Thoughts on the future of HiFi

[theaudiohobby]

We are not talking about speech here, we are talking music only. If you choose not to beleive what I have stated, that's your business. I am not here to convince anyone of anything, I gave up that battle years ago, and it will not start again with you.

Differentiating between music and speech for the purposes of telecommunications is a false distinction.

Who said anything about agitation? I said less relaxed. There is a difference.

Ok

Are you asking me to produce the paper? Hey, I know it exists, but I didn't purchase it, and I have no interest in doing so just to appease you. So if you think the evidence is not there, bask in your own ignorance, I don't care. Even if I did purchase it, I could not post it as it would be against AES rules.

I did not ask you post the paper, I asked you to state the name of the paper or any other pertinent paper for that matter. By definition, an accepted theory would be documented in multiple places.

Once again, that is your opinion, and you are welcomed to keep it. My claim is the fact they are not identical goes along way in explaining why each sounds different even when using the same microphones mixer, amps and speakers(and with zero processing).

Here you are simply wrong, I have posted a link to an academic digital theory primer to make the point. You previously said "Let's start with the waveform. Once a analog signal is encoded to 16bit resolution, it appears as a stair stepped waveform, unlike a recorded analog which is more rounded. A DAC see's that waveform, and attempts to transcode that waveform exactly as it is represented - a series of snapshots that represent a stair stepped pattern" . However the primer says

"Diagram 2a shows an acoustic sine wave with sample points at regular intervals in time from right to left on the x-axis. The sample values shown in Diagram 2b create the digital representation of the wave file shown in Diagram 2c. A low pass filter is used upon output to smooth the edges of the new digital wave, which represent added high frequencies, with the resultant wave shown in Diagram 2d."

In otherwords, the DAC produces a pretty decent analog waveform, not a stair stepped pattern, once anti-aliasing is applied, which is as one would expect given sampling theory.

I never said that what leaves the DAC is not a analog signal, my point is that its waveform is different from an all analog signal, and that can account to why our ears respond differently to each's output.

Well, as the primer clearly illustrates your original contention is simply wrong, the output of the DAC is not a stair-stepped waveform, after anti-aliasing, the resultant analog waveform is smooth and similar to (or in your words, looks like) the original waveform . and by extension your deduction about sound differences due to differences in waveform pattern is also invalid.

[Feanor]

...
Well, as the primer clearly illustrates your original contention is simply wrong, the output of the DAC is not a stair-stepped waveform, after anti-aliasing, the resultant analog waveform is smooth and similar to (or in your words, looks like) the original waveform . and by extension your deduction about sound differences due to differences in waveform pattern is also invalid.

To me it's all very confusing. As I understood the Nyquist Theorem, (granted I'll never understand it perfectly), a sound wave can in principle be perfectly reproduce up to 1/2 the sampling frequency. E.g. if the sampling rate is 44 kHz, then 22 kHz can be perfectly reproduced. But there are various demons which a technically unsophisticated person like myself can see affecting the source signal to the ADC => DAC result ...

• Equipment problem #1: incorrect measurement of the amplitude of the source signal, i.e. getting the bits right
• Equipment problem #2: getting the ADC and subsequently DAC timing right, i.e. sampling at precisely the right rate, i.e. preventing "jitter"
• Technical problem #1, 2, ... etc.: eliminating the spurious sound of samples about 1/2 the sampling frequency, i.e. how do you dump the frequencies above 22 kHz (in case of CD) which are noise? I suppose this might be dealt with more or less well. The techniques of anti-aliasing, dithering, and filtering are very obscure to me, Except I think dithering has to do with randomizing the digital noise -- signal above 22 kHz -- which would otherwise variy directly with the "good" signal below 22 kHz). Also, I understand that filtering can (or must?) affect the phase of the signal below the 1/2 sample rate frequency.
• Psycho-accoustic problem: do sounds above concious audibilty, say 20+ kHz, really have no effect on sound perceptions?

In any case it's obvious that the source sound, (analog), will include frequencies above 1/2 the sample rate, so will not be identical to the ADC => DAC's signal however prefectly the sound below 1/2 are reproduced.

I'd appreciate further clarification of these issues though I can't guarantee I'd understand what you're saying.

[poppachubby]

I'm with you Bill, it's alot to digest. One thing I am wondering about is oversampling.

My Magnavox has 4 times OS. Is less OS substantial in terms of arriving to a sound closer to NO oversampling? Or is it once you OS, regardless of times, the result is the same.

