[Redbeard]

Interesting discussion, but I'm confused.  I was under the impression that the process that Oxia described is known as oversampling, which is different than upsampling.

This link has a pretty good explanation of the difference between the two:

http://www.simaudio.com/pdf/Upsampling.pdf

Any thoughts?

[Hantra]

Seems like there are only a few guys who have an opinion on this, and it's always different.  

I disagree with Sim's paper because they state that filtering is required unless you upsample at a high rate.  There are many non-oversampling, non-upsampling, filterless DAC's on the market.  And they are all worlds better than the alternative.  The best upsampling I ever heard was a $30K Accuphase setup.  But for long term listening, I'd hide my non-oversampling DAC behind that pretty gear. . .  

B

[JohnR]

I believe the theory still says that the filtering is necessary to recover the original signal. If you have any pointers to a technical explanation of why it isn't I'd be interested to see them. My first guess would be that the combination of relative lack of energy in the higher "audible" frequencies combined with the ear's rolloff (and possibly, rolloff in the rest of the playback chain) makes the high frequencies in a non-filtered signal a non-issue, audibly speaking.

JohnR

[Hantra]

That's right John. . .

There are many theories, and one of them suggests that just because we can see it on the scope does not mean that we can physically hear it.  I believe that when the first such DAC's came out with no filters, this was the guiding theory behind them.

[eico1]

the question to ponder is, is it the lack of filters that make us like these dacs or is it the addition of gross amounts of hf images causing our pre-amps and amps to generate non-harmonic distortion!

steve

[Hantra]

is it the lack of filters that make us like these dacs or is it the addition of gross amounts of hf images causing our pre-amps and amps to generate non-harmonic distortion!

Well, try one out for a week, and then try letting it go.  Then that question won't matter. . .

 

B

[Redbeard]

Sheesh, I'm confused again!  This time it has to do with the use of the term 'filter'.

Every DAC, whether oversampled, upsampled, upconverted, or just plain vanilla will have an output that is a step waveform.  If I'm not mistaken every DAC, no matter what flavor, is followed by an analog filter to smooth out the step function.  

OTOH, the term 'filter' is also used to describe the digital interpolation performed by the DSP in an oversampled, upsampled, or upconvertd DAC.

To me, the best way to keep this sorted out is to use the term 'analog filter' (which they all have) and 'digital filter', which only some have.

Oh, here is another interesting link.

http://members.chello.nl/~m.heijligers/DAChtml/Digital%20Theory/Digital%20theory.html

Cheers

[JohnR]

There are many theories, and one of them suggests that just because we can see it on the scope does not mean that we can physically hear it.

I think I just said that. I'm interested in something a little deeper, if it exists.

eico1, interesting thought!

[ehider]

Based on my experience I totally agree with Hantra on upsampling.

In a nutshell: Upsampling probably causes as many problems as it trys to solve due to the filtering. Based on my extensive listening to very expensive DACs (some with and some without upsampling), I've come to the conclusion that upsampling can create as many problems as it attempts to solve. Additionaly, the DACs that I thought were the most harmonically pure and the least "grainy" sounding were the ones without upsampling!

I have heard DAC's with upsampling that sounded excellent, but NONE that sounded superior and unique due to this single topological difference.

[nathanm]

Oxia:  Sorry if I wasn't entirely clear.  I am indeed referring to 'bicubic' interpolation of an image and not 'nearest neighbor'.  The former does indeed add information to smooth things out whereas the latter simply scales up the original pixels.  Neither method is actually getting you more objective detail, but subjectively Bicubic has the edge since it looks much smoother.

In either audio or visual, the upsampling is not creating additional resolution  from the source but simply playing with the data to create a potentially more pleasing result.  As I said, it's a grey area.  If you really want higher fidelity you have to add bits.  Anything else is a clever magic trick of sorts!

[Oxia]

Hi Nathan,

I see what you meant, and you have raised a good point regarding what interpolation can actually achieve. In the end, the information that is lost due to undersampling (as in scanning an image at a low resolution) is gone forever.  Interpolation can never really get it back. At best, it can only hope to approximate what was lost. Whether this succeeds in bringing us closer to the (original) music, subjectively or otherwise, is debatable.

