Quote Barish:

---

You are making us write the same things over... and over... and over... again.


PLEASE READ THOROUGHLY BEFORE POSTING.


...Including the links.

B.

---

Barish,

It is obvious when you look at it (if you take the time), but many people don't believe the obvious anymore. In this "Quantum" world people are sceptical about scientific truths. There has obviously been problems with digital recording, processing and playback. Many people believe that the answers lie in the sampling rate. They feel if the maths don't add up, then the maths must be incomplete like classical physics, etc, etc. You are not going to win any arguments by hitting them again and again with the maths.

I personally have my own beliefs on why digital has often failed to satisfy, (not sampling rates) but I will keep those to myself.

Cheers

Dave

--------------------
Lighthouse Mastering

Thanks James - that's my internet research project for the afternoon set out then.

Quote kris:

---

But it's wrong, there IS approximation, as we're restricted by limited bit-depth, and higher sample rates CAN give a better approximation of frequencies at <1/2 sample rate, although the differences are probably inaudable.

---

I can't agree with this I'm afraid. Sure, quantisation is an imperfect process and the resolution is determined to a large dgree by the word length used. But changing the sample rate has no effect on the wordlength, and thus it can not alter what you refer to (incorrectly) as 'approximation of frequencies at >1/2 sample rate'

hugh

--------------------
Technical Editor, Sound On Sound

Quote Pangloss:

---

The algorithms used by the reconstruction filters - are they open source? I.e. is there a "correct" implementation, one that can be found in a numerical recipes-type text book. Or are they proprietry? That is to say, are some manufacturers' reconstruction filter designs inherently better than others (i.e. do some more intelligently interpolate between samples than others).

---

Tricky question to answer. This is the cutting edge of the art and science. I'm not very knowledgable in the intricate details of converter design I'm afriad, but I don't think there is anything around by the way of open-source reconstruction filters and the like. Having said that, though, the principles involved are all pretty standard stuff.

Perhaps the best place to seek more information is the converter chip manufacturer websites. You'll be able to see spec sheets there with loads of frequency ploys showing the different filter characteristics used. AKM, Crystal, Cirrus, Burr Brown and so on all use slight variations on the themes.

The high end converter box makers then choose specific chips over others because they feel they sound better -- largely because of the way the filtering has been implemented. Read my review of the Benchmark DAC1 for a little insight into this -- I had a fascinating chat with the designer about converter filter issues and some of that chat made it into the review.

One of the main issues is that of out of band noise -- probably more so than the details of pass band ripple and stop band slopes, in fact.

Most, if not all, audio converters these days are of the delta-sigma type, which operate at a very high sample rate with low bit depth, and then use decimation (essentially a digital filtering process) to translate to a lower sample rate and much higher wordlength.

Part of this process generates a lot of quantisation noise which is piled up above 20kHz. Exactly how much, and how it is shaped spectrally, varies a lot between different chip-makers. At the last AES one chip maker was making a big song and dance at its show stand about how much less ultrasonic noise it produced than its competitors.

Hugh

--------------------
Technical Editor, Sound On Sound

Actually what probably makes 96 or 192 khz superior to plain
44.1khz sampling is not the fact that the sampled data is more accurately reproduced but rather the fact,
that when working on the recorded data in the DAW, the
DSP-algorithms work far nicer if the sampling-rate is
higher. This is the reason why quite a few VST-developers
resample stuff from 44.1 khz upwards internally for
processing and later on convert it back to 44.1.

Obviously this concept is a little bid absurd if you
have a chain of 4 VSTs on one track each interpolating
and decimating, causing much more load than necessary.

I really don't understand why the developers of DAW-Hosts
like Steinberg or Apple don't support some kind of
hybrid-mode, where sampling & playback is done with 44.1khz,
but the processing-chain itself is run at a higher sample-rate. This would prevent multiple chained SRCs and still
allow effects to use the increased bandwidth for better
DSP operation...

Tobias

Quote:

---

James Perrett:
Different designers make different choices. Some choose a slightly lower corner frequency with a shallower slope while others choose a higher corner frequency with a steeper slope. The steeper slope is possibly harder to implement and you may have to put up with a small amount of frequency response ripple in the passband. There are also issues of phase to think about - whether you go linear phase or minimum phase.

---

And no one filter will work perfectly with every programme source in every configuration, at every sample rate.

In their newer A-D converters, dCS have made a move towards addressing this by allowing the user to choose from 4 different anti-aliasing filters and three different types of noise shaping according to what suits the project in hand. Very effective they are too. The D-A converters also allow you a bit of choice with their filters letting you chose to trade off transition curves/impuse responses/stereo imaging against Nyquist imaging. Again, a useful feature and not as subtle as one might expect. It's still possible to vanish up your own backside when comparing filters and noise shaping though!

