Wow, congratulations for being brave enough to tackle the question of how measured non-linear distortion is (or can be) correlated to perceived sound quality. This is not a trivial task, especially if you're doing this with limited resources. I briefly glanced over the other threads and posts you mentioned, and there is some good stuff (both from you and others who commented on your work).
It would be great if your work could be scaled up such that (many!) more people could take the tests in a well-controlled manner. Would it be possible to set up a bunch of test files or maybe even a website that provides the playback of the test tracks such that the listener does not know if he/her is listening to "distorted" or "undistorted" test tracks, and then reports his/her perceived results back into the system? This might allow you to establish a much bigger data set, which would provide a sound basis for a statistical assessment of the "measured vs. perceived" question.
Keep it up!
Thought experiment as a simple case study:
Tube amplifiers + high efficiency speakers without demodulation rings.
Simultaneous bass and high notes through the speaker would ordinarily create intermodulation, but because the amplifier happens to be a current source, dynamic changes in the inductance in the speaker are ignored so that mechanism of IMD is reduced.
So that's one example of 2 high distortion "components" performing well together, which could be easily confused for a pleasing addition of distortion to make things sound better.
I can think of others, eg crossover distortion (common with low distortion solid state). Aberrations with a fixed amplitude mean that loud signals are disproportionately cleaner than quiet ones. Again, resulting in potential confusion as to why a "high distortion" design may be preferable to an allegedly lower distortion one. The simple answer could be the wattage.
Don't know about the general practicality of canceling nonlinear distortions between two different pieces of gear, but would agree that systems have to be taken as a whole. After all that's what the test subject is really evaluating: How does a DUT sound in a particular system?
The other thing probably pointed out earlier is IME some people will accept more HD in order to more of something else that they like. Sometimes there can be a tradeoff between measured distortion numbers and subjective quality of the stereo illusion. In addition, sometimes a lot of feedback is used to improve measured distortion numbers. It may be less easily measured side-effects of high loop gain that affect sound in a way that some people want less of. I know the latter statement may raise some eyebrows, but we know some people have been complaining about the sound of feedback for a long time, particularly in regards to global feedback. IMHO what they are complaining about is most likely not entirely attributable to imagination.
The other thing probably pointed out earlier is IME some people will accept more HD in order to more of something else that they like. Sometimes there can be a tradeoff between measured distortion numbers and subjective quality of the stereo illusion. In addition, sometimes a lot of feedback is used to improve measured distortion numbers. It may be less easily measured side-effects of high loop gain that affect sound in a way that some people want less of. I know the latter statement may raise some eyebrows, but we know some people have been complaining about the sound of feedback for a long time, particularly in regards to global feedback. IMHO what they are complaining about is most likely not entirely attributable to imagination.
Last edited:
At the end of this post reported initially I described how I would like to run controlled tests, yet to start.
At the moment I only performed blind listening tests on my system or those of friends, with encouraging results. For other people, I had their reference music tracks sent to me and then I returned multiple versions of them: the original plus other versions with the addition of different amounts of non-linear distortion. Of course, the person does not know which track is the original and which is the modified one, for a blind listening. Less formal, but still I think it's useful.
This last procedure could also be extended, yes. With the support of an administrator of this site it could be done... I'm available!
However, there is a small detail to add. As already stated, the non-linear distortion model adopted at the moment is of the static type. This means that the input/output curve (fig. 2 of the same post) is univocally identified by setting the amount of the desired distortion for each harmonic. Also the relative phases of harmonics result univocally identified with specific ratios. Well, I compared these values with those obtained from the measurement of dozens of preamps of all types: the agreement is weak.
This because our audio devices are also affected by dynamic non-linear distortions, caused by the non-linearity of the components inside the amplifier, together with thermal effects. The effect on the model is a variable input/output curve, dependent on the frequencies of the input signal, and probably also on its level. Therefore, if we want to represent more faithfully the non-linear distortions injected by our audio devices, we need to adopt a more sophisticated model, the Volterra Series, which can take these aspects into account.
I am currently working on this model; the reference scheme is as follows:
Basically, a digital Test signal is generated, a Synchronised Swept Sine which through the DAC reaches our preamp or power amp (DUT) to measure in analog. The property of this test signal is that of separate non-linearities for single harmonic. The DUT output is then converted back to digital via an ADC and stored. Then, the two digital input and output signals are processed (de-convolved) in the Kernel Identifier block to identify the parameters of the model (the Volterra Kernels). These Kernels hi[n] allow the Simulator to reproduce the behaviour of the preamp digitally, and then validate the model using any signal x[n]. A Test NLS component can be used as known block instead the DUT to further checks.
