When is upsampling a bad thing?

[chillysalsa]

I've noticed people talking about the Philips 963SA, and saying that for poorly recorded CDs, they prefer 96K upsampling, while newer stuff sounds best on 192K.

Without having experienced upsample, am I right in assuming that it's basically anti-aliasing the sound with a similar 'smoothing' effect (see the attachment).

So why would this be bad for some discs? Does it seemingly take away detail?

 

[elrod-tom]

I seem to recall reading somewhere that older, more flawed CD's will be exposed by upsampling. I know that I get clicks and pops on some CD-R's when using my GW Labs DSP with upsampling on. If I turn it off, the problem goes away. Not sure exactly why this is....

 

[Wodgy]

FYI, upsampling is completely unlike anti-aliasing. The main point of upsampling is to make the digital filter (part of the DAC) easier to design. When you upsample, you can inject zeroes or random data into the data stream, it doesn't change the end result. Strange but true; you'll need to read a book on signal processing to understand why things work that way. Anti-aliasing is different... if you filled in every second pixel automatically with black, an antialiased image would look terrible.

 

[chillysalsa]

elrod-tom:
I too, have noticed a couple of CDs that I had burned from MP3 (oh the heresey, I know!) have clicks and pops on them. I never noticed with my other CD players, but I'm now using a Adcom GDA-600. It's not upsampling, but the 8x oversampling with 20-bit filter has brought these clicks out much more. So maybe the affect is simlar with upsamplers...

Wodgy:
Thanks for the info. Yes, I had also read about the benefit of pushing the Nyquist tone to higher frequencies, and allowing a 'gentler' filter to be used rather than the 'brick wall' ones of the past.

I got the impression that anti-aliasing (or something like it) is common to upsamplers - something I read in an article by Paul Bergman in an article for UHF magazine..

 

[morphsci]

Right now there is a lot of disagreement about the difference between oversampling and upsampling. My understanding is that oversampling is a D to A procedure but upsampling is D to D. Essentially they both seem to have a similar effect, in that they are both interpolation methods. In both cases, there is no real increase in resolution because the amount of real information is the same. Therefore, a 44.1 kHz CD upsampled to 96 kHz will still have less real information than a disk originally mastered at 96 kHz.

In my limited experience with upsampling it increases my ability to localize instruments in space (this is on a speaker system). My guess is that this is because it is allowing better resolution of finer details by allowing shallower filtering of the output. Thus on poorly mastered/recorded CD's you get better resolution of the poor mastering and/or recording.

Of course my take on this could be off, so if someone can correct anything I said, please do.

 

[JaZZ]

Upsampling is part of the antialiasing filter, as is «oversampling». Both are just variants of the same procedure: creating (intrapolating) sample points between the existing 44,100 ones per second. Common oversampling does exactly that with an integer multiple (4x ... 16x), whereas «upsampling» creates a non-integer multiple of them, actually without using the existing ones further. With both variants the unwanted high frequencies resulting from the square form (steps) of the unfiltered analog signal after digital-to-analog conversion are reduced by up-/oversampling due to the smoothed signal with finer steps. This allows a flatter analog filter at a higher frequency. Oversampling sometimes is also called «digital filter», since the interpolation algorithm which creates the smoothed curve represents itself a low-pass filter.

Both combined, up-/oversampling and analog filter, have the same function like a non-oversampling analog filter: to prevent aliasing (the virtual reflection of frequencies higher than ½ of the sampling rate on the sampling frequency towards audible frequencies).

Upsampling and oversampling are virtually the same – but surprisingly they don't sound the same.

 

[aeberbach]

wodgy, would I be right in saying that CD upsampling is like scanner interpolation?

 

[David Mitchell]

Informative post, JaZZ. For more on the topic I recommend John Atkinson's article "Upsampling or Oversampling?" at http://www.stereophile.com/showarchives.cgi?344; he says that upsampling and oversampling are essentially the same, and that the "sonic differences ... are due to the different choices in digital filters ... with respect to the number of taps, passband ripple, and stopband rejection, and to changes in the jitter performance." The Madrigal whitepaper which Atkinson references is also interesting. Someday I'll come to understand it all.