This isn't particularly a new thing - In the 1990s, Harmonia Mundi Digital (early Daniel Weiss gear) allowed the user to select filters, noise shaping and dither algorithms in their 102 mastering processor, and Digital Integration and ADT both made really dinky SRC boxes which let you choose from loads of different dither types and filters. On a simpler level, things like Apogee UV20/UV22/etc. and all the multiplicity of dither options now available in mastering/recording software all give the ability to optimise conversions to different jobs via various presets and selectable options.


There's some interesting general reading on the Tech Papers section of the dCS website.

Quote Lighthouse_Mastering:

---

There has obviously been problems with digital recording, processing and playback. Many people believe that the answers lie in the sampling rate.

---

You are right, of course. There have been problems with the first few commercial generations of digital equipment. In the main I think we can largely blame the marketing hype over the engineering reality.

Some of the first experimental machines operated at 60kHz sample rates, and they showed enormous promise -- personally I think that is the rate that should have been retained from the start. Sadly, Sony's cost-cutting to use existing video recording hardware as the data storage device imposed 44.1kHz on us, and the rest is history.

But the real problem with early digital equipment was that of the inherent practical problems associated with the anti-alias and reconstruction filtering, and the limited bit depth (many so-called 16 bit machines were actually struggling to achieve 14 bit performance, for example). It was the practical implementation of the theory that was flawed, not the theory itself. Indeed, the theory has been proven to be complete and accurate coutless times, and in countless industries -- not just audio.

If the sampling theorum was flawed we wouldnb't have telephone systems, planes would fall out of the sky on a regular basis, TV and films wouldn't work, and so on!

As far as audio is conerned, though, things have come on in leaps and bounds in recent years. The introduction of oversampling and digital filtering cured most of the practical problems with analogue anti-alias and reconstruction filtering -- and delta-sigma techniques have improved the situation further still, allowing the move to 24 bit word lengths along the way. We also now have infinitely better clock accuracy and stability, better interface standards and much more besides.

The first practical magnetic tape recorders started to appear in the early 1940s, but the technology didn't reach the peak of its performance potential until the late 1980s. That's a forty year development curve.

The first practical digital audio equipment started to appear in the early 1980s -- that's only 25 years ago. For a new technology I think it is mighty impressive and I wouldn't want to turn the clock back. Although I'm not saying we have an entirely perfect implementation of the technology yet -- or that it necessarily sounds the same -- although that particular aspect cuts both ways, of course.

Quote:

---

They feel if the maths don't add up, then the maths must be incomplete like classical physics, etc, etc. You are not going to win any arguments by hitting them again and again with the maths.

---

Er... but the maths do add up. Of that there is nodoubt whatever. The problem is that the maths involved relies on concepts and techniques which are above the understanding of the vast majority of users.

Few people understand the thermo-dynamic equations involved in defining how petrol burns in a combustion chamber and makes a car engine go faster.... doesn't stop them from using the car to get to work though! For some strange reason, people seem to want to argue about digital audio technology even though they don't undertsand it, yet not about the merits of fuel additives and octane ratings. Maybe I'm on the wrong forum

hugh

--------------------
Technical Editor, Sound On Sound

/Groans

> I can't agree with this I'm afraid. Sure, quantisation is an imperfect process and the resolution is determined to a large dgree by the word length used. But changing the sample rate has no effect on the wordlength, and thus it can not alter what you refer to (incorrectly) as 'approximation of frequencies at >1/2 sample rate'

I didn't incorrectly refer to frequencies of >1/2 sample rate, you incorrectly quoted me . I said frequencies of <1/2 sample rate and it's easy to show this mathmatically, especially with signals represented with low bit-depths.

Here's a simple example, take a 1-bit system (or indeed a signal which only modulates the lowest bit) with a 5hz sampling rate. A 2.5hz sine wave could clearly be encoded as 1010101. Now show me how 2.4, 2.3, 2.2, 2.1 or 2hz sine waves (ALL of which fall below the Nyquist limit) can be encoded.... Do you think a 100hz sampling rate could reproduce these frequencies more accurately?

Now of course digital system don't work with 1-bit precision, and this error is inversely proportional to bit-depth, so is almost certainly inaudible. That however doesn't change the fact that's it's still there, and higher sampler rates can reduce it.


> This is a common argumant, but is technically incorrect I'm afraid. If you re-assess the Nyquist theorem you quote, you'll see that the signal can be fully reconstructed -- without any waveshape distortion and with perfect accuracy -- provided there are at least two samples per cycle. The maths proves this without any doubt whatever.

No, provided there are two samples of INFITINE PRECISION per cycle, something no digital system will ever have...

This thread gets better and better.
Any chance of an SOS article on the merits of 192kHz.
e.g. A fact and fiction guide for dummies like me.

--------------------
.. is this thing on?