Of the processing components described, I have already created all of them except the "Kernel Identifier", which is currently being completed. Suggestions/comparisons with similar works are always welcome.
Last edited:
Possibly of interest: http://www.sintefex.com/docs/appnotes/dynaconv.PDF
Also perhaps: https://scholar.google.com/scholar?q=dynamic+convolution+audio&hl=en&as_sdt=0&as_vis=1&oi=scholart
Also perhaps: https://scholar.google.com/scholar?q=dynamic+convolution+audio&hl=en&as_sdt=0&as_vis=1&oi=scholart
As far as I can see you're two posts are self-contradicting but you probably overstated your opinion in your first post?
Anyway, an interesting investigation from Bob Katz, an audio mastering engineer, "Adventures in Distortion" which I think Pinox already posted?
IMHO there are problems with Mr. Katz's conclusions. As people often tend to do he has oversimplified the situation. For one thing, tube distortion is not exactly static 2nd harmonic distortion. One can add static 2nd harmonic fairly easily to find out if it does what is claimed in the 'investigation.' Two ways of doing that here: (1) ESS dac chip harmonic distortion compensation registers, and (2) Crane Song HEDD 2nd and 3rd harmonic distortion processing. Both sources of distortion sound better turned off, at least to my ears. Doesn't matter which of the available power amps I use.
I see Pinox already mentioned that the right level of 'good distortion' is probably needed to produce a 'realistic illusion'.
And the level & type of this distortion varies depending on all the factors in the reproduction chain from recording, listener, playback system & room.
This is a highly complex scenario & the illusion itself, created by a stereo system of reproduction, is very fragile & on the edge.
In order to be perceived as realistic sounding it needs to match the internal processes of auditory perception that have been learned & refined from the audio world we encounter daily.
The fragility is likely arising from the fact that auditory perception itself is thought to involve a competition between internal models which best match the nerve impulses arriving through our auditory system. These aspects are being carried out in the ongoing research into auditory perception, particularly ASA (auditory scene analysis).
ASA is concerned with how we internally create & therefore perceive auditory objects, auditory streams & auditory scenes out of the nerves impulses arriving along the auditory nerves from two ears. These nerve impulses are just a continuous stream in linear time of electrical signals, not grouped in any way into objects or streams. It is our internal analysis which groups such signals into auditory objects. In an orchestra in full flight we extract & group the nerve impulses which arise from each instrument playing. It appears that we are continually analysing this electrical impulse stream & doing a best matching process based on the rules we have embodied about how auditory objects behave in the real world. In other words, it's a speculative or heuristic internal process. As a result we can perceive 'levels of auditory realism' when listening to our stereo playback systems which are just about producing enough audible cues to fool us
It may be that a certain level & mix of distortions overcomes the fragile nature of stereo reproduction & creates a stronger/more stable perception which better satisfies what our auditory perception considers a better match to our internal real world auditory model.
In a way, this stereo audio illusion is a victim of its own success - we don't look at reproduced visual images as immersive or giving the illusion of 'being there', unless we use some form of 3D glasses which create different signals arriving at our retina.
On the other hand, I'm open to the idea that it is not added distortion that is needed for 'realistic audio' but perfect reproduction of the signal, at every level in the chain i.e the full playback chain as a whole.
And the level & type of this distortion varies depending on all the factors in the reproduction chain from recording, listener, playback system & room.
This is a highly complex scenario & the illusion itself, created by a stereo system of reproduction, is very fragile & on the edge.
In order to be perceived as realistic sounding it needs to match the internal processes of auditory perception that have been learned & refined from the audio world we encounter daily.
The fragility is likely arising from the fact that auditory perception itself is thought to involve a competition between internal models which best match the nerve impulses arriving through our auditory system. These aspects are being carried out in the ongoing research into auditory perception, particularly ASA (auditory scene analysis).
ASA is concerned with how we internally create & therefore perceive auditory objects, auditory streams & auditory scenes out of the nerves impulses arriving along the auditory nerves from two ears. These nerve impulses are just a continuous stream in linear time of electrical signals, not grouped in any way into objects or streams. It is our internal analysis which groups such signals into auditory objects. In an orchestra in full flight we extract & group the nerve impulses which arise from each instrument playing. It appears that we are continually analysing this electrical impulse stream & doing a best matching process based on the rules we have embodied about how auditory objects behave in the real world. In other words, it's a speculative or heuristic internal process. As a result we can perceive 'levels of auditory realism' when listening to our stereo playback systems which are just about producing enough audible cues to fool us
It may be that a certain level & mix of distortions overcomes the fragile nature of stereo reproduction & creates a stronger/more stable perception which better satisfies what our auditory perception considers a better match to our internal real world auditory model.
