[gregorio]

Sejarzo - It would definitely make sense or at least more sense than 24/96. The exception would be if you had an ADC with good filters at 44.1. In this case there are only disadvantages in using 88.2 over 44.1: It would halve your processing power, halve your storage space, halve your transport bandwidth and introduce re-sampling errors (although fewer than resampling from 96k).

 

[sejarzo]

Does simply taking every other sample from a 88.2k stream really introduce significant errors in the resulting 44.1k stream?

 

[gregorio]

Even if the SRC was absolutely perfect, it's not going to give you anything more than if you recorded in 44.1 in the first place. As no SRC process is absolutely perfect, it will introduce errors. How noticable these errors are likely to be is completely dependant on the the ears and equipment used to listen to the music. When you get to the point with an ADC where the SRC errors are more damaging than the effects of the anti-alias filter, then there seems absolutely no point in recording at a higher sample frequency.

Bob Katz, who wrote the bible on mastering, warns in the strongest terms possible against using SRC.

 

[gyrodec]

sejarzo

Unfortunately removing every other sample isn't quite the same as saple rate conversion. Stereophile did an indepth technical article on this, and if the files has rapidly falling volumes above about 15khz, then you can get away with it and it is supposed to sound better than normal SRC. THe trouble with just removing samples is you end up with something akin to a 44khz file with saples of above 22khz in it, i.e. sample wrap-around. If the >22khz level are low enough, it all works out, if not, then it can sound awful.

 

[gregorio]

Sorry gyrodec but you can't just remove every other sample, you would just have digital garbage. There is no alternative to SRC, part of the process of SRC includes an anti-alias filter. There's no way around it, it is not possible to have any >22,050Hz frequencies in a 44.1 format file. Even if it's low level, it makes no difference only minus infinity will work, anything else breaks the rules of digital audio theory. Nevertheless, sejarzo's basic premise is correct, that the SRC from 88.2 to 44.1 is likely to introduce fewer errors than going from 96 to 44.1.

 

[12Bass]

Quote:

Originally Posted by gregorio /img/forum/go_quote.gif Bob Katz, who wrote the bible on mastering, warns in the strongest terms possible against using SRC.

If the major concern is retaining digital signal integrity, then it stands to reason that any alteration of the data stream will be undesirable. However, the relative importance of maintaining digital integrity is somewhat less for the end listener than it is for a mastering engineer. For the listener, even though mathematical errors are introduced by employing SRC, sample rate conversion may serve to mitigate the undesirable reconstruction artifacts of the digital to analog conversion process resulting in a net overall improvement in fidelity.

It seems reasonable to caution against putting too much weight on the matter of retaining absolute data integrity when the errors produced by the D/A process may in fact be more serious in nature....

 

[gregorio]

12Bass - I agree broadly with the points you made but I have two additional points:

1. Depending on your DAC, the errors introduced by SRC may be more significant than the resonstruction artifacts.

2. Doing a SRC (which will introduce errors) to avoid the errors in a poor reconstruction process seems a ridiculous solution. The solution should be to improve the reconstruction process at 44.1 so that it matches the reconstruction at 96k. This way we are reducing errors rather than deliberately creating them.

 

[gyrodec]

gregorio

I'm extra-sorry, but you really can remove every other sample in 88.1 or 3 of 4 in 176.2, with the proviso of rapidly falling volumes above 12-15khz. I have seen a paper on it, with graphs showing the acceptably low level of sample wrap-around in selected files.

I'm trying to find the Stereophile reference right now, but not having much luck. But from sampling theory, for say a simple 1Khz sine wave, it should work perfectly.

 

[gyrodec]

gregorio & all

It seems like my failing memory slightly simplified the actual process. It was invented by mastering guru Tony Faulkner, and invloves averaging blocks of 4 samples, and does have the low volumes of high-frequency energy limitation I mentioned (no rock cymbals). This averaging process creates a notch filer and so removes the need for explicitly using a filer. Here is a link to the Stereophile paper:
Stereophile: The Law of Averages

Having reread the paper, I do think you could just remove alternate samples for a simple sine wave signal, but agree that it would not work for any normal music signal.

I also found another interesting paper by the same author (Kieth Howard) about issues with DAC filers, showing definate issues, but showing the 3 blind testers didn't find much difference and couldn't agree which ones they liked best: Stereophile: Ringing False: Digital Audio's Ubiquitous Filter

 

[12Bass]

Quote:

Originally Posted by gregorio /img/forum/go_quote.gif 2. Doing a SRC (which will introduce errors) to avoid the errors in a poor reconstruction process seems a ridiculous solution. The solution should be to improve the reconstruction process at 44.1 so that it matches the reconstruction at 96k. This way we are reducing errors rather than deliberately creating them.

This seems to me to be an example of a disagreement based upon the difference between theory and practice:

According to sampling theory, if all potential sources of error in the data stream are eliminated then the D/A conversion provides a perfect reproduction of the original analog waveform. However, in practice, even if we can achieve a bit-perfect data stream, D/A converters will inevitably introduce phase shifts (and other anomalies) due to the requirement of steep low-pass filter.

There are various sources of error in the A/D -> D/A process. What I'm suggesting is that achieving the highest realizable level of fidelity involves mitigating the largest sources of error, wherever in the chain they might occur - be it in the data stream, the D/A reconstruction, or both.