Quote kris:

---

Now of course digital system don't work with 1-bit precision,

---

Err - yes they do... it would be well worth doing a little reading on how digital audio systems really work before trying to tell us what is wrong with them. One bit sampling is extremely common - bitstream, MASH and DSD are all names of one bit sampling systems used in digital audio.

Cheers

James.

--------------------
JRP Music - Audio Mastering and Restoration.
http://www.jrpmusic.net

Quote kris:

---

I didn't incorrectly refer to frequencies of >1/2 sample rate, you incorrectly quoted me .

---

Apologies for the misquote. Fingers faster than brain sometimes...

Quote:

---

Here's a simple example, take a 1-bit system (or indeed a signal which only modulates the lowest bit) with a 5hz sampling rate. A 2.5hz sine wave could clearly be encoded as 1010101. Now show me how 2.4, 2.3, 2.2, 2.1 or 2hz sine waves (ALL of which fall below the Nyquist limit) can be encoded.... Do you think a 100hz sampling rate could reproduce these frequencies more accurately?

---

Sorry to say it yet again, but your view on this is far too simplistic. The maths are complex, and hard to understand if its not your thing (and I'm no mathematician), but they aren't wrong! Once again, the flaw in your example is the lack of the equivalent of the reconstruction filtering.

Quote:

---

Now of course digital system don't work with 1-bit precision

---

Er... talk to Sony about that and mention DSD...

hugh

--------------------
Technical Editor, Sound On Sound

> Sorry to say it yet again, but your view on this is far too simplistic.

My views on it certainly aren't simplistic, I'm simply trying to word it that way so people can understand.... and I'm obviously failing.... miserably.... as usual.....

> The maths are complex, and hard to understand if its not your thing (and I'm no mathematician), but they aren't wrong!

Which maths aren't wrong? Nyquist certainly wasn't wrong, but he didn't say you could perfectly recreate all frequencies at <1/2 sample rate when your sampled values were quantised!


> Once again, the flaw in your example is the lack of the equivalent of the reconstruction filtering.

No, the flaw is that the Nyquist theorem requires a higher sampling resolution (not rate) than a low-bit depth signal can provide (in fact to perfectly recreate the source would require an infinite bit depth). This has nothing to do with reconstruction filtering!

Quote Hugh Robjohns:

---

But the real problem with early digital equipment was that of the inherent practical problems associated with the anti-alias and reconstruction filtering, and the limited bit depth (many so-called 16 bit machines were actually struggling to achieve 14 bit performance, for example). It was the practical implementation of the theory that was flawed, not the theory itself.

---

The real problem was the need to achieve a flat 20-20K freq response, yet avoid the aliasing problems. “Brick wall” filters sounded horrible, and digital filtering and oversampling seemed necessary. I think we are still suffering, and our engineering solutions have only moved us sideways. If you listen to early multibit converters, they have qualities that modern Delta Sigmas have lost. Oversampling, Upsampling, and Delta Sigma conversion all have their own sonic signature, that we now equate to the digital sound. Sure we have lost the hard glassy sound of the early digital, but we have gained other forms of distortion.

Quote:

---

Indeed, the theory has been proven to be complete and accurate coutless times, and in countless industries -- not just audio.

---

Agreed, the theory is sound, and I do not think that very high sampling rates are by themselves the answer. The one advantage that I believe high sampling rates allow, are the movement of the aliasing products higher up the frequency range, which allows for gentle analog filtering. I don’t believe they allow us to hear any “missing information”.

Quote:

---

As far as audio is conerned, though, things have come on in leaps and bounds in recent years. The introduction of oversampling and digital filtering cured most of the practical problems with analogue anti-alias and reconstruction filtering -- and delta-sigma techniques have improved the situation further still, allowing the move to 24 bit word lengths along the way. We also now have infinitely better clock accuracy and stability, better interface standards and much more besides.

---

Each of these “advances”, has brought its own set of problems. Oversampling and digital filtering were solutions to aliasing problems with low sample rates. They should not be present in higher sample rate based digital audio. I am sure that a 96Khz based system using no oversampling or digital filtering, and based around high quality multibit converters would sound amazing. Unfortunately we cannot break free from the solutions to our past mistakes, and digital is always likely to sound nothing like analog.

Quote:

---

Er... but the maths do add up. Of that there is nodoubt whatever. The problem is that the maths involved relies on concepts and techniques which are above the understanding of the vast majority of users.

---

I agree, the maths do add up, but there are people who are always going to believe that it is incomplete. The fact is, if the maths IS right, then where is the problem? Could it be that our implementation still stinks

Quote:

---

Maybe I'm on the wrong forum

hugh

---

No, from what I have seen, I think you do a great job at bringing some common sense and knowledge to discussions that are often missing in those departments.

Cheers

Dave

--------------------
Lighthouse Mastering

Quote Hugh Robjohns:

---

The problem is that the maths involved relies on concepts and techniques which are above the understanding of the vast majority of users.

*snip*

For some strange reason, people seem to want to argue about digital audio technology even though they don't undertsand it...

