Article: My Lying Ears

[jabbr]

Sample and hold adds distortion, not information. A perfect interpolation filter adds nothing above the Nyquist frequency of the original and alters nothing below it. Such filters are not practically realisable, so any real-world filter will introduce some amount of distortion, and the aim is to minimise this, especially in the audible band.

 

I basically agree with this but think that people are arguing at cross purposes. It all comes down to what we mean by "information" as opposed to "data points" or "bits". In the way of defining "information" in the way I define information, no filter *ever* adds information to a system regardless of whether it interpolates new data points or causes distortion, or improves the sound etc.

 

The "problem" with this view is in the definition of "system". So if I start with a file F that I obtain, then I define F as the source of information and any transformation (filter) as T(F).

 

Now if I use T(F) to create a new file G and then distribute that (eg "remaster") then for G there is essentially a new set of information, and in that case the remastering can be said to change the information... so this is essentially a matter of perspective.

[Paul R]

I basically agree with this but think that people are arguing at cross purposes. It all comes down to what we mean by "information" as opposed to "data points" or "bits". In the way of defining "information" in the way I define information, no filter *ever* adds information to a system regardless of whether it interpolates new data points or causes distortion, or improves the sound etc.

 

The "problem" with this view is in the definition of "system". So if I start with a file F that I obtain, then I define F as the source of information and any transformation (filter) as T(F).

 

Now if I use T(F) to create a new file G and then distribute that (eg "remaster") then for G there is essentially a new set of information, and in that case the remastering can be said to change the information... so this is essentially a matter of perspective.

 

If you mean no filter ever exactly restores lost information, then I would agree with you. However, the information interpolated by an appropriate filter is most certainly used by most DACs- whether that is a good or bad thing, or whether one filter is better than another, or better than not filtering at all, is not the question.

 

On the other paw- filters almost always add information to a system. Pretty much by any definition.

 

Actually the Wilipedia entry is for once, pretty good on this subject. Normal Warning about wiki sources... https://en.m.wikipedia.org/wiki/Interpolation

[Jud]

I basically agree with this but think that people are arguing at cross purposes. It all comes down to what we mean by "information" as opposed to "data points" or "bits". In the way of defining "information" in the way I define information, no filter *ever* adds information to a system regardless of whether it interpolates new data points or causes distortion, or improves the sound etc.

 

The "problem" with this view is in the definition of "system". So if I start with a file F that I obtain, then I define F as the source of information and any transformation (filter) as T(F).

 

Now if I use T(F) to create a new file G and then distribute that (eg "remaster") then for G there is essentially a new set of information, and in that case the remastering can be said to change the information... so this is essentially a matter of perspective.

 

While I empathize with you and xyzzy1 regarding what you would *like* to call information, I despair of ever being able to communicate effectively if everyone gets to use his or her own special meaning of a word depending on point of view. So let's acknowledge that according to the scientific definition from information theory, in part developed by the very same Shannon who is one of the people responsible for the Sampling Theorem, the interpolation of data points in fact *does* add information.

 

OK, fine, what sort of information? This is one of the places where digital audio is counter-intuitive (not many of those, are there? ). Under the Sampling Theorem, if I can over-simplify using a metaphor, as soon as you've got 3 points on a curve whose rate of change is appropriately limited, you've located the curve along its entire length. No more points are needed to more exactly locate the curve. What the greater number of sample points is useful for is letting you throw the necessary number of them away with a gentler filter. Yep, those extra points are essentially cannon fodder, but they allow the result to sound better than if you didn't have them to throw away.

[mansr]

While I empathize with you and xyzzy1 regarding what you would *like* to call information, I despair of ever being able to communicate effectively if everyone gets to use his or her own special meaning of a word depending on point of view. So let's acknowledge that according to the scientific definition from information theory, in part developed by the very same Shannon who is one of the people responsible for the Sampling Theorem, the interpolation of data points in fact *does* add information.

 

Forget about information, seeing as the word appears to be contentious. Interpolating a signal does not reveal any additional detail about the source of the signal. While they added bits may be useful, they are not meaningful.

[Jud]

Forget about information, seeing as the word appears to be contentious. Interpolating a signal does not reveal any additional detail about the source of the signal.

 

OK

 

While they added bits may be useful, they are not meaningful.

