• 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!

Alternative method for measuring distortion

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
How do you think, if I suggest mass testing of modern portable audio devices with df-metric for the purpose of creating alternative view of the situation on this market, will this be interesting for ASR community?

I have no idea how many folks will like to participate in such endeavors but there will be some that would like to experiment with the idea and/or software out of curiosity.
Most of them will already have commented in your threads.
Concentrating on portable stuff like DAPs, phones, tablets is mostly outside of ASR scope so that may be of interest to some people.
The limitations of said devices may influence the metrics in a non realistic way.
Also a device may test well yet have way too little output or have quite different output under certain load conditions.
So even with great metrics it might be under performing on some headphones that need a bit more than the DUT can provide.
Maybe include power ratings under various loads as well for instance.

There are certainly people willing to help here but you may need to come off from the high horse and take comments (especially by specialistst, which I am not) at heart.

I, as well as other technical folks, are quite happy with the current set of measurements and can make educated guesses on its performance based on those. When you ask 10 of them not all will come up with the exact same answer. This would most likely be based on personal experiences of those individuals.
To get a little merit for your efforts you will have to measure some of the devices that have been measured here and come up with metrics they can relate to in order to get them on your bus.

I am sure there is a market for metrics that are easy to understand and are confirmed by others to be really good.
Those that don't understand a set of measurements (fully) will welcome a simple plot showing fidelity.
To get it accepted by the more technical oriented folks at ASR you will need to make a few steps yet.
 
Last edited:

pma

Major Contributor
Joined
Feb 23, 2019
Messages
4,592
Likes
10,728
Location
Prague
Agreed. My problem is that I am absolutely not interested in portable DAP devices, small monitors regardless active/passive, and other cheapest stuff. If the method would say anything about adult hifi, then fine. However, I am sceptical about any single number approach to "sound quality".
 
OP
S

Serge Smirnoff

Active Member
Joined
Dec 7, 2019
Messages
240
Likes
136
Concentrating on portable stuff like DAPs, phones, tablets is mostly outside of ASR scope so that may be of interest to some people.
The limitations of said devices may influence the metrics in a non realistic way.
Also a device may test well yet have way too little output or have quite different output under certain load conditions.
So even with great metrics it might be under performing on some headphones that need a bit more than the DUT can provide.
Maybe include power ratings under various loads as well for instance.
I like DAPs as DUTs for the following reasons:
- their df measurements do not require expensive analogue input interface and ADC
- they are popular, which is important from "educational" point of view as results of df measurements can be presented in a very attractive and understandable way
- it is possible to measure almost all current market of DAPs and try to influence it, estimate the results of such influence
- many DAPs already tested by me (old ones) and by csglinux from headfi forum (modern ones); so, the procedure is well polished now and there is good understanding of what can be expected.
Otherwise, it does not matter, other devices are ok as well.

Measurement conditions can be discussed. Output level I used is max. level allowed for DAPs by EU standard (hearing loss ...), load now is resistive but can be complex (real headphones with Olive target response or, even better, an electric equivalent of such phones).
 
OP
S

Serge Smirnoff

Active Member
Joined
Dec 7, 2019
Messages
240
Likes
136
There are certainly people willing to help here but you may need to come off from the high horse and take comments (especially by specialistst, which I am not) at heart.
:) Yes, I know. I don't like to use that horse, only in situations when other part come to me on the horse and try to teach me from )). If you look at this thread and the one about dithering you will notice the difference in the tone of discussion. So, when I detected that kind of "teaching" I thought it is a good reason to start that "little war", which could help to gain more attention to my ideas. This happened mostly by chance and I'm not sure of its effectiveness. I'm a peaceful man )).
 

pkane

Master Contributor
Forum Donor
Joined
Aug 18, 2017
Messages
5,632
Likes
10,207
Location
North-East
OP
S

Serge Smirnoff

Active Member
Joined
Dec 7, 2019
Messages
240
Likes
136
However, I am sceptical about any single number approach to "sound quality".
As almost all of us. A 'single number' approach could work only when this number is critically low. That is why I propose not just measure/assess "distortions" but to force manufacturers to make it that low. In order to stop endless speculations about their audibility. I'm pretty sure they can not convolve to any agreed numbers.
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
That is why I propose not just measure/assess "distortions" but to force manufacturers to make it that low.

That.... will never, ever happen.

Those trying to teach you about dither where forced to react more strongly and I fully agree with their (multiple persons) viewpoints in this.
Dither is not just adding (extremely low) noise levels. Any noise in recordings is many many times higher and is reproduced more accurately in 16 bit because of dither. In 24 bits the added noise level is even far below self noise of any electronics.
 

SIY

Grand Contributor
Technical Expert
Joined
Apr 6, 2018
Messages
10,386
Likes
24,752
Location
Alfred, NY
If you're trying to solve nonexistent problems, you can enjoy the endless quest.