---

In this age of wide-spread political correctness, everyone apparently has a right to feel good about themselves, and everyone's ego deserves a lil' message. The line between facts and opinions has become blurred in so many people's mind to the point where they don't consider the difference, and all expressions are considered equally important and valid, all deserving of protection.

It doesn't matter if someone actually has FACTS on their side, or if they actually have any knowledge or experience on the subject matter: everyone with a modem is "entitled to an opinion". We can't let FACTS stand in the way of what we believe, can we?

It's gotten to the point where some school administrators discourage teachers from marking students' exams with a red ink pen, since the red ink is considered "too harsh and traumatizing". Eventually, teachers no doubt will feel compelled to erase the student's incorrect answer, and simply write in the proper answer, since you don't want to bruise the little punter's fragile ego by pointing out they're wrong.

Obviously, there is a price to be paid for subordinating objective facts to desires for diplomacy and political correctness; namely, a dummying-down of our culture.

Chris

--------------------
1.5GHz Al 17" Powerbook G4 (2.0GB RAM, Hitachi 60GB 7,200 rpm drive), running Logic Pro 7 under OSX 10.4.5

Quote Feefer:

---

Obviously, there is a price to be paid for subordinating objective facts to desires for diplomacy and political correctness; namely, a dummying-down of our culture.

Chris

---

Did I tempt this response with my 'for dummies' suggestion?
Not fair.

--------------------
.. is this thing on?

Edited by Stan (31/01/06 07:29 PM)

Quote Stan:

---

Did I tempt this response with my 'for dummies' suggestion? Not fair.

---

No, you didn't tempt the response. Education is fine.

Just realize there's a big difference between writing a book for dummies vs. letting all the dummies author a chapter in the book! And believe me: I won't be writing a chapter in that book, as some of the people here have gotten in way over my head (and I ain't in a learning mood, right now).

Chris

--------------------
1.5GHz Al 17" Powerbook G4 (2.0GB RAM, Hitachi 60GB 7,200 rpm drive), running Logic Pro 7 under OSX 10.4.5

Quote kris:

---

>My views on it certainly aren't simplistic, I'm simply trying to word it that way so people can understand.... and I'm obviously failing.... miserably.... as usual.....

---

Well, I've got to say in all the years I've been dealing with digital audio in one form or another I have never once come across this argument before. I'm not convinced by it, but I'm open minded enough to look into it further if you can suggest where to go looking....

Quote:

---

Nyquist certainly wasn't wrong, but he didn't say you could perfectly recreate all frequencies at <1/2 sample rate when your sampled values were quantised!

---

True enough, he wasn't dealing with quantisation. But I'm still struggling to see how quantisation noise can affect sampling accuracy in the way you claim.

Quote:

---

No, the flaw is that the Nyquist theorem requires a higher sampling resolution (not rate) than a low-bit depth signal can provide (in fact to perfectly recreate the source would require an infinite bit depth).

---

Hmmm....

hugh

--------------------
Technical Editor, Sound On Sound

Quote Lighthouse_Mastering:

---

“Brick wall” filters sounded horrible, and digital filtering and oversampling seemed necessary.

---

Yes, most of the analogue brickwall filters sounded grim, but we still have them -- it's just that they are in the digital domain now where we can make them more nearly 'perfect' than we ever could with analogue designs.

Quote:

---

Sure we have lost the hard glassy sound of the early digital, but we have gained other forms of distortion.

---

Very true. No such thing as a free lunch. Delta-sigma converters do have their own unique set of issues to deal with... but on the whole they come closer to the holy grail than anything else so far.

Quote:

---

I am sure that a 96Khz based system using no oversampling or digital filtering, and based around high quality multibit converters would sound amazing.

---

But it can't be done with current technology. It was hard enough getting multibit converters to operate with 16 bit accuracy, and it really was rocket science trying to squeeze 20 bits out of them -- and that was at 48kHz. Variations on the delta-sigma theme coupled with oversampling and digital filters is the only practical way at the moment.

Quote:

---

digital is always likely to sound nothing like analog.

---

True... but is that really a bad thing? I loathe the digital v analogue comparisons. They are pointless. What we should be comparing is analogue and digital against the source. And if you do that, I'm happy these days that good 24/96 digital wins every single time -- you get back exactly what you put in as far as my ears can tell. 16/44.1 is close but the top end tends to sound a little congested. Analogue comes back sounding warm and fuzzy. Not unpleasant, but not accurate.

Analogue is great and will be with us for a long time to come because of its simplicity and sound character, but let's not confuse it with precision or accuracy.

Quote:

---

The fact is, if the maths IS right, then where is the problem? Could it be that our implementation still stinks

---

I wouldn't go as far as saying it stinks, but agree with you that most budget to mid level implementations are still lacking in various small but significant ways. The really high end stuff is stunningly good, but it will be a while before that quality reaches our daily lives.