I have always made the naive assumption that less oversampling would be desirable, for someone who enjoys Non-OS.

I hope my question is clear enough.

[E-Stat]

Well, as the primer clearly illustrates your original contention is simply wrong, the output of the DAC is not a stair-stepped waveform, after anti-aliasing, the resultant analog waveform is smooth and similar to (or in your words, looks like) the original waveform

How can anyone possibly disagree with a cute little picture like that?

rw

[E-Stat]

Or is it once you OS, regardless of times, the result is the same.

The problem remains with the need to filter content beyond the converter's ability to sample. Any and ALL content not filtered will necessarily be perceived as 100% distortion. A Nyquist sampler (non OS) must perform the steep and absolute filtering before the A to D converter requiring complex filters. In this case, it is not necessarily the simpler solution. Over sampling allows multiple filters used at higher ranges that do not effect the audible band to the same degree. Red Book playback must necessarily trade some bandwidth for phase integrity. There is an Ayre player which allows you to choose which profile you want for a given recording. Which is why hi-rez is inherently superior - because simpler filters can be used at far higher frequencies leaving the audible band completely intact. It obviates the technical challenges that transcend theory which does not work in practice.

Sir T is among many recording engineers who have taken the feed directly from mics and compared the result stored in multiple formats and found Red Book lacking. That is where the rubber hits the road, not the presentation of simple graphs.

rw

[Sir Terrence the Terrible]

Differentiating between music and speech for the purposes of telecommunications is a false distinction.

Ok,

I did not ask you post the paper, I asked you to state the name of the paper or any other pertinent paper for that matter. By definition, an accepted theory would be documented in multiple places.

Since I read the paper while passing the exibit, I didn't pay attention to the name of the paper or who did it. I just ask what the long line was for while attending the convention, and the person I was with put that white paper in my face. I asked was it credible, and the guy I was with(who actually WAS interested in it) told me it passed peer review, nobody is challenging it. I read it, I know it exists, but it was not my interest. It is still not my interest, and I am not going to spend my time chasing it.

Here you are simply wrong, I have posted a link to an academic digital theory primer to make the point. You previously said "Let's start with the waveform. Once a analog signal is encoded to 16bit resolution, it appears as a stair stepped waveform, unlike a recorded analog which is more rounded. A DAC see's that waveform, and attempts to transcode that waveform exactly as it is represented - a series of snapshots that represent a stair stepped pattern" . However the primer says

"Diagram 2a shows an acoustic sine wave with sample points at regular intervals in time from right to left on the x-axis. The sample values shown in Diagram 2b create the digital representation of the wave file shown in Diagram 2c. A low pass filter is used upon output to smooth the edges of the new digital wave, which represent added high frequencies, with the resultant wave shown in Diagram 2d."

In otherwords, the DAC produces a pretty decent analog waveform, not a stair stepped pattern, once anti-aliasing is applied, which is as one would expect given sampling theory.

I am not debating on whether you see a analog waveform after reconstruction(anti aliasing is a front loaded process not a rear loaded one, reconstruction is done after conversion) we both agreed what is done is an analog waveform. My contention(and what I have seen) is the digital waveform does NOT look like an all analog waveform, even though they are both analog. I said it sawtoothed BEFORE conversion, and that is what the DAC is reconstructing, snapshots of the signals, not the signal in its entirety like analog represents(at least that was what I was trying to say.)

Well, as the primer clearly illustrates your original contention is simply wrong, the output of the DAC is not a stair-stepped waveform, after anti-aliasing, the resultant analog waveform is smooth and similar to (or in your words, looks like) the original waveform . and by extension your deduction about sound differences due to differences in waveform pattern is also invalid.

Your dependency on theory shows that you have zero experience in real life. If all things were perfect, you would be correct. In the real world, all things are not perfect. When a signal passes through a real life DAC, it is not a perfect process that mirrors you visual example. Theory often fails in the presence of real life ADC and DAC chips. I have not heard a perfect digital system(or analog for that matter).

If all is perfect as you describe, then why oversampling? You shouldn't need it all is perfect right?

Once again, if the two waveforms are exactly alike as you say, then why do they sound so different even when the same equipment is used?

[theaudiohobby]

Equipment problem #1: incorrect measurement of the amplitude of the source signal, i.e. getting the bits right

This is a quality assurance issue, If the ADC is unable measure amplitude correctly, it’s broken. That said, there are overload conditions, where the ADC input is overloaded and under this conditions, bits recorded would be wrong, herefore care is typically taken overloading the ADC. Sir Terence is a professional engineer, so he will be in a better position to discuss this one.