Current measurement regimes can completely capture audible issues in electronics.
 
OP
S

Serge Smirnoff

Active Member
Joined
Dec 7, 2019
Messages
240
Likes
136
Those trying to teach you about dither where forced to react more strongly and I fully agree with their (multiple persons) viewpoints in this.
Dither is not just adding (extremely low) noise levels. Any noise in recordings is many many times higher and is reproduced more accurately in 16 bit because of dither. In 24 bits the added noise level is even far below self noise of any electronics.
The discussion on dithering is about the question - is it pure mathematical process helping to preserve information during quantization of an audio signal and thus it is universally helpful operation for any cases of quantization; or it is a psychoachoustic operation, which has sense only in regards to human hearing and in case of other non-audio signals can lead to loss of information.

The problem with dithering is that it plays two different roles during quantization for signals below LSB and above LSB. For below-LSB signals it definitely helps to preserve them in company with some noise. I agreed that in this case dithering can be considered as pure mathematical operation and can be helpful for quantization of signals of any nature, not only audio signals (radar systems as an example).

But during quantization of the above-LSB signal the role of dithering is different. Now step-by-step. We have an original non-quantized signal. Adding noise to it (of any type and level) always results in degradation of its waveform and loss of information (increase of entropy). Subsequent quantization of the signal further degrades the signal and can not cancel that loss from dithering. This degradation can be easily measured by correlation coefficient, what I showed yesterday in that thread (and will show further with xr100 examples that he sent). So, dithering of above-LSB signals (with or without further quantization) degrades them, mathematically.

In many technical systems the dithering before quantization is harmful operation (financial markets as an example). Audio in this sense is lucky. Degradation of an audio signal by means of dithering before quantization results in - !!! - improvement of its perception. The reason is simple and well described in wiki:

Quantization yields error. If that error is correlated to the signal, the result is potentially cyclical or predictable. In some fields, especially where the receptor is sensitive to such artifacts, cyclical errors yield undesirable artifacts. In these fields introducing dither converts the error to random noise. The field of audio is a primary example of this. The human ear functions much like a Fourier transform, wherein it hears individual frequencies. The ear is therefore very sensitive to distortion, or additional frequency content, but far less sensitive to additional random noise at all frequencies such as found in a dithered signal.

In other words, dithering of above-LSB signals has all characteristics of psychoacoustic processing - it reduces accuracy of a signal but increases accuracy of its perception by ear. Much like psychoacoustic encoding, which degrades the signal but makes this degradation unnoticeable for ear.

The root of misunderstanding is in that word "information". If one use it in the Shannon's sense (as a signal wich carries the message, sintactic level of information in semiotics) then the statement: dithering preserves the information of above-LSB signals during quantization is false. If the term information is used as a message intended for perception by human auditory system, (semantic level of information in semiotics) then the above statement is true. In our discussion we use information in Shannon's sense, because we research mathematical aspects of its processing, trying to decide whether dithering is math operation or not.

In a nutshell. In audio the dithering before quantization is twofold - it is formal/mathematical operation for below-LSB signals and psychoacoustic operation for above-LSB signals.

In order to explain this I climbed to my high horse as nobody listens otherwise )).
 

SIY

Grand Contributor
Technical Expert
Joined
Apr 6, 2018
Messages
10,386
Likes
24,752
Location
Alfred, NY
In a nutshell. In audio the dithering before quantization is twofold - it is formal/mathematical operation for below-LSB signals and psychoacoustic operation for above-LSB signals.

In order to explain this I climbed to my high horse as nobody listens otherwise )).

Despite the wordiness... no. Not even close. You still fundamentally do not understand information, dither, and the basics of quantization. Substituting goalposts on roller skates and redefining terms on the fly to sound knowledgeable do not transform lack of understanding into understanding.
 
OP
S

Serge Smirnoff

Active Member
Joined
Dec 7, 2019
Messages
240
Likes
136
Current measurement regimes can completely capture audible issues in electronics.
And this is great. There will be something to compare results of df-measurements with.
Despite the wordiness... no. Not even close. You still fundamentally do not understand information, dither, and the basics of quantization. Substituting goalposts on roller skates and redefining terms on the fly to sound knowledgeable do not transform lack of understanding into understanding.
You could just say that I'm ignorant. It's shorter. Otherwise, please, try to elaborate what you consider as fundamentally wrong in my reasoning.
 

SIY

Grand Contributor
Technical Expert
Joined
Apr 6, 2018
Messages
10,386
Likes
24,752
Location
Alfred, NY
As Scott politely said, your premises are incorrect, so any “reasoning” from those premises is incorrect. It helps to understand the fundamentals of the area in which you’re trying to do work, and multiple people who have a solid background in this field have already tried, fruitlessly, to persuade you to gain that knowledge before engaging in trying to extend it. Start by actually absorbing basic literature on sampling and information theory, the stuff with actual equations and proofs rather than fashionable terms like “semiotics” and “syntactics.”

