Audibility thresholds of amp and DAC measurements

[bennetng]

I am not going to comment about audibility threshold as such discussions often lead to endless debates.

One point I want to address is 16/44 audio with various dithering and noise shaping technique can achieve 10 to 20dB better dynamic range than 96dB at some frequencies. On the other hand most modern DACs usually have a flat noise floor at least up to 30kHz or so before modulator noise showing up.

I created some test files for evaluation. Blumlein 88 also provided other test files later in the thread below.
https://www.audiosciencereview.com/forum/index.php?threads/philips-red-book.2138/post-57936

 

[Krunok]

A side note:

When attempting to measure speaker distortion here, the noise floor seems to dominate the reading, until the SPL set exceeds the levels at which I would listen.

SPL, from 50 to 91 at 1kHz. Ambient "noise" intrudes here at the bottom left...

View attachment 19015


And the THD for the same measurements. Only at the highest SPL (traces at the bottom) does the measured distortion seem to start rising out of the noise. The top six traces, I interpret as ambient noise being mistaken for harmonic distortion.

View attachment 19016

Anyway, makes me wonder about the tiny amounts produced by various DACs.

Which would be waaaaay down here someplace:

View attachment 19018

Maybe the difference in THD between 0.5% and 0.5005% can also be heard if you get to compare the two immediately one right after the other.

 

[j_j]

How about "THD is a useless measure".

What one must do is compare error spectrum to signal spectrum. This is not something easily explained at 2AM on a Tuesday morning.

 

[sergeauckland]

How about "THD is a useless measure".

What one must do is compare error spectrum to signal spectrum. This is not something easily explained at 2AM on a Tuesday morning.

I don't think it is a useless measure. It's limited, certainly, but it is a very good measure of the non-linearity of transfer function. It may or may not correlate with audibility, and as such can be misused in setting specs, but as an honest metric, together with frequency response and noise, it can characterise an amplifier pretty well.

If I know that an amp is flat to 1dB within 20-20kHz, has noise better than -90dB below full output and distortion less than 0.1% at all frequencies and levels above, say, 100mW - even then, it'll be noise-limited, there won't be a lot wrong with it. Add a 'scope trace of the distortion residual that shows no crossover distortion, and there's not much else I need to know.

S.

 

[Kyle / MrHeeHo]

Isn't the SMSL 10th and example from before a separate USB power supply was plugged in?

 

[j_j]

I don't think it is a useless measure. It's limited, certainly, but it is a very good measure of the non-linearity of transfer function. It may or may not correlate with audibility, and as such can be misused in setting specs, but as an honest metric, together with frequency response and noise, it can characterise an amplifier pretty well.

If I know that an amp is flat to 1dB within 20-20kHz, has noise better than -90dB below full output and distortion less than 0.1% at all frequencies and levels above, say, 100mW - even then, it'll be noise-limited, there won't be a lot wrong with it. Add a 'scope trace of the distortion residual that shows no crossover distortion, and there's not much else I need to know.

S.

If you insist. If it's over 120dB below one watt, I'm happy, ok?

 

[Francis Vaughan]

A caveat when talking about CD resolution. Whilst 16 bits gets you your 96dB, it doesn't take into account the use of noise shaped dither. It is pretty universal that when a CD is mastered, it is dithered with one of a family of shaped dither functions, and that this shaping of the dither trades off resolution in the higher frequencies (say above 10kHz) against higher real resolution in the mid frequencies, with a common result of trivially another 6dB, and often 12dB of resolution in the bands where the ear has highest resolution. Thus it isn't true to say that CDs have a flat 16 bits of resolution. One could employ a useful limit that is "exceeds CD resolution" but it would need to take into account the use of shaped dither.

One sees the occasional claim that dither is some form of "trickery" and the dynamic range not valid - which is sadly ignorant. The critical point again is that you are trading off dynamic range across the frequency spectrum, and reducing it in bands where the ear is insensitive (raising the noise floor) and gaining it in bands where it is more sensitive. The figures of merit for measured performance of equipment needs to take this into account.

ETA - I see I was beaten to the punch by bennetng above.

 

[sergeauckland]

If you insist. If it's over 120dB below one watt, I'm happy, ok?

