Head-Fi.org › Forums › Equipment Forums › Sound Science › Interesting Video on Analog vs. Digital
New Posts  All Forums:Forum Nav:

Interesting Video on Analog vs. Digital - Page 3

post #31 of 41
Quote:
Originally Posted by Hudson View Post
Theoretically yes, assuming you had a perfect reconstruction filter.
How good are the reconstruction filters used in modern DACs? How close to perfect do they get?

I've read Dan Lavry's presentations. In one he talks about using the sinc function for reconstruction. Doing the reconstruction that way eliminates the stair steps. They're gone completely. Pretty amazing. But how many DACs are using a reconstruction filter like that? How many are taking an easier route that isn't quite so perfect?
post #32 of 41
^ if it just takes a low-pass filter it shouldn't be so hard for every manufacturer to do. Every point in a sample gets multiplied by the entire span of a filter that can comprised of perfectly smooth oscillations, so nothing is left from the original that is not continuous and smooth. So you can see what a low-pass filter looks like here's a close-up below of (a relatively low-resolution) one I made before for analyzing EEG.

post #33 of 41
Quote:
Originally Posted by Hudson View Post
Theoretically yes, assuming you had a perfect reconstruction filter.
With the 22KHz example represented by just 2 samples.

I am sure you could get rid of the steps and make it 'nice and smooth', thats easy, but I'm not sure how it is possible for the resultant waveform to accurately represent the origional.

For example if the two sample points are away from the peaks of the sine wave, how could you possibly know what the amplitude of the peaks was? This information just isn't there any more. I'm not an expert in signal processing but I can't get my head around how you could know this by taking just two samples for each wavelength.

So in the real world what sort of reproduction of the original 22KHz wave can you get? I don't know but for the reasons I give above I cannot believe it will be that accurate. I am sure there are experts on here that can answer this?
post #34 of 41
Having Fremer testify in an expert capacity about the technical merits of digital audio is the result of an editorial stance that's either hilariously misguided or deceptive and insulting to the viewer.
post #35 of 41
the precondition for discrete time sampling is to band limit the analog input signal to less than 1/2 of the sample rate

this was originally done in the analog domain with very high order analog filters, now lower order analog filters are used in front of a oversampling ADC

in delta-sigma converters the oversampling rate can be >256x, then digital filters are be used to give the sharp cutoff for final low sample rate output

if band-limiting is properly done then your sampled signal has all of the information of the band limited analog signal - according to Nyquist’s theorem

we do potentially loose some info by quantizing to a fixed # of bits - the relation of bit depth to analog S/N is given in Shannon-Hartley channel capacity theorem

the reverse order of upsampling and digital filtering is used in the reconstruction filtering process to allow less extreme analog filters for the final output

no one would record or process at 16/44.1 in a serious studio
post #36 of 41
Hey JCX, you look like you know what you are talking about.

Could you answer my question specifically:

with CD 16/44.1 playback (not recording) how can you recreate a 22KHz sine wave by reading just 2 samples. If the sample point is not exactly at the peak sinewave value, how can it know what this peak value is?

Are you saying that you can't?
post #37 of 41
Quote:
Originally Posted by Shark_Jump View Post
So in the real world what sort of reproduction of the original 22KHz wave can you get? I don't know but for the reasons I give above I cannot believe it will be that accurate. I am sure there are experts on here that can answer this?
It's not very accurate. It's called quantization error in the digital realm. The difference between the actual signal and the reconstructed signal. The error gets larger higher you go in the frequency range.

But the thing is that does it really matter that much?
At 22kHz a single wave lasts only 45,45 µs. So I am not sure if humans can even hear differences in a frequency that high, let alone minute details.

Besides while analog can represent any frequency accurately in theory. I don't think those signals can be read and played back that accurately either. High frequencies are really small differences in analog formats which makes them hard to read accurately.
post #38 of 41
the links contained in posts on and after 2009-12-17 from this thread

http://www.head-fi.org/forums/f133/b...88/index2.html

should be helpful for those seeking basic info wrt sampling theory and quantization.

(see especially Pohlman, Principles of Digital Audio)

hth
post #39 of 41
Quote:
Originally Posted by tuoppi View Post
It's not very accurate. It's called quantization error in the digital realm. The difference between the actual signal and the reconstructed signal. The error gets larger higher you go in the frequency range.

But the thing is that does it really matter that much?
.
OK Thanks and point taken. But that also means:

At 11Khz frequency, one wavelength is only represented by 4 sample points.

If you showed the waveform at the start of this thread for frequencies greater than 10KHz it would actually look far worse!

It seems like the amplitude and shape of any waveform over 10KHz frequency can only be very approximated due to low numbers of sample points over a wavelength.

I agree what you say about high frequencies though. I did a test at college and no one could detect a difference between a sine wave and a square wave over 5KHz (try it yourself).
post #40 of 41
^ Practically speaking, low pass filters are made to process frequencies up to 20kHZ, which would deal with the problem you point to above. Even filtering at SampeFreq/2-1 Hz would be enough to sample at different places in the wave and resolve that ambiguity. In our electrophysiology lab we don't extract frequencies over ~45% of the sample rate to better deal with nonstationary signals. But multiplying samples by the filter above (post 32) would have a best match (least squares or largest product) that would exactly fit even a short segment of a sine wave with ~2.001 + samples per cycle.
post #41 of 41
Quote:
Originally Posted by leeperry View Post
It's been thoroughly agreed that SACD sounded better due to superior mastering....listen to DM's Violator, and become a believer.
Ummm....that was my point to begin with. Superior mastering = better sound. It's another's engineer belief that SACD doesn't faithfully reproduce the master tape. It sounds different.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Sound Science
Head-Fi.org › Forums › Equipment Forums › Sound Science › Interesting Video on Analog vs. Digital