• 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!

Study of the effects of Nonlinear Distortion on the Perceived Sound Quality

0bs3rv3r

Active Member
Joined
Oct 24, 2019
Messages
248
Likes
234
If adding distortion makes music more pleasing, I wonder why the music mixers have not found this. Some guitar effects are distortion, so it should not be too hard for them to relate to mixing. There must be some experiment or maybe accident that make people find out adding distortion is better. Of course those are just guesses.


But they have. Plugins for recording and mixing software, that add "pleasing" distortion, already exist, and I have to presume, are being used. Edit: sorry, I see others have already replied about this. Maybe I should add, that I have used just such a plugin and I quite like the effect on the sound. Definitely more enjoyable/pleasing!
 
Last edited:

0bs3rv3r

Active Member
Joined
Oct 24, 2019
Messages
248
Likes
234
I've noticed songs are much dryer and sterile on headphones, that is defiantly one place where tubes can.

With headphones you get no room interractions. A better way, than just adding tubes, would be a room simulation algo - maybe simple light reverb even
 

audio2design

Major Contributor
Joined
Nov 29, 2020
Messages
1,769
Likes
1,830
Music is art. Intentional distortion (tube guitar amps) has been used for ages. Ditto the compression of tape as it enters saturation (was always a process variable). Now distortion and noise is added during post as noted above. Some of use have even made money on it :) Again, music is art. There is no right or wrong, but the artist(s) get to choose what they release.

We know that people prefer in many cases added distortion or noise, but the big question is, does this (always) mean less information reaches the conscious brain from the original pure recording or are we in some cases increasing the information content from the original signal that reaches the brain?

We know that added noise can increase the ability to pick up low level signals (in machines and the brain). And while white noise can mask other noises, it can also keep us focussed on important sounds in an otherwise too quiet environment where everything becomes a distraction. Intentional cross talk can help center an image in a poor acoustic environment.

So what does distortion do ....
 

RHO

Major Contributor
Joined
Nov 20, 2020
Messages
1,182
Likes
1,087
Location
Belgium
but the artist(s) get to choose what they release.
I think most record companies make that decision. I think these days many artists have to make compromises because of requirements put on them by the record company they signed with.
 

audio2design

Major Contributor
Joined
Nov 29, 2020
Messages
1,769
Likes
1,830
I think most record companies make that decision. I think these days many artists have to make compromises because of requirements put on them by the record company they signed with.

I am not sure that is a bad thing :)
 

RHO

Major Contributor
Joined
Nov 20, 2020
Messages
1,182
Likes
1,087
Location
Belgium

RHO

Major Contributor
Joined
Nov 20, 2020
Messages
1,182
Likes
1,087
Location
Belgium
The real one, or the perceived one from use of dynamic range testing software that does not work properly with vinyl?
The real one.
You don't even need to measure it. You can hear it. And artists complain about it.
 

GimeDsp

Senior Member
Joined
Oct 30, 2020
Messages
418
Likes
362
Location
Earth
I think I posted about all the rack gear and plug ins designed for distortion a few pages back, its been popular since the 80s and the"distressor" is a very common piece of gear now.

Heres the thing though. In a treated room with boss hog speakers a mastering engineer can apply distortion carefully to the mix. With side chains from mid/side possessors/eq they can also treat the center and sides seperatly.

In a non treated room by the time you hear the distortion you are likely a great deal beyond hearing threshold in a treated room.

I am not saying all mastering engineers use subtly just that that can with high resolution systems.

We know from the loudness wars and from famed engineer Bob Katz that the artist/label gets what they want!
If its high distortion and low dynamic range they get it!
 
OP
Pinox67

Pinox67

Member
Editor
Joined
Apr 8, 2020
Messages
85
Likes
148
Location
Italy
Some friends have pointed out to me that some aspects of my previous post are described too “hermetically”.

One of the aspects on which I struggled the most in providing explanations is that relating to the sentence where I affirm that "the amount of distortion digitally injected is that actually audible only at a certain level of listening", set as a parameter in the model. Therefore, even the listening tests are only valid for this listening level.

To better explain this point, it is necessary to spend few words on how a preamp, or the preamplication section of an integrated amplifier, "arrives" at the output signal with the desired level, based on the volume control.

In short, the amount of the amplification of the input signal applied by the preamp stage is always the same: it is the one reported in the technical characteristics under the item Gain (G below). For a preamp it is typically between 10dB and 20dB. For example, if we have a sinusoidal signal with a peak-to-peak voltage value Vin = 1V in input to an amplifier with G = 20dB, we will have at output the same sinusoid with Vout = 10V. Since this value determines also the listening level, it must be controlled by the user. This naturally happens with the volume knob, which attenuates the signal. Based on the point of the preamp circuit where this attenuation is performed we can distinguish two design approaches:

1. Attenuation is performed on the input signal and then amplified.
2. The input signal is first amplified and then attenuated at the output.

