spl peaks. There aren't many speakers that can do that. Note, yes, I am saying that almost all listening rooms make the noise level of the atmosphere irrelevant.

On the other hand, a full examination of signal spectrum and noise spectrum gives a much more complicated result, which applies ONLY to the signal and listening room under test. Still, a situation where more than 16 bits, properly dithered, fails to be enough for playback (note, PLAYBACK!) is very, very unusual in the real world.
So,in practice,I now have about 38db (A) SPL noise floor in my room.
Where would I go from here and what gear performance would I need to get a good result? (both SPL,IMD and THD+N across,level,frequencies,etc)

Is that the reason that mains studio monitors advertise 130db (some more) SPL?
 
Last edited:
So,in practice,I now have about 38db (A) SPL noise floor in my room.
Where would I go from here and what gear performance would I need to get a good result? (both SPL,IMD and THD+N across,level,frequencies,etc)

Is that the reason that mains studio monitors advertise 130db (some more) SPL?
There is a painful trend in studios to listen at ear-destroying levels. In most recordings, there is not a lot of really low-level signal. But yes, it's a hard problem, and studio monitors are ALSO often directional, and you need to understand how a given radiation pattern relates to your setup.
 
Still, a situation where more than 16 bits, properly dithered, fails to be enough for playback (note, PLAYBACK!) is very, very unusual in the real world.

+1000. This is where I think “strict” transparency really lies where strict is used in contrast to the “lenient” description as opposed to “absolute” definition.

Is that the reason that mains studio monitors advertise 130db (some more) SPL?

The way I look at it is that it’s a bit of a marketing.


The first bullet claims 122.4 dB SPL for the bass. But as you go into the specifications:

Sine wave output with a THD < 0.5 % at 1 m in half space: 95 dB > 100 Hz

Max. long term SPL with pink noise at 2.3 m, in typical listening conditions (single): 90 dB

Max. short term SPL with music material at 2.3 m in typical listening conditions (pair / full range): 103 dB

Looking at Amir’s data

It was no longer flat at 106 dB at 1m. At 2m that means it wouldn’t be flat with the 100 dB peaks and at 3m, it would not be flat with 94 dB peaks…

So all of a sudden, you have gone from “it should have more power than anyone needs” to “hmmm. At 3 ft listening distances, in transients, it might not be fully accurate.
 
The way I look at it is that it’s a bit of a marketing.


The first bullet claims 122.4 dB SPL for the bass. But as you go into the specifications:

Sine wave output with a THD < 0.5 % at 1 m in half space: 95 dB > 100 Hz

Max. long term SPL with pink noise at 2.3 m, in typical listening conditions (single): 90 dB

Max. short term SPL with music material at 2.3 m in typical listening conditions (pair / full range): 103 dB

Looking at Amir’s data

It was no longer flat at 106 dB at 1m. At 2m that means it wouldn’t be flat with the 100 dB peaks and at 3m, it would not be flat with 94 dB peaks…

So all of a sudden, you have gone from “it should have more power than anyone needs” to “hmmm. At 3 ft listening distances, in transients, it might not be fully accurate.
No,I'm not talking about these,I'm talking about mains monitors,the ones we usually wall mount:
(example)

bit same with big Genelecs like 8381A or the ones who soffit mount like 1236A,etc.
 
Discussing that -120 dB is not enough. In an atmosphere that is not your living room.
That can only happen on ASR. Everything you say can and will be used against you. :p

*sitting here absolutely unaware of any distortion or noise from my cheap PA amp and miniDSP.
 
OK, so in the first post, it means guaranteed audible transparency, for example Amir's use of -120 dB SINAD. That is a very strict (demanding) threshold, because much lower levels of SINAD are most likely going to be undetectable too.

IIRC Amir got his number by summing a string of assumptions, plus safety margins. It wasn't by referring to a published reference on SINAD audibility. Any such research is more likely to come up with -60 dB to -90 dB. Hence -120 dB is (exceedingly) strict.

cheers
And added to that is the requirement for engaging special listening techniques to get anywhere near Amir's most demanding threshold, including gainriding the tails of reverberation and listening to floor-level sounds at high volume. That's how Amir can distinguish between recordings with low levels of distortion and noise.

When I have noise at -90 dB, I have to put my ear right up to the tweeter to hear it. That's the equivalent of listening at high volume--if my ear is there when full-scale signal appears, it's gonna hurt.

But, yes, the 120 dB SINAD threshold exceeds any that is found in the literature, and least that has been cited where I've seen it. Published thresholds are much lower, even for listeners able to adjust the volume during the test. For controlled-level tests, it's lower still. A search on "thresholds" in the title of threads yield a page and a half, and many of these threads include citations. But here is the best attempt at a summary:


EDIT OF SHAME: Well, hell. It just occurred to me that I linked the very thread in which I was responding. Maybe me and the person Newman and I were responding to should go back to the head of the thread, recognizing that it was posted back when my high-frequency hearing was better by about 2 kilohertz.

Rick "who, with music and without gain-riding, can't hear harmonic distortion below 1% in blind testing" Denney
 
Last edited:
Back
Top Bottom