More fuel to the fire : analog vs digital

Found this youtube video. It is intriguing, VERY intriguing. But do read some of the comments below the video.

https://www.youtube.com/watch?v=4eC6L3_k_48

Good one.

But there are some who would argue stubbornly that anything above 20KHz (and below 20Hz) doesn't matter because their highly perceptive ears (and human body in general) are insensitive, and therefore irrelevant.

I'm not one of them.

These two videos are even more intriguing (to me at least).

Look at the unique patterns certain sound frequencies can create out of evenly spread particles (sand)

https://www.youtube.com/watch?v=oCmGjD9j9bU

To be honest today only I learned about the 'saw' frequency. What really fascinates me is the 'melody' formed as the frequency goes beyond a certain range.

https://www.youtube.com/watch?v=G71yKMtXDoA
https://www.youtube.com/watch?v=s5QIkGCxRHk&NR=1

There are lots more and extremely interesting, like the below comparison in output of a non-oversampling and 4X oversampling CD player.

https://www.youtube.com/watch?v=vgi-r_z1cS4

But I think we need to have fundamental understanding of digital signal processing to fully understand them.

WongKN wrote:There are lots more and extremely interesting, like the below comparison
in output of a non-oversampling and 4X oversampling CD player.

https://www.youtube.com/watch?v=vgi-r_z1cS4

But I think we need to have fundamental understanding of digital signal processing to fully understand them.

If the non-oversampling CD player's output was passed thru an analog brickwall filter (the way it normally would work, had the lab dude not bypassed that stage in order to just read off the unfiltered output), the scope screen would not seem as horrible or "digital" as it did. The analog filter would have removed all those higher order freq components and it would more closely resemble the over-sampling player's output.

Remember the basics of a Fourier Series. A square wave is comprised of its fundamental Sine component and every other odd order harmonic components to infinity. So without either a digital filter or analog filter stage, what you're seeing are all those odd order higher harmonics that gives it all the square blocks and dot and whatnots.

Just in case anyone reads too much into that video and to avoid a scenario where "a little knowledge is a dangerous thing".

KN, that "saw" video demonstrates aliasing as the sampled frequency approaches the Nyquist frequency.

You can see it in digital photographs as well:



The aliasing is the wavy lines on the scarf - a spurious pattern. This is why anti-aliasing filters are so important.

I like illustrative explanation of technique matter, even formula.. but the very technique book end up having more wormish sentence of words rather than having it illustrated.

Now it is even much clearer with video~

WongKK wrote:KN, that "saw" video demonstrates aliasing as the sampled frequency approaches the Nyquist frequency.

You can see it in digital photographs as well:

Spoiler:

The aliasing is the wavy lines on the scarf - a spurious pattern. This is why anti-aliasing filters are so important.

Doc, actually that isn't aliasing (referring to the youtube video's scope screen). Those are actually the peak-hold stepped output of the D/A conversion stage, which aptly should have been filtered out by the analog filters (or digital filters, depending on how mfgs implement the D/A conversion).

Digital audio Aliasing happens as a combination of the A/D & D/A stages, where if the A/D converter where to digitize some signal thats more than twice the sampling freq, then it would have been captured as a foreign component in the digitized signal and when played back in the D/A stage, gets folded down below the Sampling-freq.

For example, if Fs = 44.1KHz, and the recording engineer purposely used it to record a 23KHz tone, then when played back, the original 23KHz tone would be reconstructed as a 21.1KHz signal. This 21.1KHz is totally foreign and unrelated to the original analog signal in the first place. (except due to aliasing in the A/D - D/A stages)

The folddown freq is calc'ed as such: 23KHz - 22.05KFs (aka Nyquist Freq) = 950Hz

Digital photography aliasing isn't quite the similar analogy here.

Again, those squarish and dotty outputs as seen on the Scope, are NOT aliasing artifacts, but the fact that the lab guy did not run thru a filter stage to "clean up" the reconstructed analog signal.

This thread is generating the interest, as well as the technical input I have hoped for. For those who are not inclined to the technicalities, the main purpose, or rather MY main purpose of starting this thread is to illustrate the kind of artifacts that A/D and D/A can generate. Many thanks to the forumers who have responded with their technical explanations.

The original youtube video also clearly illustrate the term 'low-pass brick wall' filter on RBCD and the more recent posts by the forumers explains why it is needed. The scoping of the turntable output also shows the meaning of 'frequency response'. Yes, frequency response of old LPs and budget turntables may be "30Hz - 18kHz" (just an example). But that is telling HALF the story. Frequency response is specified with the dB tolerance as well. So the actual frequency response can be "30Hz - 18lHz" +/- 3dB for e.g. And what happens after that ? How fast does the response drop ? What happens at say 30kHz ? Is it -6dB or lower or higher ?

So just be aware that digital sound and the improvement of digital sound is not just simply "increase the sample size, increase the sampling frequency". So just the magic "16/44.1 -> 24/96 -> 24/192" does not automatically generate better sound. There are lots of other considerations.

