more info on CD upsampling...please

[Unregistered]

I am seeing some DVD players available that offer CD upsampling from the standard 44.1khz to frequencies as high as 176.4khz. I am pretty sure this does not raise regular CDs to meet the quality of DVD-A or SACD. However, there must be some benefit (hopefully not just marketing). I am still a little confused on how this will help. The CD will still be a 16-bit source, right? What are your experinces with CD upsampling? If it will make my regular CDs sound better...I am all for it. Please shed some light on this issue. Thanks

 

[Unregistered]

The main benefit to upsampling, and many believe the ONLY benefit, is that the design of the reconstruction filter is much easier.

As you said, cds are 44.1/16. Sampling at 44.1khz theoretically gives you a good reproduction of frequencies up to 22050hz. But before the DAC, the waveform has to be filtered. Reconstruction filters are not brickwalls, just like crossovers on speakers are not. The filter will not exactly cut off at 22050 so some of the frequencies below that will get attenuated. Since those frequencies are now in the audible range, the sound can be degraded.

If you upsample to a much higher rate, like 192khz, now the filter can start working at 96khz and it would have to be incredibly poor to affect frequencies in the audible range down below 20kHZ.

Just how much better upsampling a waveform originally recorded at 44.1khz is has been debated to death. Even many well known mastering engineers aren't fully convinced it's all that important. Google for Bob Katz or read his book: The Art and Science of Mastering Audio. Fairly technical but a good read if you are into digital audio.

 

[docferdie]

another analogy would be resizing a 1 megapixel image in photoshop to become 8 megapixels. You have a bigger file but you don't have better image detail. You can't see or hear detail that wasn't present to begin with no matter how much you upsample or resize.

 

[Rip Van Woofer]

"Unregistered" basically has it right. Here is the straight poop on the subject of upsampling (and other digital audio matters) in an easy to read form.

 

[Unregistered]

They say it's time...

Or rather, the time domain. A filter which has to cut off totally between 20 and 22 kHz must be very steep. It has a lot of ringing (evil) and likely also pre-ringing (not found in nature, and therefore particularly evil, according to some schools of thought).

A filter which has to cut off totally between 20 and 44 kHz (or 88 or 96) can be much less steep, and therefore have less evil effects in the audio band.

For an interesting and different viewpoint, see http://www.mlssa.com/pdf/Upsampling-theory-rev-2.pdf

It could also be that machines with oversampling tend to be better designed than you average run-of-the-mill player.

 

[Audioholics]

New Article: Upsampling vs. Oversampling for Digital Audio

Here's our take on the whole issue - an article just completed by one of our new contributing writers, Nauman Uppal.

Upsampling vs. Oversampling for Digital Audio

 

[Unregistered]

So upsampling and oversampling will both yield about the same frequency response benefits, right? Upsampling also reduces the jitter artifacts by spreading it over a wider spectrum and thus reduce its debatably audible effects. So it seems to me that while upsampling's benefit is debatable and over-hyped, there's nothing wrong with it to degrade the sound. So it has a place among audiophiles.

So will upsampling a 16/44.1 wav file to 24/96 sound better? Maybe a more concrete question is, will a VOB file containing 24/96 wav files played in a DVD player sound better than a 16/44.1 version of the same sample played on a CD player?

 

[UMD_terp]

There is nothing wrong with it as long as it is implemented correctly...

As for your second question, it all depends. Is the DVD wav file originally sampled at 96kHz, while the CD wav file is sampled at 44.1kHz? The benefit of having more samples does help with quantization noise and allows the use of 'better' filters. Simply taking a source file that has sampled an analog waveform at 44.1kHz, and then upsampling it may improve it... will it be noticeable? I don't know. If we have a 44.1kHz sampled signal, and then interpolate it to a higher rate and follow that with an interpolation filter, it may help cut out some quantization noise and jitter. The signal may be cleaner, but there is no more information in this new signal than the original...

 

[Unregistered]

I would like to add a correction/addition to the explanation given about oversampling and upsampling in this article. The description of oversampling given here confusingly refers to oversampling during the A/D conversion processes in which samples are actually taken from the analog signal. Unfortunately the Oversampling/Upsampling discussion refers to oversampling digital filters that are used in the digital domain preceding D/A conversion. The various oversampling strategies utilize for instance 'zero filling' and 'sample and hold' (basically repetition of each same sample for 4 or 8 times in a row). This is then followed by low pass digital filters (FIR filters) which effect various kinds of interpolation depending on their sophistication. Finally the newly 'oversampled' digital data is sent to the D/A followed by analog low pass filtering.

Upsampling is identical to Oversampling if the sampling rate is integral (4x, 8x etc...). The use of asynchronous sample rate converters which allow non-integral resampling (e.*. 44.1kHz to 192kHz) is what birthed the marketing distinction between upsampling and oversampling. They are really just the same thing. As always the latest generation of filters/dacs are always better than the previous generation and what better way to help sales than rebrand your technology.

 

[Rob Babcock]

I'd say the new generation of DACs (and almost everything else) is almost always cheaper but not necessarily better. Most advances in electronics seem more aimed at the bottom line than the high end.