A note on that image attached to the first post: the upper text is anti-aliased (it would look fairly smooth at the normal size), while the lower text is additionally blurred. There's a little difference. Basically, you could think of anti-aliasing as being blur applied in a focused way along curves, which can't be represented properly on our rather low-resolution displays.

 

[chillysalsa]

Thanks all, very informative.

I'm guessing that the way these technologies 'bring out' lower detail lets you hear improved soundstage and imaging (even though they are not *adding* detail).

A note on the picture: I made it by taking a 100 pixel image, and upsampled it to 600 pixels without anti-aliasing (above) and WITH anti-aliasing (below). That UHF article I found does a similar analogy with the Mona Lisa picture.

 

[Orpheus]

...and there's a third possiblity not explored yet here: upsampling does nothing for the sound.

 

[morphsci]

Quote:

Originally posted by Orpheus ...and there's a third possiblity not explored yet here: upsampling does nothing for the sound.

Unfortunately, that is not the case to my ears. I haven't done any dbt but the difference toggling the upsampling on and off on my Philips is pretty noticeable and as I stated, not always for the better. JMO.

 

[lini]

aeberbach: Yes, that's a good example. But usually scanner interpolation factors are integer numbers only. So I'd wonder, whether 88,2 or 176,4 kHz wouldn't be more ideal for CD audio than 96 and 192 kHz, by the way. Anyway, following JaZZ' definition, which is probably the most common over here, scanner interpolation would in most cases be over- instead of upsampling.

A nice visual example for the effects of integer and non-integer multipliers are TFT displays. Try a native 1600 x 1200 pixel model. Fed with 1600 x 1200 input pixels by the graphics card, the picture should look smooth and crisp - that would be similar to hi-res audio like DVD-A or SACD. Fed with 1024 x 768 pixels, the picture should look less detailed, but also smoother due to the non-integer multiplier. Fed with 800 x 600, the picture should look even less detailed, but crisp, because the multiplier is an even number.

Of course, the example is not that good, as long as we don't also bring bit depth into play - as most modern upsamplers are 24 bit, too. That would be more like a scanner interpolating its (theoretically insiting for this example) 275 dpi / 24 bit CCD to 1200 dpi / 32 bit output.

So upsampling in fact usually involves an uneven multiplication of samples plus bit depth increase plus interpolation.

Greetings from Munich!

Manfred / lini

P.S.: Edited for being dumb. I usually have a good picture of things in my head, but I tend to mess up verbal explanations.

 

[JaZZ]

Manfred...

...good analogy! I'm not completely opposed to the idea that the audible (!) effect of upsampling consists of a blurring with high fequencies, thus a masking of digital artifacts such as ringing due to the sharp low-pass filter – maybe the rounding errors arising from non-integer sampling-rate conversion produce some specific harmonic distorsion which do so. Nevertheless I am pleased with the sound created by upsampling and don't care too much how it comes about.

 

[jpelg]

Quote:

I'm not completely opposed to the idea that the audible effect of upsampling consists of a blurring with high fequencies, thus a masking of digital artifacts such as ringing due to the sharp low-pass filter

I don't doubt that you hear a difference. But it sounds like an equalizer could have the same effect, in the end, to me. How is this giving you truer sound to the original source

?

 

[JaZZ]

An equalizer emphasizes or attenuates certain frequency ranges. The subtle sonic differences between upsampling and none don't sound like this at all. Don't misunderstand the comments regarding the incompatibility of upsampling with some bright recordings – in fact it doesn't generally make the sound brighter, just different, particularly influencing the perception of space.

How is this giving you truer sound to the original source? That's not necessarily the case, but it sounds a bit more natural and less technical to me. As others have stated: the diference is similar (though in a clearly minor degree) like after switching from CD to SACD (which isn't generally brighter than CD, too).