[Ricky Monk]

Now.. I'm a technical guy.. and I generally understand all these terms.. but I keep hearing people talk about CD players that can upsample, and look at this as a feature.

Now, I can see how this could be useful for interconnecting with other equipment.. but what I *can't* see is how it could possibly improve the sound quality. The original data is 44.1Khz, 16 bit PCM. The most you can get is a perfect upsampling, which would sound exactly the same.. more likely, you will get some kind of degradation (whether you can hear it or not).
The only question should be, if you need upsampling, which device can do the best job for you.. correct?

Am I totally misunderstanding something here, or is this just marketing hype coupled with the odd person who bought into it?

 

[radrd]

It doesn't seem to be marketing hype, but it doesn't seem like anyone can really explain exactly why it does sound better, just how it works and the fact that it does sound better.

This link might help a bit:
http://www.aslgroup.com/dcs/upandover.htm

 

[jude]

I've used two upsampling DACs for review, and both sound very good -- I'm buying one of them. I don't really know what upsampling is relative to oversampling, and I hope to have one of the manufacturers of the DACs I'm reviewing answer some questions from me for the review article regarding any differences between upsampling and oversampling (if any).

At the end of the day, here's what I can say: the two upsampling DACs I used both had some sonic traits in common, despite the fact that they used different sample rate coverters. The things they did in common included significantly improving soundstage, image layering and placement, as well as to enhance overall resolution relative to almost all other redbook players I've had in my rig that weren't "upsampling" players. The Creek CD53 (a fantastic one-box player that isn't an "upsampling" player) was as detailed as the two upsampling DACs, but without the level of soundstaging and image layering presented by the two upsampling DACs. Long story short, the two upsampling DACs had sonic traits in common that none of the non-upsampling gear I've used shared with them. I'm inclined to think there's something different going on, but I'm hoping at least one of the manufacturers can clue me into what that is through a Q&A for the review.

 

[JaZZ]

Some say over- and upsampling is the same. But usually upsampling implies sampling rate conversion, while oversampling actually doesn't, just calculates some sample values between the original samples by interpolation. And that seems to be the cause for the sonic differences. But there's obviously no explanation for them.

Jude, I guess the two mentioned upsamplers are the MSB GoldLINK III and the Bel Canto DAC 2. Well, I have ordered (and prepaid) the latter – not least inspired by your review – from the French distributor (there's no one in my country), and on Saturday morning I was going to collect it at the local post office – but the parcel is missing!

 

[jude]

[size=xx-small] Quote:

Originally posted by JaZZ ....Jude, I guess the two mentioned upsamplers are the MSB GoldLINK III and the Bel Canto DAC 2. Well, I have ordered (and prepaid) the latter – not least inspired by your review – from the French distributor (there's no one in my country), and on Saturday morning I was going to collect it at the local post office – but the parcel is missing!

[/size]Well, good luck finding the parcel, man. When you do, I think you'll be very pleased with what's inside (once it's plugged into your rig, that is -- it ain't much to look at).

 

[Ricky Monk]

Okay. So the only benefit is really some kind of sound coloring due to the upsampling algorithm combined with the high frequency DACs used.

I rank that in there with tubes. It might bring some sounds out in some cases.. but that doesn't necessarily mean they were meant to come out (even if it sounds good. But if it sounds good, who cares, right?)

 

[jude]

[size=xx-small] Quote:

Originally posted by Ricky Monk Okay. So the only benefit is really some kind of sound coloring due to the upsampling algorithm combined with the high frequency DACs used. I rank that in there with tubes. It might bring some sounds out in some cases.. but that doesn't necessarily mean they were meant to come out (even if it sounds good. But if it sounds good, who cares, right?)

[/size]I'm not the guy to answer this question beyond saying that the two upsampling DACs I've heard sounded very good.

John Atkinson, in a fairly recent Stereophile did perform some measurements of the Musical Fidelity A324 upsampling DAC. You might want to see if you can find that issue. Whether or not it's helpful information for you is for you to decide after you've seen it.

 

[AndreYew]

Upsampling and oversampling are identical. The only difference in implementation is that most upsampling is asynchronous, which means any jitter in the incoming stream is translated into noise or other upsampling artifacts, whereas oversampling is usually done synchronously, which means it's not affected by jitter, and it can be (but not necessarily must be) more accurate.

The interpolation necessary is the same for both: low-pass filter the sampled datastream to prevent aliasing in the new sample rate.

--Andre

 

[JaZZ]

Andre...

...I think it's not exactly the same: Interpolation with oversampling can be linear, and that's probably even the standard process. While non-integer upsampling comprises a sine-wave function algorithm. But even if you let oversampling use the same algorithm, the two resulting curves (after the compulsory low-pass filtering) won't be identical, I guess.

