• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). Come here to have fun, be ready to be teased and not take online life too seriously. We now measure and review equipment for free! Click here for details.

Audibility thresholds of amp and DAC measurements

Blumlein 88

Major Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
8,130
Likes
10,048
Yes, I know.
It is something different than writing a DSP program that undoes the damage though.

Have you characterized a device and fed the opposite of it in your distortion program ?
I was already thinking of trying that with Deltawave.
 

solderdude

Major Contributor
Joined
Jul 21, 2018
Messages
4,301
Likes
7,987
Location
The Neverlands
Isn't that rather what feedback does? The distortion undoer signal. Some audiophile company can create a story about that, and suddenly feedback will be good again.
Amplifiers add distortion even with feedback loops because feedback can't undo all distortions perfectly.
Perhaps adding the opposite error signals is more like feed forward ?.

I don't think it will work very well with power amps driving difficult loads unless the characterization would be done with the speaker connected.
 

pma

Addicted to Fun and Learning
Joined
Feb 23, 2019
Messages
631
Likes
849
Location
Prague
1) depends on loopgain frequency response and amount of loopgain, how effective is feedback as a result
2) the more linear is the circuit before the feedback is applied, the better the results and less work for feedback.

I think that endless jokes on audiophiles are unnecessary.
 

pkane

Major Contributor
Forum Donor
Joined
Aug 18, 2017
Messages
1,264
Likes
1,784
Location
North-East
Isn't that rather what feedback does? The distortion undoer signal. Some audiophile company can create a story about that, and suddenly feedback will be good again.
Funny thing is, I just started adding feedback "distortion" to the DISTORT app. It seems to reduce distortions :)

With negative feedback, THD of -110dB:
1579106138052.png


And without, THD of -82dB:
1579106174728.png
 

solderdude

Major Contributor
Joined
Jul 21, 2018
Messages
4,301
Likes
7,987
Location
The Neverlands
1) depends on loopgain frequency response and amount of loopgain, how effective is feedback as a result
2) the more linear is the circuit before the feedback is applied, the better the results and less work for feedback.

I think that endless jokes on audiophiles are unnecessary.
1: Of course it does, no one here thinks otherwise.
2: Of course, no one here thinks otherwise.

The point was that J_J mentioned the difference between linear and non linear distortion and that linear distortion was easy to undo by applying the reverse filter again.
Paul then coined that if one feeds an amp (with a certain transfer function) with the opposite transfer function errors that the end result would be undistorted amplification.

This (IMO) is not that easy. Paul is experimenting with his programs (which I applaud)
 
Joined
Dec 7, 2019
Messages
207
Likes
76
Linear operations (in the absense of noise) are reversible. Your all-pass filter can be convolved with the time-reverse of the allpass filter, and presto, the "distortion" is gone. That's because IT IS NOT DISTORTION, it is a linear filtering operation.

When a nonlinear operation happens, the result is lossy, and information is lost. This is a fundamental difference, and one you need to learn.
The result is lossy/irreversible in both cases because corrupted waveform at the output of audio device can not be corrected in any way; it is on the way to your brain for perception and comprehension.

Does your “fundamental” distinction between linear and non-linear distortion manifest itself during listening? Can a listener discern those types of distortion? Is linear distortion less audible/more pleasant?
 

solderdude

Major Contributor
Joined
Jul 21, 2018
Messages
4,301
Likes
7,987
Location
The Neverlands
There is a substantial difference between linear and non linear distortion.

EQ is linear distortion and when applied correctly will improve things.
All speakers and headphones have vast amounts of linear distortion and can have small amounts of non linear distortion.
The linear distortion of speakers can be compensated for (to a certain extend).
Sometimes many dB's are needed which is directly audible and fully reversible on the input of the amplifier.

Non linear distortion is very low level and in general is not sound improving when it is reaching clearly audible levels and also cannot easily be compensated.

One can have only linear distortion and only non linear distortion (assuming audible band) these are not related.
Of course one can also have both where linear distortion is directly audible.
 
Joined
Dec 7, 2019
Messages
207
Likes
76
There is a substantial difference between linear and non linear distortion.

EQ is linear distortion and when applied correctly will improve things.
All speakers and headphones have vast amounts of linear distortion and can have small amounts of non linear distortion.
The linear distortion of speakers can be compensated for (to a certain extend).
Sometimes many dB's are needed which is directly audible and fully reversible on the input of the amplifier.

Non linear distortion is very low level and in general is not sound improving when it is reaching clearly audible levels and also cannot easily be compensated.