Technology advances at quite a pace though. Compare cheap mixer mic amps now with ten years ago and you'll be shocked with how bad they were back then! Compare a modern cheap mixer mic amp with a good high end outboard one, and you'll be shocked again at how limited they still are!

Hugh

--------------------
Technical Editor, Sound On Sound

I hate to say this Hugh, but you're really letting down the whole forum with your inappropriate attire this morning. A chap should keep his evening wear for the evening.


(Incidentally, does your dinner jacket have the regulation BBC leather elbow patches?)

That is an interesting point that Lighthouse_Mastering is making regarding the quantisation of the available samples. Fitting some sort of sine curve over infinitely quantised samples, while non-trivial, at least looks fairly do-able. However, if each of those samples has a small quantisation error in the time direction then fitting a curve starts to look much more like a statistical fit and therefore more subjective? No? Isn't there going to be some uncertainty in the phase accuracy whichever sine wave you choose?

Edited by Pangloss (01/02/06 10:58 AM)

Quote Hugh Robjohns:

---

Delta-sigma converters do have their own unique set of issues to deal with... but on the whole they come closer to the holy grail than anything else so far.

---

We will have to disagree there. The distortions that delta sigmas introduce are far more incidious than any problems with multibit

Quote:

---

Quote:

---

I am sure that a 96Khz based system using no oversampling or digital filtering, and based around high quality multibit converters would sound amazing.

---

But it can't be done with current technology. It was hard enough getting multibit converters to operate with 16 bit accuracy, and it really was rocket science trying to squeeze 20 bits out of them -- and that was at 48kHz. Variations on the delta-sigma theme coupled with oversampling and digital filters is the only practical way at the moment.

---

You must be refering to AD conversion, because there have been plenty of 24bit mutibit DA converters capable of running at high sample rates. I am sure that the dirth of multibith AD's is due to lack of demand (and the cost), and not a technical issue. Surely the best we can get out of any conversion system is just over 120db.

Quote:

---

Quote:

---

digital is always likely to sound nothing like analog.

---

True... but is that really a bad thing? I loathe the digital v analogue comparisons. They are pointless. What we should be comparing is analogue and digital against the source. And if you do that, I'm happy these days that good 24/96 digital wins every single time -- you get back exactly what you put in as far as my ears can tell. 16/44.1 is close but the top end tends to sound a little congested. Analogue comes back sounding warm and fuzzy. Not unpleasant, but not accurate.

---

Sorry, I do not want to drag this into an analog vs digital debate either. The point is that the conversion processes should be transparent. If you take an analoge recording and convert it to digital, and then convert it back, it should not be fundamentally transformed. The issue for me is that with analog processing, each step causes a gradual degradation, whilst the conversion to/from digital is a much more fundamental change.

Quote:

---

Quote:

---

The fact is, if the maths IS right, then where is the problem? Could it be that our implementation still stinks

---

I wouldn't go as far as saying it stinks, but agree with you that most budget to mid level implementations are still lacking in various small but significant ways. The really high end stuff is stunningly good, but it will be a while before that quality reaches our daily lives.

---

I don’t believe that the sonic differences between esoteric and mid/budget converters can be put down to the conversion or filter technology. It is more to do with the quality of the analog circuits, the power supply, and the overall precision of the implementation. Even so, the best digital processing equipment is in my opinion flawed.

Quote:

---

Technology advances at quite a pace though. Compare cheap mixer mic amps now with ten years ago and you'll be shocked with how bad they were back then! Compare a modern cheap mixer mic amp with a good high end outboard one, and you'll be shocked again at how limited they still are!

Hugh

---

There is a limit to what you can do within a budget. Good quality analog stages and power supplies are expensive, so is the research to get the most out of technology.
It is easy to see how cheap mixers have been improved, but I cannot see where the improvements in digital are going to come from. I think we missed a turn, and are proceeding up a dead end. The maths is perfect, we cannot improve the signal to noise, and higher sampling rates (in the current implementations) have not brought the promised improvement. What is left to work on?

Since the best quality digital replay I have heard (by ten country miles), was from an antiquated Multibit DAC, with no digital filtering or oversampling, I cannot see how further digital filter development or advances with delta sigma conversion are going to help.

Dave

--------------------
Lighthouse Mastering

Quote kris:

---

No, the flaw is that the Nyquist theorem requires a higher sampling resolution (not rate) than a low-bit depth signal can provide (in fact to perfectly recreate the source would require an infinite bit depth). This has nothing to do with reconstruction filtering!

---

Nyquist theorem is not about the dynamic range, but about the frequency content.

The dynamic range of human hearing: 110dB on average.

24 bits: 144dB, all covered and then some.

16 bits: 96dB, quite an adequate range if used properly, considering the effects of correct dithering.


Problem solved. What are you talking about?