The video from the lab guy scoping the output of a CD player, I was thinking for a while before putting it up. There is a possibility people will get confused, and thinks whatever is on the scope is what is output from the CD player's analogue outputs. But I think forumers should now be able to understand that it is not, after explanations from ppl like mugen and the good doctor. What I like about this particular video is that it clearly shows that post-processing of the signal AFTER the DAC stage is necessary and that it is not as perfect as things are made out to be. So a number of filters (anti-aliasing, the low-pass brickwall) are needed. There is also a delta-signma conversion required. All of these things can matter more than simply using more bits and more samples. I think this is what mugen tried hard to explain in the 'other' vinyl thread but which I suspect many people probably did not really understand. I would think what mugen was trying to say is now clearer.

Haha, end of the day 16/44.1 still the best resolution forwards~



hmm, few things that I need claritication, looking into the comments on the video clip, it seem quite true that those recorded beyond the 20kHz isnt exactly music info but some subharmonic contributed by the stylus and equipment attached to it.. Something like the DAC pre and post ringing...



DAC is a very complex stuff to understand properly.. even after reading so many thing and the also the spec sheet of the chipset, still dont get to know exactly what has taken place at which stage.

Adrian, that is what ONE responder on the first video CLAIMS. Whether or not it is technically correct is still open to debate. His point was that the equipments of the day at that time were not supposed to be able to capture very high frequencies. Therefore is it supposedly cannot capture and yet the turntable reproduces, then it must be artifacts. But again, the question is what is defined as 'not captured' ? Once again the frequency response of an equipment is always given in terms of a range of frequency and the dB variation within that range. I.e. 20Hz - 20kHz +/- 0.5dB in the case of some earlier CD players. But as explained, does that mean there is nothing below 20Hz in CDs ? The frequency response of the CD player can well be -0.5dB at 20Hz, but IN ADDITION, -1dB at 18Hz, -3dB at 15Hz, -6dB at 10Hz and so forth. So the frequency response specification alone doesn't tell the whole story. As mugen puts it "a little bit of knowledge....". So we must be careful not to fall into traps like this.

The technical papers on DSD/SACD gives a good illustration on the various stages involved in a typical D/A conversion. Let me search to see if I can find it.

WongKN wrote:The video from the lab guy scoping the output of a CD player, I was thinking for a while before putting it up. There is a possibility people will get confused, and thinks whatever is on the scope is what is output from the CD player's analogue outputs. But I think forumers should now be able to understand that it is not, after explanations from ppl like mugen and the good doctor. What I like about this particular video is that it clearly shows that post-processing of the signal AFTER the DAC stage is necessary and that it is not as perfect as things are made out to be. So a number of filters (anti-aliasing, the low-pass brickwall) are needed. There is also a delta-signma conversion required. All of these things can matter more than simply using more bits and more samples. I think this is what mugen tried hard to explain in the 'other' vinyl thread but which I suspect many people probably did not really understand. I would think what mugen was trying to say is now clearer.

It should be further clarified that Filters in the D/A stage aren't for preventing aliasing per se (as this role is at the A/D stage) but more to remove the gunk and harshness of those higher order harmonics from the D/A conversions.
Oversampling pushes these "junk frequencies" higher above the nyquist so thats its easier on the filter stage to have a milder rolloff and maintain better phase coherency.

Which is also why Oversampling in the D/A staqe, doesn't restore or recover any of the original signals which are higher than the half-Fs which was already lost in the first place in the A/D encoding stage. (killed by the lowpass "anti-aliasing" filter).

It's the lowpass filters in the A/D stage that would actually preventing the digitizing and quantization process from encoding any signals above the Nyquist and hence would be folded down during the D/A stage.


Here is a quick excerpt:

http://en.wikipedia.org/wiki/Digital-to-analog_converter#Practical_operation

The fact that DACs output a sequence of piecewise constant values (known as zero-order hold in sample data textbooks) or rectangular pulses causes multiple harmonics above the Nyquist frequency. Usually, these are removed with a low pass filter acting as a reconstruction filter in applications that require it.

mugenfoo wrote:Doc, actually that isn't aliasing (referring to the youtube video's scope screen). Those are actually the peak-hold stepped output of the D/A conversion stage, which aptly should have been filtered out by the analog filters (or digital filters, depending on how mfgs implement the D/A conversion).

Sorry Mugen, I was referring to the video with the swept sawtooth pattern. As the source frequency rises, spurious lower frequency sounds are heard thanks to aliasing by the sampler.

Tonight I am a bit sibuk ... will have a look at the rest of your posts later. Sorry!

No worries Doc, we are always up for a good civilised discussion based on sound facts.

As for the sawtooth readout on the scope, its still not due to aliasing but the effects of D/A peak-hold conversion. As mentioned before, the lab guy bypassed the analog filter, hence those kind of edgy and rough jagged scenes.

If you need a good reference on Digital Signals, or Digital audio in general, i'd be happy to post the ISBN numbers for your further reference. But they are the typical textbook stuff (just like those medical books) you would refer to once in a blue moon to clarify certain facts and that's it. Definitely not your coffee table top material.

More fuel...

http://news.cnet.com/8301-13645_3-20097820-47/the-lp-comes-of-age-quieter-and-better-sounding-than-ever/?tag=cnetRiver