I confess that I have limited patience about trying to nail Jello to a wall.
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
In order to explain this I climbed to my high horse as nobody listens otherwise )).

What you claim does not make sense and your explanations are flawed to the core. That's why nobody listens. They read it, see where the flaws in reasoning are. Realize its nonsense. Try to explain it. Explain it again. Others make remarks and show how it works. Shouting louder won't help here.

If you have some time watch this video by Ian Shepperd

But do you want this to happen?

No I don't think this needs to happen at all, there is already enough choice.
I don't care if others sell snake oil, illusions, crap or wonderful engineering.
It's fine if something is exposed or confirmed but to me this does not make any difference.

Everyone is free to manufacture or buy whatever they want using whatever method they like (subjective, objective, combinations of it, using tarot cards, a pendulum or have someone you think knows what's best to pick it for them).
I sure ain't gonna tell manufacturers what to make nor demand it.
It is not up to me, nor do I expect anyone to listen to me or be forced to do something I would demand.
 
Last edited:

xr100

Addicted to Fun and Learning
Joined
Jan 6, 2020
Messages
518
Likes
237
Location
London, UK
This degradation can be easily measured by correlation coefficient, what I showed yesterday in that thread

OK, let's try this another way.

Sample rate: 192kHz. 1kHz sine at ~-40dBr... (call that the "message")

Add 20kHz square wave at ~-3dBr...


"Reconstructed" waveform:
1581280580763.png


And the "reconstructed" waveform of the 20kHz square wave ONLY:

1581280714426.png


Hmm, looks identical...

Spectrogram, however:

1581280754349.png



So, if the receiver of the signal "understands" the spectral magnitude ONLY, and further disregards everything above 20kHz, then the "message" is recoverable:

1581281051208.png



Now, the same sine wave (1kHz @~-40dBr) but with a 100Hz sawtooth wave @~-3dBr...

"Reconstructed" waveform:

1581281163556.png


Compare to the 100Hz sawtooth wave alone:

1581281327925.png



The 1kHz sine almost -40dB lower has affected the shape of the waveform. However...

Spectrogram:

1581281412802.png


If, once again the receiver of the signal "understands" the spectral magnitude ONLY, how is the "message" going to be recovered?

This, BTW, is a spectacularly basic example compared to, say, spread spectrum communication systems.

In many technical systems the dithering before quantization is harmful operation (financial markets as an example).

What is the "signal" in a financial market?

If we take share price, for example, then the "signal" may not even exist in the price--"insider trading," for example.

In attempting to recover the "price signal" as information so that shares can (hopefully) be bought/sold at minimum/maximum regions (whether relatively local for "scalping" or short-term trading, or longer-term)--small magnitude, high frequency fluctuations in price are "NOISE."

(Perhaps looking at the patterns of the "NOISE" that would be removed by, say, a "moving average" window might allow the derivation of potential information as an input to decisions, but... that's still an attempt to extract the "meaningful" part out of a NOISY signal anyway.)
 
Last edited:
OP
S

Serge Smirnoff

Active Member
Joined
Dec 7, 2019
Messages
240
Likes
136
What is the "signal" in a financial market?

If we take share price, for example, then the "signal" may not even exist in the price--"insider trading," for example.

In attempting to recover the "price signal" as information so that shares can (hopefully) be bought/sold at minimum/maximum regions (whether relatively local for "scalping" or short-term trading, or longer-term)--small magnitude, high frequency fluctuations in price are "NOISE."

(Perhaps looking at the patterns of the "NOISE" that would be removed by, say, a "moving average" window might allow the derivation of potential information as an input to decisions, but... that's still an attempt to extract the "meaningful" part out of a NOISY signal anyway.)
An audio signal (in electrical or digital form) reproduces mechanical movement of some particles in a medium with time (molecules of air for example). Financial signal (or curve) reproduces movement of price on a market with time. This movement is determined by buy/sell orders placed by traders. Both signals convey some valuable information.

Dither is actually used for financial signals. It is a recent innovation. It is used not for improving quantization, it is used alone in the form of temporal dither (or jitter). Buy/sell orders placed by traders are delayed by some small and random value. This results in small degradation of price discovery on the market (spoiling financial signal/curve) but effectively prevents high frequency trading, which is considered by many as harmful for markets (such dithering is also mentioned in the wiki article about dither). In this case the intentionally introduced error/dither can be considered as preserving financial information because it leads to better price discovery on the market (otherwise spoiled by HF traders) and the same time it can be considered as degrading financial information because such errors make impossible HF trading (not all agree that HFT is bad for markets).