 

"Meaningful" can be seen as getting to "informative" or "information" through the back door, so perhaps we can say the added bits are useful, but not for the purpose of providing additional detail about the (analog) source of the (digital) signal.

[Paul R]

OK

 

 

 

"Meaningful" can be seen as getting to "informative" or "information" through the back door, so perhaps we can say the added bits are useful, but not for the purpose of providing additional detail about the (analog) source of the (digital) signal.

 

+1

[mansr]

"Meaningful" can be seen as getting to "informative" or "information" through the back door, so perhaps we can say the added bits are useful, but not for the purpose of providing additional detail about the (analog) source of the (digital) signal.

 

I think we might be in agreement, more or less.

[jhwalker]

OK

 

 

 

"Meaningful" can be seen as getting to "informative" or "information" through the back door, so perhaps we can say the added bits are useful, but not for the purpose of providing additional detail about the (analog) source of the (digital) signal.

 

Exactly.

[Sal1950]

When you said “...if you can't hear the differences they claim, either your system sucks or your ears do.” you were responding to Paul who was responding to xyzzy1. What I was saying in a very polite way was: "If one person doesn’t try to tell another person what they are allowed and not allowed to hear, then that person won’t be able to say their system or ears suck." So, basically common courtesy.

 

That is why I said: “Personally I have no problem with people being able to hear something I can't. My very affordable audio/video system surely is not as revealing as high end equipment I can't afford.” Because I don’t attack other peoples listening experiences, no one says I can’t hear something because my system or ears suck. I just accept that I don’t have the world’s best ears and or the best equipment, thus many people will hear things I can’t. Do you understand now how it relates to what you said?

 

If someone is making ridiculous claims about what they hear or see or anything else it's not a matter of them being "allowed" to hear it but about someone else being "allowed" to say BS. If you told me you saw a swarm of aliens over your house last night I'd say prove it. Did any one else see it? Do you have pictures? You can't make a response by telling me "well I saw them and if you didn't there's something wrong with your eyes or your glasses". For many reasons people have a responsibility to challenge peoples public claims and those making the claims have a responsibility to be able to back up their claims.

Thanks to Mark Waldrep of AIX Records posting of the falsified AudioQuest Cable video. We now know how badly the public was lied to and they are now protected from being hoodwinked by these deliberate lies from a bunch of snake oil peddlers. If this was about something considered really important, like medicine, someone would have legal charges against them.

Audio is a lala land in separation from the rest of the universe where anyone can claim anything and not be expected to have any backup except their word. Good thing the legal system doesn't "usually" work that way.

[Sal1950]

OMG, LOL, What timing!

Just posted by Gene at Audioholics

This is just too funny and goes directly to what you

are trying to tell me Teresa. Claims HAVE to be questioned and backed up.

ROTFLMAO

[video=youtube_share;BgDB86tkth8]

Gene DellaSala

President, Audioholics

[jabbr]

While I empathize with you and xyzzy1 regarding what you would *like* to call information, I despair of ever being able to communicate effectively if everyone gets to use his or her own special meaning of a word depending on point of view. So let's acknowledge that according to the scientific definition from information theory, in part developed by the very same Shannon who is one of the people responsible for the Sampling Theorem, the interpolation of data points in fact *does* add information.

 

"samples"/"data points" is not the same as "information" according to Shannon-Nyquist. Don't despair:)

 

I am defining the original data file as the information set F. I am defining a filter as a transform or function T. So the transformed data F' = T(F). Since F' is implied by F it does not contain more information.

 

No over or under simplification , no metaphors -- although I'm not sure what to make of a metaphor of a mathematical theorem.

[Jud]

"samples"/"data points" is not the same as "information" according to Shannon-Nyquist. Don't despair:)

 

I am defining the original data file as the information set F. I am defining a filter as a transform or function T. So the transformed data F' = T(F). Since F' is implied by F it does not contain more information.

 

No over or under simplification , no metaphors -- although I'm not sure what to make of a metaphor of a mathematical theorem.

 

OK, well just in the friendliest possible way to help smite the deceased equine here, and taking this slowly, since it is the only way I have a prayer of following it:

 

- Regarding "information," I'm talking Kolmogorov-Shannon (information theory) more than Shannon-Nyquist (sampling theorem).