In my previous sentence I was implicitly referring to the first approach, which is by far the most used. Let's briefly see both below, in relation to the static nonlinear distortion model described previously. In the following, please remember that the masking effect of our ear:
  • It is higher for frequencies close to the fundamental; in particular it is more extended towards frequencies with a higher value than the fundamental.
  • It is higher when the level of the fundamental is higher. In other words, at low volumes the masking effect is much more modest, for intensity and extension.
Type 1: Input Attenuation + Amplification
For the first type of preamp the signal management is outlined in the following figure.

Figure 1 - Input attenuation + Amplification
Pasted Graphic 6.png
Basically, the input signal x(t), which we assume of fixed intensity Vin for simplicity of exposure, is attenuated by a certain amount controlled by the volume knob, before being subjected to the fixed amplification part of G value. The volume control then determines the portion of f(x) curve (responsible for the distortion) around the origin that is actually used and therefore the maximum extension to avoid clipping. From the point of view of output distortion this implies that:
  • Given that the deviation of the linearity of f(x) increases with the rise of the value of x, then the distortion will also follow the same trend: low (in dB) at low volumes and higher at higher ones. Considering the masking effect of our ear, if the distortion is more concentrated on low orders, this behaviour is a positive aspect.
  • If the linearity deviation of f(x) is already present in the neighborhood of 0, such as crossover distortion (actually, unlikely for a preamp stage) then we will have high distortion even at low volumes and, given the harmonics of high order generated for this situation, and listening will likely be unpleasant. As the volume increases, these distortions decreases in dB and the masking effect makes it less perceptible.
For this type of preamp the digital distortion simulation is valid only for a certain volume level. On the simulator it is possible to configure how much of the f(x) curve to use, and therefore how much distortion to inject, which cannot then be modified once the distorted digital signal has been generated, as happens by modifying the volume of the real amplifier.
What has been described can also be applied to power amplifiers, as the gain G is fixed: it is the preamp that determines the part of f(x) that is used via Vin.

Type 2: Amplification + Output Attenuation
For the second type, used less frequently, the signal management is outlined in the following figure.

Figure 2 - Amplification + output attenuation
Pasted Graphic 5.png
Here the input signal x(t) is amplified first and then attenuated before the output to the desired level. In this case, a much wider part of the curve than the previous type is used. The amount (in dB) of the output distortion is practically independent of the set volume. In theory, amplifiers with crossover distortion could benefit from this design, since this is then attenuated at the output and can first be “covered up” by distortions due to the more distant parts of the curve.

For this type of preamp the digital simulation of distortions is correct for any volume level, since the amount of curve f(x) to be used in the simulation is always the same, like for the real amplifier, regardless of the volume.

Final remarks
To understand which of the two types of situations we are in, it is necessary to measure harmonic distortion for different volume settings. Assuming crossover distortion is absent, if the distortions (in dB) are practically independent of the volume then we have the second type; if the dB values increase as the volume increases, we have the first type. What about the differences in sound? In my limited experience, changing the position of the potentiometer, at input or output, on the same preamps in class A you have a greater characterization (in terms of warm or dynamic contrast) of the sound in the second case; in the first you have a feeling of greater neutrality.
 
Last edited:

tktran303

Addicted to Fun and Learning
Forum Donor
Joined
Mar 27, 2019
Messages
683
Likes
1,179
A lot has changed in the marketplace since Earl Geddes stated that
“ THD and IMD are meaningless measurements of distortion as far as perception is concerned.”

So does all this explain why the world has moved to 3” bluetooth enabled peakers and laptop speakers as being good enough; while greying hair audio enthusiasts/ audiophiles clutch onto speakers which are the size of shoeboxes (the very minimum) all the way up to refrigerators ?

Not trying to be facetious here, but to me non linear distortion is related to SPL.
Do you want to something to try to emulate the sound in a jazz club, a bar, a nightclub or concert hall?
Or do you just want to listen to your favourite tunes in the background whilst cooking dinner.

So if you’re playing at 60dB in your bedroom any 3” full range driver is good enough from 100Hz to 10Khz; where the most of music can be captured (“enough bass; mids; and treble”)

Now Take that portable speaker outdoors and turn it all the way up and it sounds
“harsh, too loud, turn it down” what’s happening is HD3 is skyrocketing up to 10-100% and all the higher order harmonics from H5-H9 have gone up too; well beyond -30dB

To say “ Pursuing a loudspeaker design to lower the distortion is a waste of time if its nonlinear distortion that you are trying to lower. It simply doesn't matter.” Is a blunt instrument because- you don’t have to care if you are playing at 60dB @1m, but you care about playing loudly and cleanly at hitting 96 dB at 10m.