Equipment problem #2: getting the ADC and subsequently DAC timing right, i.e. sampling at precisely the right rate, i.e. preventing "jitter"

This is an issue in professional recording and the traditional solution has been to slave all converters to a single clock. See some John Atkinson’s recordings notes

Technical problem #1, 2, ... etc.: eliminating the spurious sound of samples about 1/2 the sampling frequency, i.e. how do you dump the frequencies above 22 kHz (in case of CD) which are noise? I suppose this might be dealt with more or less well. The techniques of anti-aliasing, dithering, and filtering are very obscure to me, Except I think dithering has to do with randomizing the digital noise -- signal above 22 kHz -- which would otherwise vary directly with the "good" signal below 22 kHz). Also, I understand that filtering can (or must?) affect the phase of the signal below the 1/2 sample rate frequency.

Dithering is a process that decorrelates quantisation noise from the digital signal across the entire bandwidth, it’s separate and different from anti-aliasing where noise, specifically, alias images above 22kHz are removed by the anti-aliasing filter i.e. low-pass filter. At this junction a new concept comes into play the passband and the transition band. The passband of CD is typically DC-20kHz, above that you have the transition band, the anti-aliasing filter generally does most of its filtering in the transition band. Filtering affects phase below ½ sample rate frequency, however if the transition band is suffiently large, phase distortion in the passband is negligible. IMO, this is one of the area where a higher sampling rate or oversampling has some benefits.

Psycho-accoustic problem: do sounds above concious audibilty, say 20+ kHz, really have no effect on sound perceptions?

Not sure, there is only way to answer that question, run controlled listening tests.

[Feanor]

...
Dithering is a process that decorrelates quantisation noise from the digital signal across the entire bandwidth, it’s separate and different from anti-aliasing where noise, specifically, alias images above 22kHz are removed by the anti-aliasing filter i.e. low-pass filter. At this junction a new concept comes into play the passband and the transition band. The passband of CD is typically DC-20kHz, above that you have the transition band, the anti-aliasing filter generally does most of its filtering in the transition band. Filtering affects phase below ½ sample rate frequency, however if the transition band is suffiently large, phase distortion in the passband is negligible. IMO, this is one of the area where a higher sampling rate or oversampling has some benefits.
...

Thess clarifications help, thanks.

Is dithering always used? Or is it used mainly for downsampling? Or maybe I should ask for a general definition of "quantisation noise".

[Sir Terrence the Terrible]

Thess clarifications help, thanks.

Is dithering always used? Or is it used mainly for downsampling? Or maybe I should ask for a general definition of "quantisation noise".

Dither is only used during downconversion.

Quantization noise(better known as quantization error) is the difference between the actual analog value and quantized digital value. It happens when rounding or truncation occurs. Quantization refers to approximating the output by one of a discrete and finite set of values.

[theaudiohobby]

This clarifications help, thanks.

Is dithering always used? Or is it used mainly for downsampling? Or maybe I should ask for a general definition of "quantisation noise".

Quantization noise, also known as Quantization error is a rounding error which represents the difference between the Analog Input and Quantized sample, As the quantized sample is alway bandlimited, i.e. bandwith is not infinite, there is always a quantisation error, consequently the size of the quantization error is very much dependent on the sample rate and wordlength. As the sampling rate and the wordlength increase, the quantization error decreases. It's called Quantization noise when the error is modelled as noise. The big issue with quantization noise is that it is not totally decorrelated from the signal and that's where dither comes in, dither is random noise that when added to the signal in a deterministic fashion, decorrelates the quantization error from the signal. This process improves the output of the digitization process. And yes in most modern ADCs, dither is employed, due to it's beneficial impact on the quality of the digitized signal. As an aside, tape noise is a form of natural dither, it's random nature effectively decorrelates quantization noise from the signal during the digitization process.

Sorry for the late response, I have been a bit busy over the last few days.

[theaudiohobby]

I'm with you Bill, it's alot to digest. One thing I am wondering about is oversampling.

My Magnavox has 4 times OS. Is less OS substantial in terms of arriving to a sound closer to NO oversampling? Or is it once you OS, regardless of times, the result is the same.

I have always made the naive assumption that less oversampling would be desirable, for someone who enjoys Non-OS.