While I agree that well-designed DACs theoretically should not introduce audible reconstruction artifacts, I've yet to hear one which doesn't. I'm pretty sure I can hear the effects of the low-pass filter on my modified Echo Gina24 and RME DIGI96/8 PST. For comparison, I'd be curious to listen to a professional, mastering-level DAC to see how well it performed at 44.1 kHz. Perhaps you are correct that true high-end DACs don't exhibit this problem to the same degree.

 

[b0dhi]

Quote:

Originally Posted by gregorio /img/forum/go_quote.gif b0dhi - 44.1 is not inherently worse that 96k, where on earth did you get that piece of information from?

All else being equal, a recording system working at a higher sampling rate will always be superior to one working at a lower sampling rate. This is basic sampling theory. The fact that a sharp filter must be applied to record at 44.1khz while avoiding aliasing is unavoidable. How good the 44.1 filter is is irrelevant, there are theoretical limitations on how good a filter can be, and the fact that the 44.1k filter must be so sharp is a huge impediment that does not hinder the design of anti-aliasing filters used in higher sampling rates. See here.

 

[Apocalypsee]

I want to ask, does the uses of oversampling DAC means that we don't need to upsample it? Like this AK4396 have 128x oversampling and Digital de-emphasis for 32, 44.1, 48kHz sampling (I don't know what that does though)

 

[gregorio]

b0dhi - "All else being equal, a recording system working at a higher sampling rate will always be superior to one working at a lower sampling rate. This is basic sampling theory."

No it's not, where did you get that idea from? The Wiki article you linked to certainly did not make this assertion. I have been dealing professionally with hi-res digital audio since 1992 and although I have seen the odd article which states that sample rates greater than 44.1kFs/s are intrinsically better, those articles have never provided any proof. The more reputable articles state (and prove) the opposite, see AES publications. Sampling theory in fact states that all the frequencies that can be heard by a human being can in theory be encoded perfectly at 44.1k. Sampling (Nyquist) theory also states that to re-create a perfect image of the waveform, two samples per waveform are required, hence why the sample frequency has to be twice the audio frequency. Having more than two samples per waveform is not going to allow for a more perfect re-creation of a waveform, it's just going to allow for the encoding of higher audio frequencies.

The same is true of bit depth, 24bit is not going to give anymore quality than 16bit, it's just going to allow for the encoding of a greater dynamic range, approx 6.02dB for every additional bit, which for the consumer is spurious anyway.

Your argument about the anti-alias filter at 44.1k would have been valid 15 or so years ago but 256x oversampling in modern ADCs make it much easier to implement the filter at 44.1k and while it is not possible to implement a perfect filter, it is possible to get close enough for there to be no decernable artefacts. Of course anti-alias filters have to be implemented whatever the sample frequency but I agree that it is more difficult to create a good filter at 44.1k than at say 96k.

The bottom line is that providing high quality professional ADCs are used, there is no way that a human being could tell the difference between 16/44.1 and 24/96. The important part here is a high quality ADC, mine cost me nearly $10,000. At that level it's not possible to tell a difference with a finished mix between 44.1k and 96k. There are even more expensive ADCs out there (Prism, et al) and at 44.1k they actually outperform cheaper professional ADCs running at 96k.

 

[IPodPJ]

Quote:

Originally Posted by gregorio /img/forum/go_quote.gif The bottom line is that providing high quality professional ADCs are used, there is no way that a human being could tell the difference between 16/44.1 and 24/96. The important part here is a high quality ADC, mine cost me nearly $10,000. At that level it's not possible to tell a difference with a finished mix between 44.1k and 96k. There are even more expensive ADCs out there (Prism, et al) and at 44.1k they actually outperform cheaper professional ADCs running at 96k.

I don't buy that for a second. Analog mediums such as vinyl have no sample rate and the human brain has a much easier time discerning a sound that is more natural and lifelike. Every DAC I've ever listened to has always sounded more natural and lifelike when the audio was resampled to 192kHz vs. standard redbook 44.1kHz.

However, you are accepting a trade-off when you resample.... you will lose/have distorted micro amounts of detail even on the best upsampling DSPs (Anagram comes to mind), but that is usually not detectable with higher end DACs and upsamplers. What is detectable is how much more lifelike and "analog" the audio sounds, i.e. spatial information/placement of the instruments and musicians on stage, airiness and texture, seemless left & right channel integration, liquid transitions between notes and more apparent decay of instruments, greater ease hearing all of the vocals, etc.

 

[IPodPJ]

Quote:

Originally Posted by Apocalypsee /img/forum/go_quote.gif I want to ask, does the uses of oversampling DAC means that we don't need to upsample it? Like this AK4396 have 128x oversampling and Digital de-emphasis for 32, 44.1, 48kHz sampling (I don't know what that does though)

The AK4396 is one of the best DAC chips on the market. If I were to build my own DAC, I would definitely use that chip. It has better bass detail and accuracy than any top model Burr Brown chip I've listened to. I now use the following chain and it sounds better than a $5,500 DAC I had borrowed.

-- PC fed with power from Virtual Dynamics Master LE 2.0 Power Cord
-- Foobar 0.9.5 beta2, running SSRC resampler at 192kHz/Ultra mode, Dither on
-- Cobalt Cable analog interconnect pair, from the Juli@ to my amp