Sorry Hugh, I'm not really convinced that open-minded approach to give everyone's opinion a benefit of doubt is really working here. This is like giving Hitler a benefit of the doubt that he might be up to something right.


I'm off.


B.

(sorry Lighthouse - just re-read the thread and it looks like I was putting words in your mouth there. Quoting the wrong person).

Quote Pangloss:

---

(sorry Lighthouse - just re-read the thread and it looks like I was putting words in your mouth there. Quoting the wrong person).

---

That's alright Pangloss. I was actually infering to the issue you mentioned, I just had not got technical. When I saw your post, I thought you were very insightful to infer the time domain quantisation distortion. Digital filtering, over(up)sampling and delta sigma conversion all cause distortion in the time domain, as I am sure does much DSP based processing.

Dave

--------------------
Lighthouse Mastering

Quote Barish:

---

Sorry Hugh, I'm not really convinced that open-minded approach to give everyone's opinion a benefit of doubt is really working here.

---

I'm even slightly not convinced, as I said. But he seems adamant so the least I can do is recheck the concept -- as much for my own peace of mind as anything else. I see no hint of any suggested references yet though...

hugh

--------------------
Technical Editor, Sound On Sound

Quote Lighthouse_Mastering:

---

We will have to disagree there. The distortions that delta sigmas introduce are far more incidious than any problems with multibit

---

Fair enough. Disagreement it is It's a different set of problems -- more time domain-related than frequency/amplitude domain. Idle tones and limit cycles were a major problem in some systems, but even that seems to have been largely gripped in the latest designs.

Quote:

---

You must be refering to AD conversion, because there have been plenty of 24bit mutibit DA converters capable of running at high sample rates.

---

I'm struggling to think of any... pretty much everything I can think of uses some variation on the theme of a low wordlength delta-sigma topology (ie. 3, 4, or 5 bits in a highly overampled D-S system).

Quote:

---

I am sure that the dirth of multibith AD's is due to lack of demand (and the cost), and not a technical issue. Surely the best we can get out of any conversion system is just over 120db.

---

My understanding is that it is very much a technical issue, which is why I made the statement in the first place And yes, 120dB or there abouts is pretty much as good as the dynamic range gets with current technology.

Quote:

---

It is easy to see how cheap mixers have been improved...

---

Really? Compare the circuit design of the mic pre in an early Mackie mixer and the later XDR or Onyx circuits. The changes (on paper) are not that significant at all, yet the sound quality is leagues apart!

Quote:

---

but I cannot see where the improvements in digital are going to come from.

---

I expect a lot of people said the same thing when Sony were struggling to make 16 bit multibit converters work properly. And then Philips introduced oversampling. And then people found ways of making clocks far more stable. And then delta-sigma converters came along.... There will be numerous advances in the years to come -- mainly to do with the way the existing technology is implemented, but I dare say some clever new techniques will also be found. Just as the wow and flutter of tape transports and record players improved dramatically over decades. The concepts didn't change, but the ability to engineer better solutions improved.

hugh

--------------------
Technical Editor, Sound On Sound

> Nyquist theorem is not about the dynamic range, but about the frequency content.

The Nyquist theorem requires that each point must be sampled accurately for it to work properly. If you can't accurately sample the lever of each point (and you can't at low bit-depths), the ONLY possible other way to give a better approximation of the input is to increase sampling rate.

The two ARE related.


> Problem solved. What are you talking about?

Barish, if you read my reply you'll see I actually said the differences are probably inaudable. That wasn't my point, the point is people are saying NO information is lost and there's NO approximation, when information IS lost due to limited bit-depth (and sample rates), and a high-sampling rate CAN reduce this loss of information.

BTW, yesterday (to convice myself I wasn't going mad) I also did the example I posted (as well as various frequency sweeps) in Wavelab, Sound Forge and Adobe Audition at a friends. The results were exactly as I expected, higher sample rates resulted in a better approximation of low bit-depth frequencies of <1/2 sample rate.

As I said though, the differences in the real-world are almost certainly inaudable (we've got no argument there!), but they're still there.


> Sorry Hugh, I'm not really convinced that open-minded approach to give everyone's opinion a benefit of doubt is really working here. This is like giving Hitler a benefit of the doubt that he might be up to something right.

If you're referring to me, I think it's totally uncalled for.

Cheers

Edited by kris (01/02/06 05:46 PM)

I'd just like to say that this thread is chock full of good information. Hugh, bravo! (and you look good in evening attire...heh!).

Not singling out any one poster, but I'll say this. Sadly, at work, even my boss makes assumptions and forms opinions about tasks that are orders of magnitude more complex than they appear on the surface. It SEEMS so simple, yet rarely is.

As an aside, my brother is just finishing up his Math degree . He's still a little up in the air on what he's going to go though he's leaning toward theoretical as opposed to applied. Any suggestions on coursework that would lend itself to this type of situation (audio/frequency/sampling/etc. and the math's behind the field). I know he has at least a passing interest in audio.