I too would be happy with -120dB, but then I'm easily pleased as I'm equally happy with -60dB.

(at all frequencies and levels)

S.

 

[flipflop]

One point I want to address is 16/44 audio with various dithering and noise shaping technique can achieve 10 to 20dB better dynamic range than 96dB at some frequencies. On the other hand most modern DACs usually have a flat noise floor at least up to 30kHz or so before modulator noise showing up.

Whilst 16 bits gets you your 96dB, it doesn't take into account the use of noise shaped dither.

One could employ a useful limit that is "exceeds CD resolution" but it would need to take into account the use of shaped dither.

You're both right. I have rephrased the relevant text in the opening post in order to address dithering. The threshold is based solely on dynamic range, though, so it remains unchanged.

It is pretty universal that when a CD is mastered, it is dithered with one of a family of shaped dither functions

If you could provide a source for this, it would be much appreciated. According to @amirm "No one knows what percentage of distributed music uses noise shaping."

Isn't the SMSL 10th and example from before a separate USB power supply was plugged in?

Yes, but the manual doesn't specify that it shouldn't be used with a USB 2.0 port without an external power supply. Therefore it's fair to assume some people might and that's why I've omitted to comment on it.
Bonus info: Before the Sanskrit 10th review came out, I intended to use Teac HA-P50 as my example of a DAC with "poor" jitter performance. I think Sanskrit 10th's jitter spectrum is a better choice because more of the spikes are symmetrical, which is a key feature of jitter.

 

[andreasmaaan]

If you could provide a source for this, it would be much appreciated. According to @amirm "No one knows what percentage of distributed music uses noise shaping."

I’m not aware of any studies or surveys that have been done to give statistical information on this. However, if you look at the instruction manual of any widely used DAW (I just glanced at the manuals for Ableton Live, Cubase and Protools), there is a section in which the basics of dither are explained along with recommendations for its use. In each case, a noise-shaping form of dither is recommended for final mixdown.

It’s also my experience that among recording and mastering engineers (and in fact also most amateurs) the general principles are widely understood and followed.

Obviously this is not precisely the level of evidence you’re looking for, but it’s enough to satisfy me personally that dither is applied correctly in the vast majority (if not all) of cases when it comes to professionally recorded and mastered music.

I have a few friends who are professional mastering engineers and I would be happy to ask them for their perspective if people require further confirmation.

 

[Jimster480]

These are an awesome set of guidelines!
Really putting the measurements into perspective for "the masses" aswell as many of the members here.
I love that you included research done by others as well in the past.

While this isn't scientifically proofed, it is pretty damn explanatory and makes alot of sense!

 

[Guermantes]

If you could provide a source for this, it would be much appreciated. According to @amirm "No one knows what percentage of distributed music uses noise shaping."

I have quite a few Sony CDs with the SBM symbol on them. Most of the mastering limiters I use have dither functions. This one by Waves opens with dither enabled by default, so even if you didn't know what you were doing, you would probably still be applying it:

Back in 1987 it was being encouraged by AES luminaries: http://www.aes.org/e-lib/browse.cfm?elib=5173
I suspect for mastering engineers it would go into the "best practice" box.

 

[j_j]

I too would be happy with -120dB, but then I'm easily pleased as I'm equally happy with -60dB.

(at all frequencies and levels)

S.

60dB below one watt at the speaker input puts things at about 36dB SPL for me.

Not low enough.

 

[Snafu]

Not sure if it's been answered, measurements are for one single unit and it's qualities. How these numbers add up when you put these units in audio chain, for example DAC+pre+poweramps ? If you have three units that each measure just inaudible, is total still inaudible ?

 

[sergeauckland]

Not sure if it's been answered, measurements are for one single unit and it's qualities. How these numbers add up when you put these units in audio chain, for example DAC+pre+poweramps ? If you have three units that each measure just inaudible, is total still inaudible ?

Probably not. Noise adds, as do frequency response errors. Distortion may or may not add directly, given that the harmonic structure may be different, but each harmonic contribution to the total adds. However, there may be filtering and attenuation as well as gain, so the additional contribution from each item isn't directly additive. Nevertheless, it's a good idea for each item to be comfortably within transparency criteria so the overall result can still be.

