• 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

OP
F

flipflop

Addicted to Fun and Learning
Joined
Feb 22, 2018
Messages
927
Likes
1,240
It's not covered in the post, but Multitone measurement thresholds would be the same as for other FFT measurements?
Yes, the thresholds also apply to the multitone test. It's not included in the OP because when I wrote it, the 32-tone test wasn't regularly posted in reviews.
Also does the amount of "grass" compared to clean multitones actually matter as long as it's below the audible threshold?
No, the grass is just noise and distortion. There's nothing fancy going on.
 

Ron Texas

Master Contributor
Joined
Jun 10, 2018
Messages
6,076
Likes
8,911
@flipflop Thanks for getting back to me. Wrt HA, I would not expect you to go through the entire site, but a quick search through the Listening Tests section should yield some good results: https://hydrogenaud.io/index.php/board,40.0.html

Masking is a funny, but real thing... The link to Ethan's test files should give you a good idea of level of masking if you check them out.

As mentioned, if you have JRiver Media Center, you can easily and quickly test your audibility threshold with the bit depth simulator:
https://yabb.jriver.com/interact/index.php?topic=74999.0
Matt, who is one of the principals writing the software did better than I did at 13 bits, but only if he turned the volume up (cheating :) but, as folks will find out. somewhere between 12 bits and 13 bits, under "normal" or "reference" level listening conditions I mentioned in my previous posts, is typically the limit. Once an (any) artefact is below -80 dBFS, for most folks it is simply inaudible... regardless of the type of artefact...

One more point to ponder for folks performing audibility testing in a home listening environment using speakers, which also has a noise masker or noise floor. For my room it measures about 45 dB SPL broadband noise with a rise at the low frequencies. If my critical listening level is 83 dB SPL, at what level can I hear below the noise floor of my room before the noise masks the signal?

IMO, this post is extremely important. It's sort of a mantra for me. My system is similar to the author's second system. LS50's, Rythmic L12's Crown XLS 1502. The difference is he uses Audiolens manage his subs and I use the hardware crossovers built into the 1502 and L12's. Remember, he was a recording engineer for a good while, has written a book on crossovers and does audio system calibration for other audiophiles.
 
N

nhatlam96

Guest
I would like to set 16bit 96dB as threshold, but I am confused with FFT 22kHz and 90kHz.

When using FFT 22kHz, I simply draw a horizontal line at 96dB, like this:
96dB - SMSL SU-9 Balanced USB DAC Audio Measurements.png


However, I don't know if the same rule applies for FFT 90kHz, because it's way different to 22kHz.
Should I also just draw a line at 96dB and it's correct? What would be the explanation for this?
96dB - SMSL SU-9 Balanced USB DAC THD+N vs Frequency Audio Measurements.png
 

pma

Major Contributor
Joined
Feb 23, 2019
Messages
4,591
Likes
10,727
Location
Prague
Higher measurement bandwidth (here 90kHz) means higher measured noise, because noise is measured over higher frequency range and higher frequencies contribute to noise level. Noise is an integrated value over the BW chosen. THD+N reflects distortion components + noise. Your ear (and any human ear) ends about 20kHz. So you need the value of THD+N with 20kHz (22kHz) BW, or even better A-weighted. THD+N value expressed without measurement bandwidth would be a pointless number only.
 

tuga

Major Contributor
Joined
Feb 5, 2020
Messages
3,984
Likes
4,281
Location
Oxford, England
I would like to set 16bit 96dB as threshold, but I am confused with FFT 22kHz and 90kHz.

When using FFT 22kHz, I simply draw a horizontal line at 96dB, like this:
View attachment 106199

However, I don't know if the same rule applies for FFT 90kHz, because it's way different to 22kHz.
Should I also just draw a line at 96dB and it's correct? What would be the explanation for this?
View attachment 106200



Have you tried these tests?

Audible Dynamic Range Sound Test
https://www.audiocheck.net/audiotests_dynamiccheck.php

Blind testing a 24 dB Dynamic Range
https://www.audiocheck.net/blindtests_dynamic.php
 
N

nhatlam96

Guest
Any human ear ends about 20kHz. So you need the value of THD+N with 22kHz BW.
So I can simply ignore FFT 90kHz, because it's outside human range. What is the relevancy of FFT 90kHz for us humans then?

EDIT:
ultrasonic hearing is kind of a thing: "ultrasonic signals resonate the brain and are modulated down to frequencies that the cochlea can then detect."
So, the signals up to 90kHz are modulated down to 22kHz, therefore drawing a line at 96 dB would be correct:
96dB - SMSL SU-9 Balanced USB DAC THD+N vs Frequency Audio Measurements.png


I think it's up to the person to believe in 90kHz or not.
 