One can have only linear distortion and only non linear distortion (assuming audible band) these are not related.
Of course one can also have both where linear distortion is directly audible.
So, in a nutshell:
- linear distortion is less audible and less annoying; their most known negative impact is corruption of tonal balance and/or spatial sound imaging.
- non-linear distortion is more audible and annoying (at comparable levels); in some cases (dithering, adding even harmonics) they can improve attractiveness of a distorted sound.

It looks like we should pay more attention to non-linear ones as more disturbing. But having an input and output signal of some DUT it's hard to measure what types of distortion are present. Even if we discover those types there will be a question of proper accounting them for audibility. This is a road to nowhere. It will be much easier, cheaper and economically efficient to design and manufacture the audio chips, which have ALL distortions below some low threshold of audibility (just to stop these endless and useless discussions of which distortions are better)).
 

solderdude

Major Contributor
Joined
Jul 21, 2018
Messages
4,301
Likes
7,987
Location
The Neverlands
So, in a nutshell:
- linear distortion is less audible and less annoying
- non-linear distortion is more audible and annoying (at comparable levels)
Exactly the opposite.
Linear distortion is much more audible and can be an improvement or totally screw things up and every possibility in between.
Non linear distortion, certainly in music can remain undetected until it reaches audible levels. And even then, depending on the harmonic spread and frequencies it does not necessarily lead to lower fidelity but could well be in other cases.
It is very easy to see in measurements how much of each type of distortion is present.
Only when nulling it get's a bit more complicated as the null needs to be analyzed further and compared to the input to see what comes from phase and amplitude distortion (linear) and what comes from non linear distortion.

You know this very well... what's up with it ?
 
Last edited:

j_j

Addicted to Fun and Learning
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
634
Likes
916
Location
My dining room.
I'm trying to learn, but having a hard time. Any transfer function that produces a one-to-one mapping between the input and output can, in theory, be reversed. Including non-linear. Information is lost only if the mapping is many to one (e.g., clipping/compression) or one to many (e.g., noise).

Are you then defining non-linear distortion as that which is generated by a subset of all the non-linear transfer functions that are lossy?
No, if you have a nonlinear operation, in any real system, you can't reverse it. At best you can if and only if there is no bandwidth limitations, noise, or other things between the distorted and the distortion inversion. Furthermore, such nonlinearities would have to be memoryless. No real system is memoryless (barring a contrived digital system that simply amounts to a remapping, which is not really a nonlinearity).
 

j_j

Addicted to Fun and Learning
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
634
Likes
916
Location
My dining room.
Isn't that rather what feedback does? The distortion undoer signal. Some audiophile company can create a story about that, and suddenly feedback will be good again.
Feedback linearizes non-linear systems (well, if you do it right, there are cases... ). It is not memoryless, of course. If you can put feedback AROUND (note, around, not after) a nonlinear system, there is a decent chance you can improve its performance (well, for most common distortions, at least).

But that's not two systems cascaded, that's one system including feedback.
 

j_j

Addicted to Fun and Learning
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
634
Likes
916
Location
My dining room.
Exactly the opposite.
Linear distortion is much more audible and can be an improvement or totally screw things up and every possibility in between.
Non linear distortion, certainly in music can remain undetected until it reaches audible levels. And even then, depending on the harmonic spread and frequencies it does not necessarily lead to lower fidelity but could well be in other cases.
It is very easy to see in measurements how much of each type of distortion is present.
Only when nulling it get's a bit more complicated as the null needs to be analyzed further and compared to the input to see what comes from phase and amplitude distortion (linear) and what comes from non linear distortion.
A quick way to tell if you're dealing with linear modifications or distortions is to look at the error spectrum. If it contains only the frequencies in the input signal, it's linear. Another way to put this is that if you designed the right equalizer, you could null the error signal (of course there would be noise, no getting around that) by equalizing the second input to the subtractor. This would also then work, assuming you got the equalization right (both amplitude and phase), for any other arbitrary signal.

You'd still have a noise floor, which I would call "noise", of course, but you could get down to a noise floor, perhaps a shaped one.

