🚀 go-pugleaf

In article <72020018@otter.hpl.hp.com> nt@otter.hpl.hp.com (Nicolas Tripon) writes:
>  What is 'oversampling' and 'bitstream conversion' for a CD player ? As the 
>latter tends to replace the former on the front panel of a CD player, I 
>suppose it has nothing to do with the 44.1 KHz sampling rate.
>  Regards. -Nicolas Tripon

Since I bought a CD player about 2.5 years ago, I've lost touch with the new
terminology they're pushing.  However, in that time I've been working on
my MS-EE, so I've at least learned something about what oversampling is.

Basically, oversampling is a method used to increase the sampling rate of a
digital signal.  According to the Nyquist principle, at 44100 samples/sec,
you can only accurately represent a signal with frequencies up to 22050 Hz.
Any frequencies above that will start "wrapping around" and show back up as
lower frequencies when you convert back to analog, so they have to be
analog-filtered out *before* the signal is digitized.  This is prefiltering.

Now, when the digital signal is being converted to analog, it will have
unwanted frequencies above 22050 Hz because of its stair-step nature, so we
have to analog-filter these out after conversion.  The problem comes in
creating a near-ideal brickwall filter that cuts off everything above a
given frequency.  If we assume that the wanted signal only goes up to 20kHz,
then the filter's transition band can only be 2050 Hz wide.  For higher
cutoffs, the transition is even narrower, but we must always cut off
everything beyond 22050 Hz.  Designing and building analog filters to do this
gets very tricky and expensive.

The solution is to take the digital signal and increase the sampling rate
while still in the digital domain.  By increasing the rate 2x, our output
sample rate is now 44100*2 = 88200 Hz.  High-order digital filters are much
easier to implement, so we can cut out some of the unwanted frequencies
digitally.  At this sampling rate, then, our analog spectrum contains our
wanted signal from zero (or 20) to 20-whatever kHz, then nothing (nothing
significant, at least) up to 44100 Hz, the new Nyquist rate for the
oversampled signal.  Now it is much easier to use an analog filter to cut
off what we don't want, because we are allowed a much wider transition
band.  Every time we double the oversampling rate, we double the Nyquist
frequency, but our wanted signal still stays in the 0-22kHz range.

I won't go into *how* oversampling is done (unless asked), except to say
that zeros are inserted between the real samples, and the resulting signal
is filtered to smooth out the zeros.  This is not the same as linearly
interpolating the samples.  A much smoother waveform is achieved this way.

>  What is 'bitstream conversion' for a CD player?

Anyone?

--
Bill Eason - EE (DSP) grad student	  || campus: Box 28021 GA Tech Station
work: GTRI/CSITL, ERB Room 173            ||         Atlanta, GA 30332
      347 Ferst Drive   Atlanta GA, 30332   finger:  wte@cdrom.gtri.gatech.edu
internet: gt8021a@prism.gatech.edu	    UUCP: ...!gatech!prism!gt8021a
-- 
Bill Eason - EE (DSP) grad student	  || campus: Box 28021 GA Tech Station
work: GTRI/CSITL, ERB Room 173            ||         Atlanta, GA 30332
      347 Ferst Drive   Atlanta GA, 30332   finger:  wte@cdrom.gtri.gatech.edu
internet: gt8021a@prism.gatech.edu	    UUCP: ...!gatech!prism!gt8021a