There are another areas where dithering is used alone (not in combination with quantization; I mentioned some of them in the earlier post). In all such cases it introduces errors and hence reduces mathematical/Shannon's information. In other words, the dithering by itself, as mathematical operation, always degrades the signal and reduces the information. But as in the case above the degradation of math. information can result in increase of semantic information, which is recipient-dependent, not mathematical. If recipient of financial information is HF trader then dithering of financial curve is assessed as loss of information. If recipients of financial information are many others participants of the market then dithering is assessed as increase of information because it results in more fair and stable market price (more correct/meaningful curve).

One and the same math. operation can result increase or decrease of semantic information. It depends on the recipient of the information. When one low-passes some dithered+quantized signal above 20kHz, he increases information if the signal is intended for perception by ear. But if the signal is intended for analysis of residual noise after quantization then he reduces the information. So, it depends.

My POV on dithering+quantization in audio (taking into account the context above) I stated in the post #171.
 

xr100

Addicted to Fun and Learning
Joined
Jan 6, 2020
Messages
518
Likes
237
Location
London, UK
Dither is actually used for financial signals. It is a recent innovation. It is used not for improving quantization, it is used alone in the form of temporal dither (or jitter). Buy/sell orders placed by traders are delayed by some small and random value. This results in small degradation of price discovery on the market (spoiling financial signal/curve) but effectively prevents high frequency trading, which is considered by many as harmful for markets (such dithering is also mentioned in the wiki article about dither). In this case the intentionally introduced error/dither can be considered as preserving financial information because it leads to better price discovery on the market (otherwise spoiled by HF traders) and the same time it can be considered as degrading financial information because such errors make impossible HF trading (not all agree that HFT is bad for markets).

Thanks for your reply.

Interesting stuff. A problem is that there is no defined original "SIGNAL" to inform buy/sell orders; rather it is an input to a GUESS at when to do so. The example I gave of "insider trading" is sufficient to demonstrate this, i.e. that part of the "signal" is not known to outside traders and thus prior to the insider trading occurring is not at all reflected in the "price signal."

An entirely different situation to audio where the source signal is absolutely defined.

In other words, the dithering by itself, as mathematical operation, always degrades the signal and reduces the information.

The received signal is subjected to PROCESSING to recover the information. Communication systems attempt to maximise the amount of information that can be reliably transmitted within a noisy channel of given bandwidth; see spread-spectrum communication systems, etc.

Dithering, again, "linearises" the system, removing quantization distortion and allows signals below LSB to be retained. Thus the information is not "reduced." If the spectral characteristics of the added noise are such that they reside outside of the human auditory range, then the original "information" is effectively recovered with nothing added or taken away.

(Not withstanding system non-linearities resulting in frequency components within the audible range to occur.)

And the above example is an extremely basic one, considering psychoacoustics; critical bands, masking, and so on.

But if the signal is intended for analysis of residual noise after quantization then he reduces the information.

Unweighted SNR is not going to say anything about whether the "information" has been reduced or not.
 
Last edited:
OP
S

Serge Smirnoff

Active Member
Joined
Dec 7, 2019
Messages
240
Likes
136
Thanks for your reply.

Interesting stuff. A problem is that there is no defined original "SIGNAL" to inform buy/sell orders; rather it is an input to a GUESS at when to do so. The example I gave of "insider trading" is sufficient to demonstrate this, i.e. that part of the "signal" is not known to outside traders and thus prior to the insider trading occurring is not at all reflected in the "price signal."

An entirely different situation to audio where the source signal is absolutely defined.



The received signal is subjected to PROCESSING to recover the information. Communication systems attempt to maximise the amount of information that can be reliably transmitted within a noisy channel of given bandwidth; see spread-spectrum communication systems, etc.

Dithering, again, "linearises" the system, removing quantization distortion and allows signals below LSB to be retained. Thus the information is not "reduced." If the spectral characteristics of the added noise are such that they reside outside of the human auditory range, then the original "information" is effectively recovered with nothing added or taken away.

(Not withstanding system non-linearities resulting in frequency components within the audible range to occur.)

And the above example is an extremely basic one, considering psychoacoustics; critical bands, masking, and so on.



Unweighted SNR is not going to say anything about whether the "information" has been reduced or not.
It looks like we have confusion with the terms. For you, if I understand correctly, a “signal” is some information useful for recipient. Consequently, price records are, at least, bad signal as it is not quite useful for traders because of, for example, insider trading. For me a signal is just a carrier of information. So, price records is a signal which is just “follows” the price movement on the market (like temperature records register temperature of a patient). Traders can use this signal for their decisions but can use another info/intuition/insides.

Accordingly an audio signal can be interpreted as an information useful for recipient and as only a carrier of information. Can we call white noise an audio signal? In my interpretation it is a valid audio signal. What information it carries is another question, it depends on the recipient and context.
 
Last edited:
Top Bottom