 

[Unregistered]

Yes there are new less expensive DACs which are not necessarily better, but the top end digital filters and DACs are much better than anything available only a few years ago. E.*. top line BB and Analog sample rate converters have amazing THD and S/N numbers as do the Wolfson 8740 and BB 1738 DACs. Neither are they inexpensive.

 

[Rob Babcock]

I'm not really sure just how much audible improvement you'll gain my reducing the already-very-low THD & SN of the average DAC. Don't get me wrong; I'm always in favor of higher performance, but I think it's pretty well established that fairly high levels of THD are inaudible (certainly 1/2 of a percent is about the threshold for most freqs, while in the low bass double digit THD is often imperceptible- I'm sure there's an egghead here that has the exact figures or a link ). Likewise, the extra S/N might be lost on the average listener.

That's not to say they couldn't sound better, just that the improved SQ probably isn't related to the measurements usually listed to justify it.

 

[WmAx]

In controlled listening tests, 15-25% THD was the JND range for the band under 100Hz. As for above, this is far too variable to quote single values; it is highly dependant on the material being played. However, an average of 1% THD is a generally accepted average undetectable level for must program material. The rate lowers with pure sine waves. You would really need to read an article on the issue that summarizes several perceptual research studies on the subject:

Just Detectable Distortion Levels
Wireless World, 1981, February, pages 32, 33, 34 and 38.
James Moir

The already extraordinarly low THD and IMD products from average ordinary DACs is not anywhere near levels that could be identified in music program material.

-Chris

I'm not really sure just how much audible improvement you'll gain my reducing the already-very-low THD & SN of the average DAC. Don't get me wrong; I'm always in favor of higher performance, but I think it's pretty well established that fairly high levels of THD are inaudible (certainly 1/2 of a percent is about the threshold for most freqs, while in the low bass double digit THD is often imperceptible- I'm sure there's an egghead here that has the exact figures or a link ). Likewise, the extra S/N might be lost on the average listener.

That's not to say they couldn't sound better, just that the improved SQ probably isn't related to the measurements usually listed to justify it.

 

[Rob Babcock]

Egghead to the rescue! (I mean that only in jest- I couldn't remember the actual value, but I know it's low. Thanks for the link).

I get a kick out of people who tout one inexpensive receiver over another based on it having a few thousandths of a percent lower THD! As Chris has pointed out, the THD of your basic mass market black box is pretty low. If they sound crappy some other factor is at work.

 

[WmAx]

Or rather, the time domain. A filter which has to cut off totally between 20 and 22 kHz must be very steep. It has a lot of ringing (evil) and likely also pre-ringing (not found in nature, and therefore particularly evil, according to some schools of thought).

A filter which has to cut off totally between 20 and 44 kHz (or 88 or 96) can be much less steep, and therefore have less evil effects in the audio band.
.

Yes, but those 'schools of thought' are not based one credible auditory studies, now are they? What is the basis for those speculations? Let's take a look at the landmark peer reviwed perceptual research that established the steep filter at frequencies required by 44.1khz sample rate was inaudible -- in fact, even lower was inaudible:

What Bandwidth Is Nescarry For Optimal Sound Transmision
Plenge, Jakubowski and Schone
JAES, 1980, Volume 26, Number 3, Pages 114-119

I would also refer you to a later paper on the audibility of anti-alias filters:

Perception of Phase In Anti-Alias Filters
Pries and Bloom
AES Preprint: 2008


-Chris

 

[mtrycrafts]

The main benefit to upsampling, and many believe the ONLY benefit, is that the design of the reconstruction filter is much easier.

As you said, cds are 44.1/16. Sampling at 44.1khz theoretically gives you a good reproduction of frequencies up to 22050hz. But before the DAC, the waveform has to be filtered. Reconstruction filters are not brickwalls, just like crossovers on speakers are not. The filter will not exactly cut off at 22050 so some of the frequencies below that will get attenuated. Since those frequencies are now in the audible range, the sound can be degraded.

If you upsample to a much higher rate, like 192khz, now the filter can start working at 96khz and it would have to be incredibly poor to affect frequencies in the audible range down below 20kHZ.

Just how much better upsampling a waveform originally recorded at 44.1khz is has been debated to death. Even many well known mastering engineers aren't fully convinced it's all that important. Google for Bob Katz or read his book: The Art and Science of Mastering Audio. Fairly technical but a good read if you are into digital audio.

However, all CD players oversample already just for this reason, from almost day one. So, this upsampling is yet another gimmick from alternat audioworld.

 

[mtrycrafts]

another analogy would be resizing a 1 megapixel image in photoshop to become 8 megapixels. You have a bigger file but you don't have better image detail. You can't see or hear detail that wasn't present to begin with no matter how much you upsample or resize.

And, you cannot hear stuff that is already beyond the capability of detection

 

[mtrycrafts]

It could also be that machines with oversampling tend to be better designed than you average run-of-the-mill player.

Except that oversampling has been the norm, as I indicated, from very early on; first 2x, then 4x and 8x, some higher. This used to be an integral part of advertising the CD player of yesteryears.
As time passed, it dissappeared from the ads but not from the players. Today, the marketeers needed a new twist to satisfy the audiophile community who needs the latest and greates every 6 month. I shudder to think what tomorrow holds. It certainly will not be the facts.