Cheers!

P-X

--------------------
My Home Studio Build Thread

Quote kris:

---

Barish, if you read my reply you'll see I actually said the differences are probably inaudable. That wasn't my point, the point is people are saying NO information is lost and there's NO approximation, when information IS lost due to limited bit-depth (and sample rates), and a high-sampling rate CAN reduce this loss of information.

---

You really ought to get yourself a good text book on digital audio. You've just described oversampling which is a standard technique used in modern convertors.

Cheers

James.

--------------------
JRP Music - Audio Mastering and Restoration.
http://www.jrpmusic.net

Quote kris:

---

> Nyquist theorem is not about the dynamic range, but about the frequency content.

The Nyquist theorem requires that each point must be sampled accurately for it to work properly. If you can't accurately sample the lever of each point (and you can't at low bit-depths), the ONLY possible other way to give a better approximation of the input is to increase sampling rate.

The two ARE related.

---

Sorry Kris, but can you explain this a bit more clearly. What do you mean by nor sampling acurately, and low bit-depths?

Quote:

---

> Problem solved. What are you talking about?

Barish, if you read my reply you'll see I actually said the differences are probably inaudable. That wasn't my point, the point is people are saying NO information is lost and there's NO approximation, when information IS lost due to limited bit-depth (and sample rates), and a high-sampling rate CAN reduce this loss of information.

---

Again, I think you need to go into more depth in your arguement. I cannot see how there is information loss.

Dave

--------------------
Lighthouse Mastering

> You really ought to get yourself a good text book on digital audio. You've just described oversampling which is a standard technique used in modern convertors.

I've got plenty of good books on digital audio James. Oversampling requires a reasonable amount of resolution to work succesfully and I'm talking about low-level signals where you DON'T have a lot of sampling resolution (although this distortion occurs at all bit-depths, it's level is inversely proportional to bit depth).

Quote Hugh Robjohns:

---

Quote:

---

You must be refering to AD conversion, because there have been plenty of 24bit mutibit DA converters capable of running at high sample rates.

---

I'm struggling to think of any... pretty much everything I can think of uses some variation on the theme of a low wordlength delta-sigma topology (ie. 3, 4, or 5 bits in a highly overampled D-S system).

---

Off the top of my head, Burr Brown make a 24bit high sample rate capable multibit DAC, and I know that there are others.

Quote:

---

Quote:

---

I am sure that the dirth of multibith AD's is due to lack of demand (and the cost), and not a technical issue. Surely the best we can get out of any conversion system is just over 120db.

---

My understanding is that it is very much a technical issue, which is why I made the statement in the first place And yes, 120dB or there abouts is pretty much as good as the dynamic range gets with current technology.

---

I am sure it is just a supply demand thing. The only reason there are multibit DAC's still available is because they are used in the audiophile market, and they don't do ADC's.

Quote:

---

Quote:

---

but I cannot see where the improvements in digital are going to come from.

---

I expect a lot of people said the same thing when Sony were struggling to make 16 bit multibit converters work properly. And then Philips introduced oversampling. And then people found ways of making clocks far more stable. And then delta-sigma converters came along.... There will be numerous advances in the years to come -- mainly to do with the way the existing technology is implemented, but I dare say some clever new techniques will also be found. Just as the wow and flutter of tape transports and record players improved dramatically over decades. The concepts didn't change, but the ability to engineer better solutions improved.

hugh

---

I agree that digital is better today than it was 20yrs ago, but just not as good as it could be. Good to see you are an optimist.

Dave

--------------------
Lighthouse Mastering

> Sorry Kris, but can you explain this a bit more clearly. What do you mean by nor sampling acurately, and low bit-depths?

Dave, I'll give the example again and I'll attempt to make it clearer. by low bit-depths, I'm talking about frequencies that modulate only the last few bits in a digital system (i.e. at a very low level). For example, there could be frequencies that only modulate the lowest bit (of course there's dither, but I'm not discussing that here). A simple example would be at the tail end of a fade out.

So, let's say you sample at 10hz and you input a 5hz sine-wave (remember, in this example we're using a 1-bit system).

The ONLY possible way it can accurately be encoded is 010101010101 etc.... (of course we have a zero crossing point in reality, but I'm trying to make it simple so please don't hold that against me).

Now try a 4hz sine-wave. We have a problem, 101 is a 5hz sine-wave (after reconstruction), but 110011 would be a 2.5hz sine-wave. It's therefore IMPOSSIBLE to accurately sample this 4hz sine waves, despite the fact it falls below the nyquist limit.

Now Hugh makes a good point about the reconstruction filter taking care of this, but I'm afraid it's NOT applicable here because the filter doesn't have sufficient input resolution to produce anything like a valid output!! If it can only see a 1 followed by a 0, how can it possibly know the input wasn't simply a 5hz sine-wave?