I’ve yet to see Earl Geddes do the double blind tests again where the subject gets to play with the volume control freely (external validity) and see if they when (a)he prefers the JBL M2, Geddes Summa, or the diminutive the JBL Charge 3…
 
Last edited:
OP
Pinox67

Pinox67

Member
Editor
Joined
Apr 8, 2020
Messages
85
Likes
148
Location
Italy
A lot has changed in the marketplace since Earl Geddes stated that
“ THD and IMD are meaningless measurements of distortion as far as perception is concerned.”

So does all this explain why the world has moved to 3” bluetooth enabled peakers and laptop speakers as being good enough; while greying hair audio enthusiasts/ audiophiles clutch onto speakers which are the size of shoeboxes (the very minimum) all the way up to refrigerators ?
...

From what I have read about GedLee's work, certain types of distortions on the speakers are much more tolerated (or inaudible) respect to others; those injected by electronics, also of much lighter entity, are often more annoying. I have no doubt that there is a relationship between distortion perception and SPL. I can add that any audio device is designed to work in specific situations, where its behaviour can be considered let's say correct; forcing it to work in different situations will generate for sure unpleasant sound. Anyway, I don't have in-depth skills to go into more detail on speaker distortions.

As I think it is very clear, the aim of my study is to investigate the non-linear distortions introduced by the electronics, mainly by acting on the preamp stage: only this is modified with all the other elements of the listening chain being equal (including the environment, which also plays an important and often neglected role). Then, measurements, listening sensations and mathematical models are correlated. It is pure utopian to hope that it will put the word "end" to the discussion on distortions, but I would be happy if it is useful to better understand our electronics and how they make audio reproduction more or less pleasant for us.
 
Last edited:
OP
Pinox67

Pinox67

Member
Editor
Joined
Apr 8, 2020
Messages
85
Likes
148
Location
Italy
Some updates with the new year.

I have not yet been able to perform the "formal" tests as described in my previous posts. Anyway, some people send me their reference music tracks and then I returned multiple versions of them: the original plus the others with varying amounts of distortion added. Obviously, without specifying the nature of each, for a blind listening. Much less formal, but with encouraging results.

I spent most of my time working on the simulation program, to make it more representative of the behaviour of our reproduction systems. To sum up, the non linear distortion model that I have implemented at the moment is of the static type: the input/output curve (an example is in Fig. 2 of this post) is "fixed", uniquely identified by setting the amount of distortion for each harmonic, obtained by simple HD measurements. Also the changes in their value due to the variation of the volume and the relative phases result univocally identified. Although the volume dependence has to be handled as explained here, if we compare the phases of the distortions with those obtained from the measurement of preamps of all kinds, we get only a modest agreement.

The reality is that the input/output curve is variable: it “deforms” according to the frequencies present in the input signal. This effect is caused mainly by the presence of memory effects in the amplifier. Therefore, if we want to represent more faithfully the distortions generated by our audio devices, we need to adopt a more sophisticated model, like the Volterra Series, which can take dynamic aspects into account. To update the model in this direction, I am proceeding as shown in the following fairly classic scheme:

Figure 1 - Architecture for the SW simulation for dynamic systems
Screenshot 2022-01-02 at 11.30.35.png

Basically, the Test Signal Generator creates a digital test signal x[n] in high resolution. This consist of a Synchronized Swept Sine which has the remarkable capacity to separate the impulse responses for each harmonic distortion order.
The test signal, through the DAC, reaches our preamplifier or power amplifier (DUT) in the analog domain; the related output is then converted back to the digital domain via an ADC and stored by the Response Signal Collector.
Then, the two digital input x[n] and output y[n] signals are processed in the Kernel Identifier, which apply essentially a deconvolution operation. The results are constituted by the hi[n] parameters of the model, i.e. the Diagonal Volterra Kernels.

If this modelling "captures" the essential aspects of the DUT correctly, these parameters allow the Simulator, structured as depicted in Fig.10 of the same post, to digitally reproduce the behaviour of the DUT with negligible errors for any input signal x'[n]. This processing is not in real-time (that's not the goal at the moment); to take also into account the constraints already described on the signal level. A Dummy NLS component, whose behaviour is defined a priori, can be used instead of the DUT (and of DAC+ADC) for further checks of the whole procedure.
All SW processing component described are already implemented, with the exception of the Kernel Identifier, currently being completed.

Lastly, with the aim of more sharing and therefore collecting useful feedback on this work, I started a thread on the same topic on this site, linking the information shared on this thread. Here too, there have been very interesting contributions.
 
Last edited:
Top Bottom