IMO, oversampling is a bit of red herring as the key differentiating factor between Non-oversampling and oversampling DACs is the lack of digital anti-aliasing filter rather than th e presence or not of oversampling (at least that's what the marketing blogs say ). The lack of a digital anti-aliasing filter is justified on the basis that it's absence results in less ringing and phase distortion, however this is a trade-off, as the HF spurie (alias images) leak into passband and distort the signal.

The issue here is not complexity but stepness, the transition band in the CD without oversampling is very narrow between 20kHz-22.1kHz, a steep filter is necessary, however with an oversampling DAC running 4X, transition band increasea ro 20-88.4kHz, which is about 30X larger, this makes very gentle filter slopes a viable option.

Arguably, you could design an oversampling DAC without a digital anti-aliasing filter and get away with it as the HF components (see prior post) that potential distort the signal are at much higher frequency. At 4X oversampling, that's at >88kHz.

[theaudiohobby]

Dither is only used during downconversion.

You mean explicit use of dither as opposed to fixed dither used by a given ADC design for improved performance, correct?

[theaudiohobby]

Since I read the paper while passing the exibit, I didn't pay attention to the name of the paper or who did it. I just ask what the long line was for while attending the convention, and the person I was with put that white paper in my face. I asked was it credible, and the guy I was with(who actually WAS interested in it) told me it passed peer review, nobody is challenging it. I read it, I know it exists, but it was not my interest. It is still not my interest, and I am not going to spend my time chasing it.

I hear you, so much for the widely accepted theory.

I am not debating on whether you see a analog waveform after reconstruction(anti aliasing is a front loaded process not a rear loaded one, reconstruction is done after conversion) we both agreed what is done is an analog waveform. My contention(and what I have seen) is the digital waveform does NOT look like an all analog waveform, even though they are both analog. I said it sawtoothed BEFORE conversion, and that is what the DAC is reconstructing, snapshots of the signals, not the signal in its entirety like analog represents(at least that was what I was trying to say.)

I am baffled by your commentary, why would anyone care about the digital waveform that feeds the DAC? It's an intermidiate process and the DAC is the intended consumer not human ears. The DAC can read the data stream from the modulated carrier signal without issue and recreate the original analog signal, so why the contention?

Your dependency on theory shows that you have zero experience in real life. If all things were perfect, you would be correct. In the real world, all things are not perfect. When a signal passes through a real life DAC, it is not a perfect process that mirrors you visual example. Theory often fails in the presence of real life ADC and DAC chips. I have not heard a perfect digital system(or analog for that matter).

The theory is the basis for the real-life implementation. Without it, the real-life would not exist.

If all is perfect as you describe, then why oversampling? You shouldn't need it all is perfect right?

That's a separate discussion, right? By the way, this is the second time you raised the perfection strawman.

Once again, if the two waveforms are exactly alike as you say, then why do they sound so different even when the same equipment is used?

I disagree, but that's another separate discussion.

[Sir Terrence the Terrible]

You mean explicit use of dither as opposed to fixed dither used by a given ADC design to improved performance, correct?

Correct. I would not use a A/D converter that had fixed dither.

[Sir Terrence the Terrible]

I hear you, so much for the widely accepted theory.

It was accepted by those who critiqued it, those who read it, and those who experienced the test themselves. Those who read and hang on to basic theory may have missed its reach.

I am baffled by your commentary, why would anyone care about the digital waveform that feeds the DAC? It's an intermidiate process and the DAC is the intended consumer not human ears. The DAC can read the data stream from the modulated carrier signal without issue and recreate the original analog signal, so why the contention?

Maybe that is a question you need to ask yourself.

The theory is the basis for the real-life implementation. Without it, the real-life would not exist.

It may be the basis for real time implementation, but it is not a real time reality using real equipment in real environments.

That's a separate discussion, right? By the way, this is the second time you raised the perfection strawman.

And this is the second time you have refused to answer the basic question. You should be able to support your theories with answers.

I disagree, but that's another separate discussion.

And with that, I guess this whole stream minutia is concluded.

[theaudiohobby]

Correct. I would not use a A/D converter that had fixed dither.

The larger question is do you have a choice in the matter? Here's a random link from the web for you. And here's an excerpt for your education.

All A/D converters have built-in dither. It is a property of the converter chip, manifesting itself as an unavoidable noise floor. A/D converter chip manufacturer's use this noise - on purpose - to linearise the chip's behavior at lower signal levels.

[Sir Terrence the Terrible]

The larger question is do you have a choice in the matter? Here's a random link from the web for you. And here's an excerpt for your education.