Hi Redbeard,

Regarding filtering, this is something that is necessary due to the effects of aliasing distortion, which is an inescapable artifact of digital playback. The audio bandwidth is limited to half of the sampling frequency, so with the 44.1 kHz sampling rate of Redbook audio, your frequency response is limited to a maximum of 22.05 kHz, which is beyond the limits of human hearing. The problem is that when a sampled sound is reproduced, the original sound is regenerated along with images of the sound shifted up in pitch equal to (a) the difference between the frequency of the original sound and the sampling rate, and (b) the sum of the frequency of the original sound and the sampling rate. So if we are sampling a 20 kHz tone at a sampling rate of 44.1 kHz, you would get the original tone plus additional tones at 24.1 kHz and 64.1 kHz -- if no filtering was present. These high-frequency copies (or "aliases”) of the original sound, though inaudible by themselves, will intermodulate with the audible signal when played back through a loudspeaker and create spurious frequencies that are audible. This form of distortion is called Aliasing Intermodulation Distortion (AID). There’s a good paper on AID that you can read here:

www.digitalaudio.dk/technical_papers/aid.pdf

 For this reason, filtering is commonly applied to DACs to attenuate frequencies above 20 kHz in order to avoid AID. Of course, "brick-wall" filtering comes with it's own drawbacks (phase shift and consequent smearing of details in the time domain), as does up/oversampling (RFI). The bottom line is that there is no free lunch with any of these approaches. It just comes down to whether you, the listener, find the strengths of one of these approaches to overcome its weaknesses.

[eico1]

Oxia, isn't aliasing and the problems you describe only present at the a/d stage?

steve

[JohnR]

Oxia, isn't aliasing and the problems you describe only present at the a/d stage?

steve

I don't believe so. Aliasing will occur at the A/D stage if the input signal is not properly band-limited. However, the spectrum of a uniformly-sampled discrete signal with no frequencies outside this limit, is an infinite series of repetitions of the spectrum of the signal (which is within the limit).

(In other words, sampling a signal that is strictly band-limited, produces a signal with an infinite spectrum.)

When this signal is put through a zero-order hold, the spectrum is multiplied by a sinc function, but in absence of steep filtering there are still a lot of freq components present above the Nyquist freq.

[eico1]

JR, sure there are images at the d/a, but not any of the distortions mentioned until the images hit another device that cant handle the signal. The worst I'd expect are power amplifiers with limited power bandwidth. If the input isn't band limited, you can have pretty bad slewing distortion. Is that what we're talking about?

steve

[Oxia]

Oxia, isn't aliasing and the problems you describe only present at the a/d stage?

Hi Steve,

When talking about aliasing distortion, the DA and AD stages are inextricably linked. It's a little like the old question that goes: "if a tree falls in the forest and there's nobody there to hear it, does it make a sound?” The answer is no, since the falling tree only creates waves --compression and rarefaction of air, which by itself is not “sound”. Only when waves are picked by the auditory nerve of a listener, are they interpreted in the brain as sound. In other words, you cannot separate the sound from the listener. Both go together, like cause and effect.

To extend this analogy to digital audio, if you digitally sample an analog signal, is it possible to experience aliasing distortion prior to the DA stage? The answer is no. Although the cause of aliasing is undersampling at the AD stage, the effects of aliasing (distortion) is manefested only at the DA stage. Put simply, you cannot "see" aliasing distortion until you have the reconstructed analog signal and can compare it to the original source.

The following is a definition of aliasing that I found on the web. It says it better than I can.

From: http://www.atis.org/tg2k/_aliasing.html

Aliasing: In any technology or process involving (a) sampling a signal, e.g., an electrical signal or (a series of images of) a moving subject; (b) processing, storing, or transmitting representations of the samples; and (c) replicating the original signal from the representations: the production of artifacts as a result of sampling at intervals too great to permit faithful replication of the original signal

[JohnR]

To extend this analogy to digital audio, if you digitally sample an analog signal, is it possible to experience aliasing distortion prior to the DA stage? The answer is no. Although the cause of aliasing is undersampling at the AD stage, the effects of aliasing (distortion) is manefested only at the DA stage. Put simply, you cannot "see" aliasing distortion until you have the reconstructed analog signal and can compare it to the original source.

Erm, you can "see" the spectrum of the sampled signal with mathematics. Aliasing products will be there if the input is not band-limited.