Last edited by a moderator:

tuga

Major Contributor
Joined
Feb 5, 2020
Messages
3,984
Likes
4,281
Location
Oxford, England
Higher measurement bandwidth (here 90kHz) means higher measured noise, because noise is measured over higher frequency range and higher frequencies contribute to noise level. Noise is an integrated value over the BW chosen. THD+N reflects distortion components + noise. Your ear (and any human ear) ends about 20kHz. So you need the value of THD+N with 20kHz (22kHz) BW, or even better A-weighted. THD+N value expressed without measurement bandwidth would be a pointless number only.

Is there any research on the effects/audibility of ultrasonic noise on wide-bandwidth amplifiers and on hard-dome tweeters with the resonance peak below 25-30kHz?
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,522
Likes
37,050
I would like to set 16bit 96dB as threshold, but I am confused with FFT 22kHz and 90kHz.

When using FFT 22kHz, I simply draw a horizontal line at 96dB, like this:
View attachment 106199

However, I don't know if the same rule applies for FFT 90kHz, because it's way different to 22kHz.
Should I also just draw a line at 96dB and it's correct? What would be the explanation for this?
View attachment 106200
It has to do with what you want to show with your 96 db line. For harmonics of distortion it will work for any FFT length over any frequency bandwidth. When it comes to noise however FFT length in use and the bandwidth make a difference. It is why I don't like SINAD. You really want to know hum and distortion levels and then you want to know noise levels separately. SINAD combines them.

As an illustration here is the 16 K FFT of -96 db white noise and a single 5 khz tone at -96 db. Notice the cursor was at 5000 hz and shows -96 db, but the graph of the spread out noise is around -136 db because noise spread over the 22 khz bandwidth is at that level in each FFT bin. An FFT like this divides it into 8000 bins of 2.7 hz width. But the single tone has concentrated all the energy into one defined frequency. It will show the same thing with the level at -96 db for the tone at any bandwidth or FFT size.

1610706454283.png


Here is the same signal, but this one is using a 2k FFT. Each FFT bin is 21.5 hz wide and has more energy in it as it covers more frequencies. So the noise is around - 126 db now. Yet the single tone with energy all at the same frequency still shows up at -96 db.
1610706667046.png


So for power supply harmonics or harmonic distortion drawing a line at -96 db works for any FFT over any frequency. For the grassy wideband noise it doesn't translate without details of bandwidth and FFT bin size taken into account.

In my example above just the white noise over the 22 khz bandwidth equals a total of -96 db. However, if I filtered it with a segment 0-11 khz and a segment 11khz-22 khz then each segment has -99 db in it. If I segment it further into 0-5500 hz, 5500-11,000 hz, 11k-16.5 khz and 16.5-22khz then each segment has -102 db in it. This is essentially what an FFT is doing. A 2k FFT has divided the bandwidth into 2048 equal segments.
 

j_j

Major Contributor
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
2,267
Likes
4,758
Location
My kitchen or my listening room.
This question requires a great deal of learning, I'm afraid.

The 'drawing the line across' does not establish dynamic range or SNR. You must add up the ENERGY (not dB, not amplitude) in all of the bands taht are noise, and convert that back to energy, and then compare that to the signal energy.

This means if you are using, oh, say, 1024 frequency bands (2048 transform for bog-normal FFT), you have (if you have an input signal that is directly in the middle of a bin) 1023 bins to add up the energy of.

When you change the number of FFT bins, lots of things change. This is a good 2-hour class discussion. Maybe suitable for a future AES/PNW talk, I dunno.
 

pozz

Слава Україні
Forum Donor
Editor
Joined
May 21, 2019
Messages
4,036
Likes
6,825
So I can simply ignore FFT 90kHz, because it's outside human range. What is the relevancy of FFT 90kHz for us humans then?

EDIT:
ultrasonic hearing is kind of a thing: "ultrasonic signals resonate the brain and are modulated down to frequencies that the cochlea can then detect."
So, the signals up to 90kHz are modulated down to 22kHz, therefore drawing a line at 96 dB would be correct:
View attachment 106207

I think it's up to the person to believe in 90kHz or not.
It's a mistake to look at all of Amir's tests with audibility in mind immediately. First, you have to assess electrical behaviour and understand what the device is doing. This is why the 90kHz bandwidth is used. Only afterwards do you do a psychoacoustic interpretation.

That Wiki article you quoted is based on erroneous research. I'm not sure how to go about combating that sort of thing. It's really wrong to have it up. If any users are more familiar with Wikipedia, could you please add a "fringe content" warning to the page? That "28kHz" upper boundary for hearing has also made it's way into the hearing range page.

The main issue with ultrasonics is imaging and IMD, both of which can have consequences within the audible range. This is a device-level issue, not an anatomical issue.
 

j_j

Major Contributor
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
2,267
Likes
4,758
Location
My kitchen or my listening room.
. This is why the 90kHz bandwidth is used. Only afterwards do you do a psychoacoustic interpretation.
And sometimes use different analysis methods for determinations of audibility or loudness error. This is not a simple subject.
If any users are more familiar with Wikipedia, could you please add a "fringe content" warning to the page? That "28kHz" upper boundary for hearing has also made it's way into the hearing range page.