 

- The filter, or function T, is the second of two operations. Interpolation is the first. The first operation does indeed create information. In fact it is relatively trivial to have that information match virtually point for point the information everyone apparently agrees would be "new," that would be obtained from recording the analog signal at a higher sample rate.

 

- The filter then gets rid of a bunch of the interpolated samples, just as it would get rid of a bunch of the actual samples if the recording were made at a higher sample rate. The greater number of interpolated or actual samples, the easier it is to have the filter do a better job of not adding even more information in the form of ringing or aliasing distortions.

 

- So why even bother with hi res? Why not just oversample? Because the listening end of things is part of a chain that stretches back to the recording end. There the ADC runs the signal through decimation filtering, the opposite side of the coin from the DAC's interpolation and filtering. Decimation filtering is subject to ringing, aliasing and group delay as well, perhaps even more critically since it may be more difficult to keep from audibly harming the signal while doing decimation than when interpolating. Keeping the signal at higher resolution avoids imperfect conversion steps (decimation, interpolation, filtering) at both ends of the chain.

[jabbr]

- The filter, or function T, is the second of two operations. Interpolation is the first. The first operation does indeed create information. In fact it is relatively trivial to have that information match virtually point for point the information everyone apparently agrees would be "new," that would be obtained from recording the analog signal at a higher sample rate.

 

- The filter then gets rid of a bunch of the interpolated samples, just as it would get rid of a bunch of the actual samples if the recording were made at a higher sample rate. The greater number of interpolated or actual samples, the easier it is to have the filter do a better job of not adding even more information in the form of ringing or aliasing distortions.

 

It is the theory that determines the necessary sampling rate to represent the information. Interpolation or oversampling does not represent new information, nor is the noise introduced in the transmission process considered information.

[jabbr]

While I empathize with you and xyzzy1 regarding what you would *like* to call information, I despair of ever being able to communicate effectively if everyone gets to use his or her own special meaning of a word depending on point of view. So let's acknowledge that according to the scientific definition from information theory, in part developed by the very same Shannon who is one of the people responsible for the Sampling Theorem, the interpolation of data points in fact *does* add information.

 

OK, fine, what sort of information? This is one of the places where digital audio is counter-intuitive (not many of those, are there? ). Under the Sampling Theorem, if....

 

It is well known that in natural language the meaning of a term depends on its context. In science we reference specific published articles to clarify meaning. I've had the opportunity to reread Shannon 1948 and this paper handles both the analog and digital information sources explicitly..

 

In audio the information source is the analog audio signal which may be encoded as PCM in the transmitter.

 

So I am defining information according to Shannon. The two points of view that I discussed earlier also envisioned treating the digital data file itself as the information source but nonetheless and according to Shannon my point stands.

[Teresa]

OMG, LOL, What timing! Just posted by Gene at Audioholics

This is just too funny and goes directly to what you are trying to tell me Teresa. Claims HAVE to be questioned and backed up. ROTFLMAO

[video=youtube_share;BgDB86tkth8]

Gene DellaSala

President, Audioholics

 

Thanks. Yes, that was funny.

 

You still don’t understand what I was saying in my last two posts. I am not defending AudioQuest but explaining how to avoid posters saying “if you can't hear the differences they claim, either your system sucks or your ears do.” If you don’t mind people telling you your system or your ears suck, go ahead and bait them. I would never do that as I don’t want the anguish and besides it’s not my job. I am more interested in enjoying music.

 

There is a difference between false advertising and someone hearing something I don't.

 

Like I said before “Personally I have no problem with people being able to hear something I can't. My very affordable audio/video system surely is not as revealing as high end equipment I can't afford.”

 

I am against false advertising but I don't try to tell people what they are allowed and not allowed to hear, it would be nice if everyone else did the same.

[Don Hills]

... - The filter, or function T, is the second of two operations. Interpolation is the first. The first operation does indeed create information. In fact it is relatively trivial to have that information match virtually point for point the information everyone apparently agrees would be "new," that would be obtained from recording the analog signal at a higher sample rate.

