Dan Lavry, design guru new white paper on the Optimal Sampling Rate for Audio!

[magiccarpetride]

life's too short to listen to mp3, who records/mixes/masters in mp3 ? The amount of reviews of digital music reduces the chance of having to listen to shoddily recorded/mixed/mastered 24/192 FLAC.

I enjoy 16/44.1 up to 24/96 (24/192 is a bit of a luxury due to space etc) Could I pick one over the other, I doubt it. The higher sample rates seem to be more atmospheric - again could be better production.

I was talking about many 'out of print' recordings that sadly seem available only in mp3 format. Publishing houses don't seem to think it's worth their while to invest in reissuing many culturally and musically important recordings from the '60s and the '70s, so we're now dependent on the kindness of the community to rescue those obsolete recordings and share them (they usually do that by converting old out-of-print LPs to digital). For some reason, I'm seeing some of those being offered only in mp3 format, but hey, I'll take it. Better lossy than nothing.

[AndyN]

I went to the Lavry website and downloaded the article on Nyquist and sampling theory. Well I must take my hat off to Dan - this article is the best I have EVER read on Nyquist sampling theorem. I think it will take a couple of more reads before every aspect of it sinks in. If you have not had a read might I suggest its time well spent.

AndyN

[bburroughs]

Nope, repetitive waveform and sample period have nothing to do with accuracy. The math (and implementation) works the same for one cycle or 1,000,000.

If I understand correctly; for the Nyquist sampling to result in a perfectly reconstructible waveform (at the receiving end) the sampled waveform must be repetitive and the waveform must be sampled for a period that is long in comparison to the waveform. This can be seen on a 'scope where a sampling rate not much greater than 2f results in a very good looking waveform on the trace of the scope for a repetitive waveform. However, audio is not quite like that as the waveforms are not repetitive but rather they are transitory. I dont therefore see how they may be reconstructed perfectly unless they are sampled at somewhat more than the 2f of Nyquist. How much more would, I guess, be dependant upon the required allowable error in reconstruction??

Perhaps you would elucidate??

AndyN

[adamdea]

If I understand correctly; for the Nyquist sampling to result in a perfectly reconstructible waveform (at the receiving end) the sampled waveform must be repetitive and the waveform must be sampled for a period that is long in comparison to the waveform. This can be seen on a 'scope where a sampling rate not much greater than 2f results in a very good looking waveform on the trace of the scope for a repetitive waveform. However, audio is not quite like that as the waveforms are not repetitive but rather they are transitory. I dont therefore see how they may be reconstructed perfectly unless they are sampled at somewhat more than the 2f of Nyquist. How much more would, I guess, be dependant upon the required allowable error in reconstruction??

Perhaps you would elucidate??

AndyN

I am no expert but I think the answer is something like this......

The reason for the "long time" qualification is AFAIK to do with the requirement of the signal to be sampled that it be band limited. In theory nothing can be band limited (filtered to a band limit) unless it is infinitely long. I wouldn't worry about this as it is really just a sort of clever mathematical point. For practical purposes it just describes the fact that the filter which band limits has to be able to see back and forth a few dozen sample points- this means that in theory there is a problem with the very beginning and end of the sampled signal. In theory every sample affects the signal at every point, but in practice the influence tends so close to zero as be negligible very soon.

Lots of people get confused about the repetitive signal point- this is a mistake: the signal to be sampled simply has to be sampled for the duration of the signal at the sampling rate (ie every 44.1kHz for cd). Any change within the signal can be captured provided that it does not involve the addition of a frequency over nyquist (half the sampling rate). It is critcal to recognise that frequency domain and time domain are two sides of the same coin and hence the maximum frequency dictates the limit of time resolution in the sense of the maximum permitted rate of change in the signal (but actually you can resolve events occurring between the samples).

To take your example of the truly repetitive signal- once you have sampled one cycle of the waveform all further samples would eb (if we knew the signal was repetitive) redundant. If the signal changes (ie its frequency spectrum is not the same throughout) that information will be captured by the new samples. You can see this by the examples given in Dan Lavrys articles on sampling which show the effect of the sinc function (reconstrction filter) on the sample values

If you are really interested in understanding this I suggest reading the Art of Digital Audio by Watkinson, or Information and Measurement by Jim Lesurf or the Principles of Digital audio by Pohlmann. Intuition is not a reliable guide here.