 

[Anders]

JaZZ,
One of my hypothesis is that 16 bit sampling with a low sampling frequency is not accurate enough (why should they otherwise had gone to 20 bit and higher samling rates in pro audio long ago?). I assume that this gives a litte edgy curve of the musical event. That could explain the harshness that many hear on CD, especially if the A-D and D-A processes are not perfectly implemented, as we know they are not.
Could the upsampling conversion smooth out this? The interpolation should result in random changes that smooth out somewhat, and it might be possible to find algortithms that to some extent compensate for the errors in the original sample.

 

[Pepzhez]

Upsampling IS marketing hype. All CD players essentially upsample anyway. What really makes these upsampling devices work is noise-shaping; the upsampling process in itself does little or nothing.

Here's an article on the subject that explains it better than I can:

http://www.iar-80.com/page21.html

(I know from experience that posting an IAR article on this and other forums always results in controversy, as IAR is a very considered, technical magazine which more often than not draws objective conclusions that most audiophiles really don't want to hear. However, I trust them, as they really know what they are talking about and are not beholden to corporate or advertisers' interests. I trust them much more than I would a so-called "audiophile" mass-market magazine or SACD cheerleaders on net forums.)

 

[Gergor]

The way I understand is. Digitally you don't gain by upsampling, but you benefit from the design of the analog backend. You see, with the 16/44.1 data, the Nyquist frequency is at 22.05k, which is right near the upper end of the audio band, so ideally you need a brickwall filter, which is difficult to do in analog, and the phase shift starts inside audio band, which degrades (or color) sound quality. By upsampling, you can push the nyquist frequency up, which makes analog filter design a lot easier.

 

[Gergor]

Quote:

Originally posted by Pepzhez Upsampling IS marketing hype. All CD players essentially upsample anyway. What really makes these upsampling devices work is noise-shaping; the upsampling process in itself does little or nothing.

Hmmm, I missed your post when I made my last post. Anyways, I haven't read the article yet (It's long but I may read later). However, I'd like to respond to your comment.

Most (if not all) CD players use oversampling, but they only have up to 16-bit of resolution, while the latest upsampling DACs have up to 24-bit of resolution, so they are not exactly the same. What I really don't understand about your comment is about noise-shaping and upsampling does not matter: According to my limited college education, the reason for noise shaping (using integrators and feedbacks) is to reduce the amount of oversampling necessary to achieve certain resolution. Your statement is contradicting that...

 

[halcyon]

Pepzhez,

upsampling is not all hype.

If it was, why do all the professional DACs and a lot of communication devices use it as well? If it was, why do good ASRC DACs measure better than non-oversampling ones?

You are right that most CDs since the beginning of 90's upsample.

What they do is synchronous umpsampling, that is they resample at the exact multiple of the incoming signal.

This allows for some nice additional digital filtering (what in the IAR article is called, IMHO a bit simplified 'noise shaping') at higher frequencies than the original nominal sampling frequency would allow.

However that is not all.

Most modern upsampling devices that are advertised as such are in fact asynchronous umpsamplers.

They do not upsample to some exact multiple of the original sample rate.

By doing ASRC these devices (esp if it's a DAC) they can significantly reduce the jitter of the incoming signal. By reducing jitter just before the actual DA IC (and using a stable clock to feed the DAC) the device can significantly reduce THD and lower the noise floor.

Again, this needs to be implemented properly and with certain type of digital filters to ensure it works.

Additionally, not all modern upsamplers (whether synchronous or asynchronous) stop there.

Some use (like in the slightly simplified IAR article is stated) different kind of anti-alias image filters to remove the upsampling influenced hypersonic noise.

Removing this noise is essential to reducing intermodulation components in the latter audio reproduction chain, especially loudspeakers that can intermodulate the hypersonic signals back to the audible band and cause audible problems.

In addition, some upsampling DACs use filters that are NOT phase linear, but minimum phase filters. With these you can actually pass a 'perfect square wave' through the DAC without distorting it.

Those who have heard this kind of a DAC swear by it, even though it's not mathematically as correct as a minimum phase filter (at least, not in the frequency domain).

So, as you can see, it's not all 'snake oil', even though most modern cd players are (syncronous) upsamplers. I agree that many claims are actually misleading, but there are actual benefits as well, if the ASRC and digital filtering is done properly.

Regards,
Halcyon

 

[AndreYew]

Jazz,

Quote:

...I think it's not exactly the same: Interpolation with oversampling can be linear, and that's probably even the standard process. While non-integer upsampling comprises a sine-wave function algorithm.

Oversampling interpolation is not linear, and it cannot be, otherwise it will not look at all like a brickwall filter.

--Andre