S

 

[RayDunzl]

If you have three units that each measure just inaudible, is total still inaudible ?

http://www.sengpielaudio.com/calculator-leveladding.htm

for noise (incoherent signal), if the values are much different, pick the worst one as a good estimate.

If the values are not much different, they might add a little.

If adding tones - of the same frequency - (coherent), and not noise, use this calculator:

http://www.sengpielaudio.com/calculator-coherentsources.htm

The above doesn't take into account any amplification/attenuation that might occur in your chain...

 

[trl]

[...]

View attachment 18966
Audio-gd NFB-28.28 has a "Massive 3rd Harmonic Distortion" spike that puts it above all 3 thresholds.

[...]

View attachment 19064
JDS Labs The Element stays below the lenient thresholds, but doesn't stay below the strict threshold.
[...]

Congrats for the article, impressive work indeed!

Based on https://en.wikipedia.org/wiki/Amplifier_figures_of_merit#Output_dynamic_range the difference between the highest non-distorted sound and the background noise (averaged perhaps) gives us the Dynamic Range. So, I'd say that the above 2 equipments aren't breaching the 120 dB DR threshold, right? As for the linearity, raising the bar from 96dB to 120dB of dynamic isn't kinda huge difference and not perceived by our ears.

At -66dB you'll not even hear the 5th harmonic, so it's quite a decent and safe number, at least in regard with the first 2 harmonics.

At -85dB of noise below the standard 2V RMS, my 100dB 16 Ohms IEMs will hear everything (noise, hum, spikes etc.). For such sensitive IEMs I need a noise level below -115dB (source: http://www.sengpielaudio.com/calculator-db-volt.htm).

Also, two very interesting articles might be linked to this thread too: https://www.audiosciencereview.com/forum/index.php?threads/can-we-trust-our-ears.3884/ and https://www.audiosciencereview.com/...hat-level-is-noise-heard-in-your-system.1013/.

L.E.: My 100dB sensitivity IEM's need a background noise of -85dB lower than 0.125V RMS (power for 1mW to drive them to 100dB). So, very good IEMs to test background noise with my own ears.

 

[flipflop]

Based on https://en.wikipedia.org/wiki/Amplifier_figures_of_merit#Output_dynamic_range the difference between the highest non-distorted sound and the background noise (averaged perhaps) gives us the Dynamic Range. So, I'd say that the above 2 equipments aren't breaching the 120 dB DR threshold, right?

Yes and no. You're onto something.
The lenient thresholds are split up into THD (-66 dB) and noise (-85 dB). I guess you could say the strict ones are, too, but they're both at -120 dB. So while the two graphs in question don't breach the strict noise threshold, they do break the strict THD threshold.

 

[Blumlein 88]

Looks like speaker distortion is probably above .1 % almost always. Headphones may be lower. If your total distortion stays low enough noise and FR should be your only worries.

 

[trl]

Yes and no. You're onto something.
The lenient thresholds are split up into THD (-66 dB) and noise (-85 dB). I guess you could say the strict ones are, too, but they're both at -120 dB. So while the two graphs in question don't breach the strict noise threshold, they do break the strict THD threshold.

I'm speaking from my hearing experience here and I wasn't able to pick up the 2nd nor the 3rd harmonic out of the fundamental on a -60dB difference: https://www.audiosciencereview.com/...can-we-trust-our-ears.3884/page-4#post-136991 and here: https://www.audiosciencereview.com/forum/index.php?threads/can-we-trust-our-ears.3884/post-125971.

I don't get the -120dB, sorry, so I don't get the point. I know that AP has around -140dB of noise and -120dB of noise out of a Hi-Fi system will most likely not be audible. Instead -85 dB of noise is a crappy noisy DAC/amplifier, but like I said...not sure I got the point, so I'll re-read your post carefully tomorrow morning. Or you are probably referring to dynamic as the difference between the loudest and the barely perceivable sound, like here: https://www.audiosciencereview.com/...-noise-heard-in-your-system.1013/post-127920? If so, then I can hear up to about -84dB @1Khz and few dB lower @ 440Hz, so a 96dB dynamic range of the Audio CD seems more than reasonable to me. Thanks for understanding!