[bburroughs]

Well, I didn't think it was possible to answer that question in just a few paragraphs without resorting to esoteric math and engineering, but Adam's description is really excellent. Anyone interested in digital audio should read and try to understand what he wrote - it's an excellent summary of some difficult-to-understand concepts.

[QUOTE=adamdea;710373]I am no expert but I think the answer is something like this......

[SBGK]

I was talking about many 'out of print' recordings that sadly seem available only in mp3 format. Publishing houses don't seem to think it's worth their while to invest in reissuing many culturally and musically important recordings from the '60s and the '70s, so we're now dependent on the kindness of the community to rescue those obsolete recordings and share them (they usually do that by converting old out-of-print LPs to digital). For some reason, I'm seeing some of those being offered only in mp3 format, but hey, I'll take it. Better lossy than nothing.

most vinyl transfers or concert recordings I have seen are 24/96, 320 vbr sounds ok on my ipod touch though.

[adamdea]

Well, I didn't think it was possible to answer that question in just a few paragraphs without resorting to esoteric math and engineering, but Adam's description is really excellent. Anyone interested in digital audio should read and try to understand what he wrote - it's an excellent summary of some difficult-to-understand concepts.

You are very kind. But the last bit (about reading some books) is the most important bit. I banged my head against a table for ages trying to get this stuff. But Phil Leigh (who I really do hope will return to this forum) patiently explained some of it too me along with Werner who posts on Hifi Wigwam and Pinkfish; eventually I realised that there was nothing for it but to teach myself using some text books.

[michael123]

44.1/16 and dimensions of the disc were chosen just to keep 74min of audio for the Wilhelm Furtwängler's recording of Ludwig van Beethoven's Symphony No. 9 from the 1951 Bayreuth Festival..

BTW my design gurus for digital are Ed Meitner, Daniel Weiss, Andreas Koch..
Weiss MAN301 is so yammy Hopefully in couple of years I will retire my Transporter..

There is similar paper on Meridian website (I think) that states that the optimal is 88.2/20 (and I personally agree)

[adamdea]

44.1/16 and dimensions of the disc were chosen just to keep 74min of audio for the Wilhelm Furtwängler's recording of Ludwig van Beethoven's Symphony No. 9 from the 1951 Bayreuth Festival..


There is similar paper on Meridian website (I think) that states that the optimal is 88.2/20 (and I personally agree)

I think you mean http://www.meridian-audio.com/w_paper/Coding2.PDF
It's a really good summary of the issues. However his argument about 16 bits is pretty subtle stuff about possible degradation due to repeated redithering. He seems really to accept that 16 bit is quite enough in itself.

I have heard the beethoven 9 story, but i'm not sure it's true. IIRC There were other reasons for picking 16/44.1 and frankly the arguments as to why it is good enough are pretty cogent. I agree though that the 44.1 seems a bit too close for comfort. I think it would be nice if everything were available in hi rez, but look at the claims made about 24/192. Someone will no doubt pop up saying that 32/384 is the only way. I think that the whole concept of "good enough" upsets audiophiles.

[cliveb]

I have heard the beethoven 9 story, but i'm not sure it's true. IIRC There were other reasons for picking 16/44.1 and frankly the arguments as to why it is good enough are pretty cogent.

16 bits was chosen because it was the limit of what was practically achievable at the time. If anything, it was on the bleeding edge. Philips wanted to go with 14 bit, but Sony made a stand and insisted on 16. (FWIW, I believe that 16 bit is plenty as a delivery format. 14 would probably have been good enough in most cases. Most modern rock/pop CDs seem to only use about 4 bits anyway).

44.1kHz was chosen because it fitted the line rate of both NTSC and PAL video machines by storing three samples per line. (NTSC has 245 x 60 lines/sec, PAL has 294 x 50). Video machines such as U-matics were the only recorders conveniently available at the time which could support the necessary data rates, so they formed the basis of the first digital audio recorders.

I too have heard the story about making CD just big enough to hold Beethoven's 9th - no idea if it is true. What does amaze me is that had they made the CD just fractionally bigger it would have been more than enough to hold pretty much any double LP as a reissue, whereas at the size they chose, some double albums fitted and others didn't. (And of course some doubles were reissued on two CDs anyway even though they would have fitted on one, which smacks of profiteering on the part of the record companies).