 

Not quite. You would only have more information if you had recorded the analogue signal at a higher sample rate. Upsampling doesn't create new information, as you say yourself the additional samples match what is already "there" between the original samples and are thus redundant. One upsampling technique is simply to add zeroes between the existing samples. When you filter it, you simply don't bother to perform the filter calculations for the inserted zero samples. The net result is the same and you save on memory and processor cycles.

[Paul R]

Not quite. You would only have more information if you had recorded the analogue signal at a higher sample rate. Upsampling doesn't create new information, as you say yourself the additional samples match what is already "there" between the original samples and are thus redundant. One upsampling technique is simply to add zeroes between the existing samples. When you filter it, you simply don't bother to perform the filter calculations for the inserted zero samples. The net result is the same and you save on memory and processor cycles.

 

Yes, that is certainly true wih sample and hold type processing. Nothing new is created or restored.

 

But interpolation certainly can add new information, interpolating between the dat points to create new data that is meaningful information. Replacing information that was lost during the sampling operations. (*Not exactly, but close enough as to make no difference in any practical sense.)

 

Note- that is not related to Shannon/Nyquist other than to say it is a way to calculate the missing data from the samples that exist. Nor am I arguing that the extra information, which absolutely is meaningful in this case, makes the reconstructed analog output sound better or worse. It certainly is involved with changing that output though. And - this type of interpolative process is not limited to the audio world either.

 

Always amazes me that people get confused about this. How many people listen to redbook material with an absolutely non-upsampling DAC? How many people do not hear a difference when Redbook is upsampled to 24/192k? Or even 96k?

 

The extra information absolutely makes a difference in the reconstructed analog signal. Mathematically? It is not necessarily required. Practically? Almost always necessary.

[jabbr]

Yes, that is certainly true wih sample and hold type processing. Nothing new is created or restored.

 

But interpolation certainly can add new information, interpolating between the dat points to create new data that is meaningful information. Replacing information that was lost during the sampling operations. (*Not exactly, but close enough as to make no difference in any practical sense.)

 

Note- that is not related to Shannon/Nyquist other than to say it is a way to calculate the missing data from the samples that exist. Nor am I arguing that the extra information, which absolutely is meaningful in this case, makes the reconstructed analog output sound better or worse. It certainly is involved with changing that output though. And - this type of interpolative process is not limited to the audio world either.

 

Always amazes me that people get confused about this. How many people listen to redbook material with an absolutely non-upsampling DAC? How many people do not hear a difference when Redbook is upsampled to 24/192k? Or even 96k?

 

The extra information absolutely makes a difference in the reconstructed analog signal. Mathematically? It is not necessarily required. Practically? Almost always necessary.

 

I understand what you are saying and agree that upsampling of Redbook, even better conversion to high bitrate DSD gives a more pleasing reconstruction of the original analog signal -- at least according to my ears.

 

And I agree that this is a very practical implementation detail for the digital audio playback chain. I think the theory assumes that the system is optimized in this way.

[Paul R]

I understand what you are saying and agree that upsampling of Redbook, even better conversion to high bitrate DSD gives a more pleasing reconstruction of the original analog signal -- at least according to my ears.

 

And I agree that this is a very practical implementation detail for the digital audio playback chain. I think the theory assumes that the system is optimized in this way.

 

Rock my world there - we pretty much absolutely agree.

[jabbr]

- Regarding "information," I'm talking Kolmogorov-Shannon (information theory) more than Shannon-Nyquist (sampling theorem).

 

- The filter, or function T, is the second of two operations. Interpolation is the first. The first operation does indeed create information. In fact it is relatively trivial to have that information match virtually point for point the information everyone apparently agrees would be "new," that would be obtained from recording the analog signal at a higher sample rate.

 

Kolmogorov is an interesting approach. Algorithmic complexity defines the minimum string necessary to specify the transmission as an algorithm. Again, interpolation is an algorithm which is a part of the system, so representing F' = T(I(F)). Interpolation is clearly a function and as such does not "create" information.

 

Nor does recording an analog signal at a higher sampling rate ***unless*** the recording is done in order to capture higher bandwidth.

 

Several examples:

1) the analog signal is bandwidth limited to 11 khz

2) the analog signal is bandwidth limited to 22 khz

3) the analog signal is bandwidth limited to 44 khz

 

In the case of (1) and (2) recording at 88 khz does not add new information and these recordings can be appropriately compressed. In the case of (3) recording at 44 khz is not sufficient to represent the information.