If you try this with a nonlinearity, first you'll have to put in zeros at the sum/difference/etc distortion products, and that is not going to work with a different input signal. (and this would be a very difficult project for any signal with any particular complexity at all)
 

j_j

Addicted to Fun and Learning
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
634
Likes
916
Location
My dining room.
The result is lossy/irreversible in both cases because corrupted waveform at the output of audio device can not be corrected in any way; it is on the way to your brain for perception and comprehension.
Nonsense, your allpass filter can be reversed down to the noise floor. I don't think you understand what "linear" means. In the absence of noise, linear is reversible. An all pass filter is reversible via convolution, to any arbitrary limit, by taking the impulse response, reversing it in the time domain, and then convolving that with the allpassed signal. You can't make another recursive allpass, of course, that pole outside the unit circle does give a bit of trouble.

Try a small square law term along with your allpass filter, now, m'kay? Try the two orders, first allpass, then square law (say gain is x + .1 x*x) for signal x, and vice versa.

Yeah. It matters. There's a lesson in that.
 

j_j

Addicted to Fun and Learning
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
634
Likes
916
Location
My dining room.
Ok. Audibility. Linear operations do not add new frequencies. You can change timbre, tonal balance, etc, with linear operations, but you don't create new frequencies.

Your ear is a frequency analyzer. It is relatively insensitive to small changes in tonal balance, compared to having new frequencies added to a signal.

Nonlinear distortions DO add new frequencies. If these are small enough, no harm, but when "new frequencies" rise above the masking threshold for a signal, most often this is bad, bad news, unless you're creating a sound, say, by overloading a guitar amp on purpose. One of the gotcha things that nonlinear operations do is create sum and difference frequencies, and therein lies the disaster, say, if you create new frequencies in a part of the audio spectrum where there is no energy. The usual result is "ech" if done without intent, at least.

I'm not addressing artistic modifications to an original. For that, use whatever you want.
 

solderdude

Major Contributor
Joined
Jul 21, 2018
Messages
4,301
Likes
7,987
Location
The Neverlands
Is it possible to define thresholds of audibility for linear and non-linear distortion?
As J_J said this depends on the signals used. Easier to define with just 2 tones than it is with music. Also in some recordings it will be different than in others. Depends on masking and the frequencies involved.

You can test linear for yourself. Just use an equalizer and pull up or down 1 (or more) slider(s) and see where your threshold is with music and/or test tones. Switch between applied and not applied EQ.
Use Paul's software to get an idea of audibility with music and/or tones.

You will find that you will get different numbers in different situations so there won't be one number. There will be a range.
If you want to be sure.. just pick the lowest number and you are probably safe in other than tested circumstances as well.
 
Joined
Dec 7, 2019
Messages
207
Likes
76
Nonsense, your allpass filter can be reversed down to the noise floor
Quite the opposite - the essence. When the output of a DUT is corrupted - it is corrupted, or you need to insert some DSP between amplifier and headphones/speakers. For a listener this corruption is final/irreversible.
 
Joined
Dec 7, 2019
Messages
207
Likes
76
As J_J said this depends on the signals used. Easier to define with just 2 tones than it is with music. Also in some recordings it will be different than in others. Depends on masking and the frequencies involved.

You can test linear for yourself. Just use an equalizer and pull up or down 1 (or more) slider(s) and see where your threshold is with music and/or test tones. Switch between applied and not applied EQ.
Use Paul's software to get an idea of audibility with music and/or tones.

You will find that you will get different numbers in different situations so there won't be one number. There will be a range.
If you want to be sure.. just pick the lowest number and you are probably safe in other than tested circumstances as well.
It seems that the ground of the discussion has been lost. What and why we are measuring? What questions we want to answer by means of objective audio measurements? And what questions can't be answered by such measurements. Let's put aside for simplicity the measurements during R&D and focus only on measuring ready-made DUTs. IMHO the situation is very similar to the one we have with blind listening tests (BLT). In short - they can't discover which DUT is better, but only - which DUT is closer to reference. If most of subjects do not percept the difference during BLT then all done - the DUT is transparent. If it is not transparent then we can research further and find what kind of difference in sound was audible. The same approach works for objective measurements as well. We can't discover by measurements what DUT is better; for the purpose we need elaborated model of human perception/comprehension of sound (actually doable but hardly reasonable). Instead we can measure how close an output signal to an input one. If the difference is small enough (we need practical thresholds for the purpose) then all done - the DUT is transparent. If it is not transparent then we can research further distortion product and TRY to assess its audibility/annoyance. At this point j_j's fundamental distinction between linear and non-linear distortion can help. In df-metric I have another solution but analysis of residuals can also be appropriate.

You may have a different view of the objective measurements, please, share it. We should have a common ground for the discussion of audibility of distortions and corresponding thresholds.
 

Similar threads

Top Bottom