Now take a 100hz sampling rate with the same input. 5hz could be 10x1s, followed by 10x0s. 4hz could be 12x1s followed by 13x0s.

Despite the fact that ALL freqencies are below the Nyquist limit in both cases, the higher sampling rate gives us a better approximation of the same source.

Now as we all know, we work with 16/24 bit systems, so this error is almost certainly inaudable (as it decreases exponentially with bit-depth), but it's still there!

Anyway, I'm obviously in a minority of one here (not for the first time), so it's probably best if I leave it there, although it would be nice to know which part of the above people think erroneous (apart from "all of it!" ). Oh and I think it's the first time I've ever been compared with Hitler!

Take care

kris, please please dont be offended, i have had this exact same conversation from yr point of view about 2 yrs ago now, and i got everyones backs up the same as youve done. i got pointed in the direction of some helpful links like you, and i tried to follow it up, i really did. but i CANT get me head round the maths, so now i just accept that it works, more bits + higher resolution = better sound, and the only way of judging that is with yr ears .. and thats all there is to it, just cos we dont get the equations doesnt mean we is thick - you really cant say the maths is wrong cos you dont understand it. you say you are trying to make it simple, thats the problem it just IS NOT simple

cheer up - make some music instead of worrying about reconstruction filters and that nasty little nyquist man

oh and as for comparing YOU to hitler, i think HE shouted a lot, thought he was really clever and called people he didnt like 'subhuman' ... remind you of anyone else around here ?

grainger

Quote kris:

---

If it can only see a 1 followed by a 0, how can it possibly know the input wasn't simply a 5hz sine-wave?

---

Have you ever taken a third, fourth or fifth degree function and taken its derivative, and then taken the derivative of that derivative, and another derivative from that derivative, and then taken the integral of that derivative, anthen another integral of the result, and another integral of that result and compared the final result with the original function in your life?

I'm assuming that you know how to take derivatives and integrals, judging by your cunning attitude about this subject.

B.

Quote Barish:

---

Quote kris:

---

If it can only see a 1 followed by a 0, how can it possibly know the input wasn't simply a 5hz sine-wave?

---

Have you ever taken a third, fourth or fifth degree function and taken its derivative, and then taken the derivative of that derivative, and another derivative from that derivative, and then taken the integral of that derivative, anthen another integral of the result, and another integral of that result and compared the final result with the original function in your life?

I'm assuming that you know how to take derivatives and integrals, judging by your cunning attitude about this subject.

B.

---

Barish, respect to you for your patience, but 155 posts on I've just lost the will to live on this topic!

If or when I can hear a discernible difference (or when my clients tell me it matters to them), I'll invest in it. Unless "it" is a stupid price.

Until then, as Shakespeare put it, the rest is silence.

--------------------
Dynamite with a laser beam...

> Have you ever taken a third, fourth or fifth degree function and taken its derivative, and then taken the derivative of that derivative, and another derivative from that derivative, and then taken the integral of that derivative, anthen another integral of the result, and another integral of that result and compared the final result with the original function in your life?

> I'm assuming that you know how to take derivatives and integrals, judging by your cunning attitude about this subject.

Barish yes I've studied calculus, but I really don't see the need for your hostile tone here, I'm not being agressive at all.

Anyway, a 1-bit waveform has no slope, it moves instantly from 0 to 1, so why in this case would derivatives help?

EDIT: Steve, yeah I think we've got as far as we're going to go on this one.

Edited by kris (01/02/06 11:15 PM)

I'm not agressive at all. I just asked you a question your reply to which answered your last question.

Had you really studied those parts of the calculus and done what I asked, you would have known how it knew the input wasn't simply a 5hz sine-wave.

Steve, you are right. I'm about to slash my wrists here too.


Frank Zappa was right too. There's more stupidity than hydrogen in the universe.


Everybody else is right as well. World is big enough for all of us.


I remember my uncle's theory of life: "If everyone in the society had a masters degree, who would we get to sweep the streets?"


There must be some superstitious people out there so that snake-oil traders can make a living too.

On you go, ladies and gentlemen.

B.

Who would we get to sweep the streets?
The people with masters in street sweeping?

Ive not read all of this post because it is to long.
But please correct me if I am wrong?

The way i see 192kh is like this yall?!

If ur baking a cake and you have some flour that you sive, it will not cause many lumps. If ur sive is in 192kh then when you run it thu your waves R-verb or whatever, it will process it in finer peices and so be a more refind product,I THINK??

If i produced that album i would use the max Sample rate. I dont think there is a concpiracy to fool people, i just think he doesnt see it as relavent?

Ya Raass clot!!

--------------------
Every day we are given building blocks.
Do we build a bridge or a wall?

I thought the purpose of it was to gave me an excuse to get a new soundcard?

--------------------
Signature (up to 200 characters). You may use UBBCode in your signature