 

To be very clear, recording at a higher than necessary sampling rate does not add new information above and beyond the information in the analog signal.

 

Here is a classic Kolmogorov example (and I'm going on here because it is interesting :):

 

What is the Kolmogorov representation of the following sequences:

 

A) "asdEtrxttgWQxtxxxttxxrr"

B) "GGGEbFFFDGGGEbFFFDGGGEbFFFD"

C) "Beethoven's Fifth"

 

[Jud]

Not quite. You would only have more information if you had recorded the analogue signal at a higher sample rate. Upsampling doesn't create new information, as you say yourself the additional samples match what is already "there" between the original samples and are thus redundant. One upsampling technique is simply to add zeroes between the existing samples. When you filter it, you simply don't bother to perform the filter calculations for the inserted zero samples. The net result is the same and you save on memory and processor cycles.

 

Kolmogorov is an interesting approach. Algorithmic complexity defines the minimum string necessary to specify the transmission as an algorithm. Again, interpolation is an algorithm which is a part of the system, so representing F' = T(I(F)). Interpolation is clearly a function and as such does not "create" information.

 

Nor does recording an analog signal at a higher sampling rate ***unless*** the recording is done in order to capture higher bandwidth.

 

Several examples:

1) the analog signal is bandwidth limited to 11 khz

2) the analog signal is bandwidth limited to 22 khz

3) the analog signal is bandwidth limited to 44 khz

 

In the case of (1) and (2) recording at 88 khz does not add new information and these recordings can be appropriately compressed. In the case of (3) recording at 44 khz is not sufficient to represent the information.

 

To be very clear, recording at a higher than necessary sampling rate does not add new information above and beyond the information in the analog signal.

 

Here is a classic Kolmogorov example (and I'm going on here because it is interesting :):

 

What is the Kolmogorov representation of the following sequences:

 

A) "asdEtrxttgWQxtxxxttxxrr"

B) "GGGEbFFFDGGGEbFFFDGGGEbFFFD"

C) "Beethoven's Fifth"

 

 

First, I apologize to everyone else for the OT diversion. Like jabbr and Don, I find this stuff fascinating, so I'll ask your indulgence for these few OT comments.

 

So here's my layperson's take, and you can tell me where I'm wrong so I learn something (which for me is the exciting part):

 

Working with a Shannon-Kolmogorov definition of "information" (the intersection of Shannon entropy and Kolmogorov complexity, see, e.g., http://homepages.cwi.nl/~paulv/papers/info.pdf), gives us something very much akin to what we would think of as compressibility. Two files that can be losslessly compressed to the same size have equivalent amounts of information/entropy/complexity.

 

Take a RedBook file and do the interpolation and filtering required to convert to 24/192. Is there a mathematical operation that will convert the 24/192 file back to the RedBook file losslessly? (Does closed form filtering make any difference here?) If not, to me this necessarily implies that the 24/192 file cannot be compressed to the size of the RedBook file, and therefore under the Shannon-Kolmogorov definition(s), the 24/192 file has more information. Thus information was added by the sample rate conversion operations of interpolation and filtering.

 

Through all of this, I am agreeing with just about everyone who's talked about it thus far, including my own previous comments, that this additional information will do nothing to further specify the analog signal.

[Paul R]

Take a RedBook file and do the interpolation and filtering required to convert to 24/192. Is there a mathematical operation that will convert the 24/192 file back to the RedBook file losslessly?

Only if you saved the equivalent of metadata during the interpolation, which is not usually done as far as I know.

 

(Does closed form filtering make any difference here?)

It does in targeting tracking/firing solutions. The greatest impact is on processor usage. Not sure about the impact on music, but I think Mike Morrow is working on just such a filter...

 

If not, to me this necessarily implies that the 24/192 file cannot be compressed to the size of the RedBook file, and therefore under the Shannon-Kolmogorov definition(s), the 24/192 file has more information. Thus information was added by the sample rate conversion operations of interpolation and filtering.

 

Two different things here - first, can the 24/192K be compressed to the side of a rebook file? Probably, but it would still have more information in it that the redbook file, so long as the compression is completely reversible. Second, can the 24/19k be used to precisely reconstruct the redbook file? Probably not.