In a way, this stereo audio illusion is a victim of its own success - we don't look at reproduced visual images as immersive or giving the illusion of 'being there', unless we use some form of 3D glasses which create different signals arriving at our retina.
On the other hand, I'm open to the idea that it is not added distortion that is needed for 'realistic audio' but perfect reproduction of the signal, at every level in the chain i.e the full playback chain as a whole.
Last edited:
Yes, agreed, masking is an oversimplification, it's appears not to hold in all situations - auditory perception is much more nuanced than simple rules imply. The problem with a lot of auditory testing/research in the past (which is now beginning to be recognised), is that not enough complex signals were used. When we are talking about a 'realistic auditory illusion', we are inferring this arises from complex signals - the interrelationships of these signals & the stability of these relationships are seemingly important aspects in creating & maintaining this 'realistic illusion'
There's one important, fundamental thing missing : music. Music as a block, comprehensive of all the human efforts that have been made to arrive at today. So stereo illusion, as has been called, is just position of the scenario created by all the instruments in the orchestra. Or the position of a single instrument...just because the way we catch a sound is the same way when live or recorded...
It's to be repeated that sound is a slow propagation phenomenon, contrary to light, the wave nature is the same but different...
So, we have incorporated that 3D glasses when listening to music, but works in sphere that comprises personal taste, knowledge, culture and might be corrupted by mental, physical status...attention can be solicited by various means.
That's why we need preamps and amps with the pot on the front to adjust the spectrum to our liking.
There are those, no need to name them, who are totally against any form of distortion, even pleasing. I was told few times when working on tube pre that trying to find pleasing distortion is 'disrespectful to the artist' and its 'putting shades on mona lisa' and that my tube pre is not an audio device but effect box.
Many people, many opinions.
Sometimes, i believe, we do not even agree why we listen to the music. How can we agree what sounds good.
There are those, no need to name them, who are totally against any form of distortion, even pleasing. I was told few times when working on tube pre that trying to find pleasing distortion is 'disrespectful to the artist' and its 'putting shades on mona lisa' and that my tube pre is not an audio device but effect box.
Many people, many opinions.
Sometimes, i believe, we do not even agree why we listen to the music. How can we agree what sounds good.
Regarding pure accuracy verses some 'effects box' sound, IME it turns out we can make a dac so accurate that can reveal all the flaws and limits of resolution in every CD quality recording. This assumes the rest of the system is up to the same level of accuracy. The problem then becomes that we don't enjoy listening to the sound for very long. It doesn't sound 'musical,' for want of a better term. Some of the old amplifier technology, the best of it, happens to introduce some complex yet subtle time-variant musical distortion. It very small doses it can sound extremely 'musical.' Listening enjoyment returns. Not many people know how to do it well, although much useful information is scattered around threads in the forum, some from many years ago. Some of it is also wrong. Personally, I find the result both disappointing and interesting. Its a sort of imperfect solution to the sound of CD quality we are more or less stuck with for now. High quality analog sources, phono and tape, don't seem to benefit as much. At least the foregoing is my opinion at the moment. Things are still evolving so could change my mind later.
Not possible with speakers and room acoustics. Those two are still orders of magnitude higher in inaccuracy than the cheapest dac sold these days.
Seems to be true. Stopped using opamps except maybe one for I/V with switched resistor dacs. Helps some to power and bypass it as though it doesn't have infinite PSRR.
Esa Merilainen (www.current-drive.info) or Pavel Macura (www.pmacura.cz/speaker_dist.htm) have some material on the matter, and probably others as well.
No amplifier has zero distortion. But if we can made distortion very low but we did not do it, it is effect box. I design amplifier with a lot of negative feedback, but the sound is not fatigue and dry. It can sound warm in some recording. Dr. Kolinummi design amplifier without global negative feedback and have very low distortion. It proved that with or without global negative feedback, amplifier can have very low distortion. Of course sound perception is not only about distortion. And distortion is not only THD, but harmonic profile of the distortion, slew limiting distortion, thermal distortion, etc.
bimo
If you add something in reproduction and called it musical, then the recording is not musical. Like Monalisa painting, you want add sun glasses because you think it better.
Mark Tillotson
If you want to change the sound from what was recorded, then you are using something called an effects unit, not a preamplifier, there's nothing wrong in that, so long as you call it what it is and don't confuse sound reproduction from sound alteration.
And with a good recording trying to change its character isn't very flattering to the artists and recording engineer! Its rather like putting sunglasses on the Mona Lisa ...
What is the probability that both of you came up with the idea, that designing audio pre for pleasing sound is like putting sun glasses on monalisa? Can you explain?