This is why I've given up on Wiki. You can sort this stuff, and anyone can come in and pile crap.
 

howard416

Active Member
Joined
Sep 12, 2019
Messages
246
Likes
143
I wish we could just move away from SINAD and THD altogether. The noise level just has to be low enough that you can barely hear it when your ear pressed up to the tweeter (or in the case of headphones, just low enough that you can't tell blind between signal chain on and everything off). As for harmonic distortion, it's pretty readily shown that 2nd- and 3rd-order dominant THD has a high threshold of audibility, and that there's tons of (anecdotal) reports that higher orders are much more audible. This also aligns with Geddes's initiative with the GedLee metric (Gm) which correlates well with audibility much better than THD does.

I think the current horizon is to understand the function of the music to mask higher-order harmonic distortion and IMD. When we look at the frequency spectrum of solo instruments, for simplicity's sake, some interesting patterns begin to emerge. Below, a flute playing G4 which is about 400 Hz:

G4.B.sound.gif


We see the 3rd harmonic nearly at the same level as the fundamental, with the 2nd harmonic nearly 15 dB down. The higher orders descend linearly, dB-wise, from 4th to 7th with slightly higher 8th and 9th. An idea that I subscribe to is that if the audio reproduction chain introduces less than 10% (or -20 dB) harmonic distortion, for any given order relative to the associated harmonic in the music, it should be readily masked by the music. In this case, if the DAC/amp/speaker produces less than -35 dB H2, -25 dB H3, -40 dB H4, etc. (just eyeballing here) it should be indistinguishable from a reproduction chain that produces 10 or even 100 times less harmonic distortion. Is this a stretch? Maybe... but a DIYer of some repute noted that nearly 10% THD was inaudible, given that the distortion was primarily HD2.

Now for piano:

C2-ff-pp.gif

C4-ff-pp.gif

C6-ff-pp.gif


Comparison of ff playing:
C1C4C6C8.gif


I don't know about you, but the above suggests to me that loud piano playing is much more forgiving of harmonic distortion in the playback chain, compared to quiet piano playing, whose harmonics are fewer and more muted in level. That is especially apparent in the middle- and higher-registers of the piano, where quiet playing has no harmonics above the 3rd/4th/5th to allow for any musical masking of harmonic distortion. What happens if we end up with -60 dB HD5 coming out of the speaker when the music has it at -93 dB, or perhaps even -70 dB HD7? Hmm... of course, those distortion products might also end up below your threshold of hearing. Fun stuff!

This doesn't even begin to touch on the aspect of IMD audibility.
 
Last edited:

j_j

Major Contributor
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
2,267
Likes
4,758
Location
My kitchen or my listening room.
What do you mean exactly by "error spectrum"?

Um, that has to be discussed in this thread somewhere. It's the spectrum of "original minus test signal", basically. There is a bit more to it, but no, it's NOT just subtracting the magnitude spectra.
 

scott wurcer

Major Contributor
Audio Luminary
Technical Expert
Joined
Apr 24, 2019
Messages
1,501
Likes
2,821
Um, that has to be discussed in this thread somewhere. It's the spectrum of "original minus test signal", basically. There is a bit more to it, but no, it's NOT just subtracting the magnitude spectra.

This kind of opens up a can of worms, doing a broad band null/difference test with phase correction is VERY difficult.
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,522
Likes
37,050
This kind of opens up a can of worms, doing a broad band null/difference test with phase correction is VERY difficult.
Yes it is. The best short cut for digital files I know of is Paul's Deltawave. It does sub-sample alignment to a high degree of accuracy. Also corrects for timing drift between the source and the recording device. So with precise time alignment you can extract the other information accurately.

Then once you have the null or difference file you get into interpretation of it.

Paul has just added a metric of his that goes a little ways toward helping with that taking into account some of the psycho-acoustic properties of our ears.

Maybe @j_j or @scott wurcer could look at what he is doing and give some advice.

https://www.audiosciencereview.com/...c-discussion-and-beta-test.19841/#post-651986
 
Last edited:

j_j

Major Contributor
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
2,267
Likes
4,758
Location
My kitchen or my listening room.
This kind of opens up a can of worms, doing a broad band null/difference test with phase correction is VERY difficult.

No. Extracting the linear part isn't so hard. You do have to extract the linear part. That includes phase and frequency shaping.
 

scott wurcer

Major Contributor
Audio Luminary
Technical Expert
Joined
Apr 24, 2019
Messages
1,501
Likes
2,821
No. Extracting the linear part isn't so hard. You do have to extract the linear part. That includes phase and frequency shaping.

I was thinking of including level dependent effects too like poorly chosen transformers and other non-linear effects. I have some vinyl rips from "The Final Cut" using a notorious pre-amp reviewed here. I might rum them to see what comes out. That LP was one of very few that used Zuccarelli Holophonics and some difference experiments that I did had bizarre results.
 
Last edited:
Top Bottom