 

Through all of this, I am agreeing with just about everyone who's talked about it thus far, including my own previous comments, that this additional information will do nothing to further specify the analog signal.

 

Not sure anyone was arguing with that... - the analog signal is the hard limit of information available in this case.

[jabbr]

So that we don't stay OT here, I'm forking further discussion to a new thread where we are free to discuss any and all aspects of information theory:)

[mansr]

Working with a Shannon-Kolmogorov definition of "information" (the intersection of Shannon entropy and Kolmogorov complexity, see, e.g., http://homepages.cwi.nl/~paulv/papers/info.pdf), gives us something very much akin to what we would think of as compressibility. Two files that can be losslessly compressed to the same size have equivalent amounts of information/entropy/complexity.

 

Agreed.

 

Take a RedBook file and do the interpolation and filtering required to convert to 24/192. Is there a mathematical operation that will convert the 24/192 file back to the RedBook file losslessly? (Does closed form filtering make any difference here?) If not, to me this necessarily implies that the 24/192 file cannot be compressed to the size of the RedBook file, and therefore under the Shannon-Kolmogorov definition(s), the 24/192 file has more information. Thus information was added by the sample rate conversion operations of interpolation and filtering.

 

Not all information/entropy is equal. If I take a signal (e.g. a music recording) and add white noise, the result will by your definition contain more information since it can't be compressed as much. Does this mean the original+noise sounds better or is otherwise preferable to the original? Hardly. Any information added by a filter (interpolation or other type) is noise/distortion caused by finite precision implementations, i.e. something undesirable but unavoidable.

 

A DAC consists, somewhat simplified, of the following stages (the first two optional):

 

1. Upsampling/interpolation

2. Sigma-delta modulation

3. D/A conversion

4. Analogue low-pass filter

 

Each of these stages introduces some amount of noise and distortion. The goal is to pick the parameters such that the total end-to-end performance is maximised (within cost/power/whatever constraints), and adding a little noise in one stage can let us reduce it a lot in another. For instance, the D/A stage is more accurate at a lower bit depth, enough so that it more than compensates for the (audio band) noise added by the sigma-delta modulation. Likewise, while a digital interpolation filter adds a little noise, it allows for an analogue filter with a higher cutoff frequency and thus less distortion in the audio band. The sigma-delta modulation also performs better the higher the sample rate.

 

When sound quality is improved by software upsampling or DSD conversion, it is because we are replacing the first stages of the DAC with a higher quality implementation (at tremendous (relative) expense in cost and power), not because it somehow adds information to the signal.

 

Second, can the 24/19k be used to precisely reconstruct the redbook file? Probably not.

 

If the original redbook file can't be precisely reconstructed, information has been lost, not added. This loss is caused by various inaccuracies in practical implementations. A perfect (infinite precision, infinite time) interpolation is reversible. For an integer multiple interpolation, the reversal is trivial: drop all but every N samples.

[Sal1950]

Thanks. Yes, that was funny.

There is a difference between false advertising and someone hearing something I don't.

 

Like I said before “Personally I have no problem with people being able to hear something I can't. My very affordable audio/video system surely is not as revealing as high end equipment I can't afford.”

 

I am against false advertising but I don't try to tell people what they are allowed and not allowed to hear, it would be nice if everyone else did the same.

 

No there is no difference, if you post that you hear a improvement using cable X you are in effect posting an advertisement for that cable, a review. In this free country you are "allowed" to post anything you like, just be prepared for some objective person to ask you to back up that statement, we are "allowed" to do that also, and have a responsibility to our fellow readers to do so. I don't mind being told a million that my ears or system suck, it give me something to ridicule in my response. It's just that it's such a lame response compared to posting the result of a ABX-DBT or some technical measurements, any repeatable, verifiable test to back up the claim, it's become more than laughable. The people using that BS retort should be ashamed by now.

Tersa, if you go to the theater and holler FIRE, your allowed to do that, that's a good thing if there is a fire. But if you can't back that one up you'll find your butt in jail.

I don't get how you can't understand the simple concept that if you make statements you need to be able to back the up. You can't just go around making any claim you want and not expect someone to want the facts. A lesson that the maker of the falsified Audioquest HDMI is sadly learning. He, AQ, and in some effect a part of the subject community is a laughing stock.