JR, sure there are images at the d/a, but not any of the distortions mentioned until the images hit another device that cant handle the signal.

Yes, you're right.

The paper Oxia linked is quite interesting, it points out how the combination of aliasing and IMD leads to audible distortion products (that would not be so if only one of the two were present). So yes, that paper points out a problem that occurs due to aliasing at the A/D conversion stage (together with IMD somewhere in the playback chain).

The earlier question was I think concerning filtering in the D/A. The distortion mentioned above occurs *regardless* of whether and how the DAC does its filtering. This is because the spectral components that are generating the IMD are within the audio band and are therefore not filtered out in any DAC.

As Steve points out, not filtering the output of the DAC could also lead to distortion being generated subsequently in the playback chain. This is a separate issue from aliasing. Slew-rate limiting and IMD both seem likely. Very interesting...!

JohnR

[Todd Krieger]

On upsampling...

The purpose of upsampling is to simply enable use of the DAC at its max sample rate to play Redbook CDs.  It basically replaces the DAC's digital filter with its own digital filter.  The output from an upsampler is a "filtered" Redbook signal at 96 or 192 kHz.

Although the DAC itself is operating at a higher D/A frequency when an upsampler is utilized, since the upsampler *already* filtered the Redbook signal, in the same way oversampling does, the resultant analog signal out of the DAC is no more resolute than more-conventional methods of D/A conversion.  (In fact, unless upsampling does 320x oversampling followed by 147x "downconversion" [which would result in a synchronous conversion to 96 kHz], the asynchronous sample conversion will be somewhat "lossy.")

In regard to listener preference of upsampling in a digital rig, it comes down to whether the listener prefers the digital filter in the upsampler or the digital filter in the base DAC.  If the base DAC is excellent to begin with, the less-likely an upsampler will improve upon it.

But technically-speaking, an upsampler does not gain anything, aside from giving the listener a choice of digital filters.  Unlike some claims by manufacturers, it does *not* improve resolution in Redbook CD playback.  (The SimAudio paper got it right, by the way.)

On Aliasing:

An alias is a frequency-reflective image about Fs/2 (half the Redbook sample frequency).  In the A/D conversion, a "brickwall" filter is required because if for example, a strong 25 kHz overtone exists *before* digitization, recording *without* a filter would encode an image tone to the CD at just above 19 kHz, which is totally unrelated to the music signal.  (Higher frequency tones before digitization would be encoded at even lower frequencies onto the CD.)  In the D/A conversion, the images would only occur above Fs/2, and as some here have said, it *could* wreak havoc with downstream electronics if no filtering is used at all.  (I think most digital filter-less DACs have some analog filtering to prevent any such problems.)

[rkapadia@ROOP]

Todd,

Glad to see you - always nice to recognize a familiar inmate.  That aside, looking at the human ear, studies conclude the following:

time/phase coherency is *proven* detectable to 3000Hz by the human ear given a pure sinusoidial source.

The human ear can measure to 20,000Hz, or 1/20,000 sec.  Given the math, that's a derivation of the audible impulse response detection of the ear.

Finally, when oversampling, the number of taps invloved add with the impulse response of the driver when calculating the net delay; if the threshold of human audibility is 2ms latency, then upsampling severly disadvanges the listener in that the taps of any FIR or IIR (requirements in any upsampling or oversampling implementation) add a fixed amount of time smear.

One has to ask themselves; if vinyl is the superior source, why is that the case?  It's not S/N ratio, distortion, or any of that...the beauty lies within the timing accuracy of the analog master and source.  In this regard, zero upsampling will always have a fixed advantage over any upsampling/oversampling implementation.

Regards.

Rup

[petermwilson]

Hi,
I use a panny rp82 for dvds and used to use it for dvd-a until i got a mltidisc player.  The newer Panny's have a RE-MASTER feature that up samples cds.  Thr rp82 upsamples to 88/24 and sounds much better than standard.

The manual had a provviso saying that it would have tobe recognised by your reciever and that even though some recievers regognize 96, they don't all recognize 88.

Anyway the upgrade to my denon5800 provided that capability so when I put in a cd the face of the denon flashes pcm88.2

I understand some of the newer Panny's have an even higher upsampling rate, so before investing in expensive DAC's you may want to try that.

Peter m.