Hojoon Chang wrote on Sat, 07 January 2006 10:34 |
Quote: | In theory, you need to take more then 40000 "slices" (40KHz sampling). In practice you may need to raise it a bit. The "reconnecting" ot the dots into a PERFECT wave (like th original analog one) is done by an analog filter. A perfect filter will make a perfect wave, at sample time and ALSO BETWEEN THE SAMPLES. That is why that "stuff" works.
|
I'm just make a comment. try to search or get info of DSD(Direct Stream Digital recording technology). It uses 2.8MHz for sampling rate which 64 times of 44.1kHz with just 1 bit conversion. You will see how digital audio works.
|
Yes, DSD is another way to do things, and I did not rule it out. Please note that I said:
"Do you need the SINC impulse to be the basic element? No, you can do it with other "basic shapes" but not with sine waves."
DSD is in some way very similar to the old analog pulse width modulation. But unlike a continues (analog) modulation, the digital nature of the process does not allow "any width" to occur. The only allowed widths that are integer multiples of that high frequency clock (such as 64fs, 2.8224MHz). That restriction "costs us dearly" - it is the reason why there is so much noise energy in the signal. In fact a pulse width modulator with a 2.8MHz quantifier would be so very noisy, to make it useless. To solve it, the noise is shaped - moved to frequencies above 22KHz, and it covers the spectrum all the way to the Nyquist (in this case, 1.4112MHz).
The noise shaping feedback loop is a one engineering method to end up with such outcome. One could get there by other design means (such as entropy, for example). The end result is a signal that contains the audio at low frequencies and a lot of high frequency above audio, preferably random noise. The idea of utilizing high frequencies is often nothing more then a modulation scheme.
There are many such schemes, from AM to FM with or without suppressed side bands... At the end of the day you call the circuit a modulator. And when you involve digital into your modulation scheme, there will be some quatization in the picture, thus quantization energy (some error signal to deal with), as well as the requirement to conform to Nyquist.
Regards
Dan Lavry
www.lavryengineering.com