| Schallfeldwebel wrote on Mon, 30 October 2006 19:36 |
Maybe I have been unclear, but my question is, what happens if a chip which is designed to run 128x oversampling for 44,1K is set to 64x oversmpling at 44,1k? Just to accomodate the possibility of recording 96K. At 96K the chip runs on a 64x oversampling.
/E
|
The answer to your question may be IC specific, but as a rule, I would expect the front end of an AD (the modulator) to operate at some fixed high frequency. In your case, the rate for 44.1KHz is 5.644MHz, and for 48KHz it is 6.144MHz. That coresponds to 128fs. Now, when you raise the sample rate to say 96KHz, the modulator can stay at 6.144MHz which is 64fs.
The main differance between 64fs and 128fs will be the back end of the AD (the decimator). In one case, the decimator converts the modulator output (few bits but at very high speed of 6.144MHz) to many bits at 48KHz (or 44.1KHz), in the other case, it "skips" the last "divide bt 2" decimation stage. The decimation is of course is synchronus sample rate down conversion.
The big differance between a 48K IC and 96K IC is in the front end. An IC made for say 48KHz, will be focused on clean usable audio range of 0-24KHz (well maybe 0-22KHz or so in practice).
An IC aimed at sat 192KHz will be designed to accomodate signals up to 96KHz (well maybe 93KHz or so).
So there is no way you could use a 48KHz to record signals above 24KHz. That region above 24KHz is "full of noise" by design. The noise shaping moves noise from under 24KHz to frequencies above 24KHz.
But an IC designed for 192KHz can record 96KH signals. If you wish to set such an IC to opertae at say 48KHz, it will remove the signals it recorded above 24KHz, and in theory yield the same results as the 48KHz IC.
But in practice, going to 192KHz does call for a performance reduction when all things are held equal. All things equal means the same modulator filter order, the same number of modulator bits, the same high frequency clock (such as 6.144MHz), the same IC process...
The amount of noise shaping - removal of noise from "signal bandwidth" (what is dedicated to audio) to the "noise bandwidth" (frequencies above the signal bandwidth) is fixed
for a given set of parameters.
This is analogous to a guy with a capability of dig say 9 cubic feet of dirt. You can ask the guy to dig a 1 foot square hole 9 feet deep. You can ask the hole to be 3 feet by 3 feet and 1 foot deep.
The qestion is then, do you want a "deep hole" for 24KHz of audio? A shallow hole for 96Khz of audio.
If you are willing to accept that we do not hear 90KHz, that our mics and speakers do not process 96KHz of audio, then keeping a wide range of say 96KHz (for 192KH sampling) is a waste. You could, for the same resources get a better 48KHz of audio, and even better performance at 24KHz of audio.
We know that 44.1KHz may be tight in some cases (mostly implementation issue). I view 96KHz sampling as already too high. The optimum is somewhere between 48 and 96KHz sampling, but the standards do not suport it.
Having said all the above, newer IC's often see impovments in many areas, from circuit design to process and more, so one can not assume that all things are equal. So a newer 96KHz IC many be better at 44.1KHz then an older 44.1KHz IC.
It is just too bad that due to the rush to accomodate higher sampling speed, the noise shaper is forced to accomodate a huge range that we do not use. The IC makers could do a better 96KHz IC, if they would agree to do away with 192KHz and apply the same capabilities they invested in 192Khz into a 96KHz IC.
I know some of the above was somewhat off subject, but it does tie in to your question.
Regards
Dan Lavry
http://www.lavryengineering.com