All A/D converters have built-in dither. It is a property of the converter chip, manifesting itself as an unavoidable noise floor. A/D converter chip manufacturer's use this noise - on purpose - to linearise the chip's behavior at lower signal levels.

Some chip manufacturers use that noise to linearize their chips, not all of them. Can't make global statements without verification can we?

If thermal noise was enough to deal with the issue, then engineers would not need to use dither at all ever. That has never been the case when I work in 16bit, so this angle is a bit of a red herring in and of itself. No thanks to your supposed education.

[theaudiohobby]

And with that, I guess this whole stream minutia is concluded.

Correct, as you are now tripping over yourself.

[Sir Terrence the Terrible]

Correct, as you are now tripping over yourself.

Posturing is totally unnecessary, move on dude.

[theaudiohobby]

Some chip manufacturers use that noise to linearize their chips, not all of them.

What do the others do to linearize their chips, sprinkle them with fairy dust?

Can't make global statements without verification can we?

If thermal noise was enough to deal with the issue, then engineers would not need to use dither at all ever. That has never been the case when I work in 16bit, so this angle is a bit of a red herring in and of itself. No thanks to your supposed education.

Where did that come from, you've gone off on another tangent, the article discusses the use of dither in ADC chips, what's with the thermal noise?

[Sir Terrence the Terrible]

There is nothing more to be added to where this last bit of foolishness is taking us.

[theaudiohobby]

There is nothing more to be added to where this last bit of foolishness is taking us.

This is where I started "the output of a DAC is an analogue signal and strictly speaking no one ever really listens to digital but to a analogue output recovered from a digital encoding scheme",
And this is were you were a few posts ago "I am not debating on whether you see a analog waveform after reconstruction(anti aliasing is a front loaded process not a rear loaded one, reconstruction is done after conversion) we both agreed what is done is an analog waveform. My contention(and what I have seen) ..."

As you are now in agreement with my original comments, it's time to call it a day have a good evening.

[Sir Terrence the Terrible]

This is where I started "the output of a DAC is an analogue signal and strictly speaking no one ever really listens to digital but to a analogue output recovered from a digital encoding scheme",
And this is were you were a few posts ago "I am not debating on whether you see a analog waveform after reconstruction(anti aliasing is a front loaded process not a rear loaded one, reconstruction is done after conversion) we both agreed what is done is an analog waveform. My contention(and what I have seen) ..."

As you are now in agreement with my original comments, it's time to call it a day have a good evening.

I have asked you this question three times now, so now that we are in agreement with what we already stated we agreed upon, answer my question.

If your contention is that the DAC reconstruction of a signal is the same as a analog based signal with no digital encoding, then why do they sound so different to the ear even on a system that is designed for transparent playback of both? It should by your adherence to theory(and ignorance for real world equipment capabilities to deliver that theory perfectly) they should sound exactly the same, with no measurable difference in frequency response, timbre and tonality, or any other audibly perceptual perimeter.

I am sure the your firm adherence to theory will give you a very convincing answer to this.

And by the way, it is never good to parcel a quote to make a point. Print all of the quote, so the the answer remains in context from beginning to end.

[Ajani]

Seeing Vincent Audio's new KHV-111MK headphone amplifier [$499.95] on the news section of 6Moons + my thread on the Asgard and Valhalla headphone amps from Schiit Audio reminds me that a major trend seems to be towards headphone users.

So Why Headphone Amps?
A realization. In the old days, audiophiles went up the food chain from the table radio to the console stereo to separate speakers the size of refrigerators and monoblocks that would cook a cat. Today, nobody starts with a table radio. Everyone—and we mean everyone—starts with an iPod and headphones screwed into their ears. Headphones are now the standard.

And what happens when someone decides to lose the earbuds and get some serious headphones? They quickly find that their iPod (or Zune, or laptop, you know what we mean) won’t drive them. Or sounds like crap. Or both.

Now it’s time for them to get a headphone amplifier. Which is where we come in.

I think that quote sums up what's happening with young, potential audiophiles... Some will stick with headphones as their primary (or even only) HiFi setup, while others will advance to speakers.

[Florian]

I would say the future goes towards downloading. Currently i run a 32Mbit line connected to my MBP and when i am lazy use the optical digital out into my Sphinx DAC. It gets a iTunes Album in less then 45 seconds. On the other hand, i buy a lot of CDs and LPs as well. Its good to feel