Re: Q: Applied Engineering Sound Card

[mjmahon@aol.com (MJMahon)]

Mitchell Spector (spec@vax2.concordia.ca) wrote:

>>>The final blow to the chip was the low-pass cut-off filter that the
>>>GS added. Apparently, from what I'm just finding out now, regardless
>>>of whether the chip or software played a sample at 44.1 kHz (or other
>>>high frequencies) the filter would essentially drop the quality down
>>>to around 26 kHz.
>>
>>Mitchell, could you post the passage from this article?  From what
>>I've read, the GS does not have such a filter, but the Amiga
>>does.  The author of the article that I read said that the Amiga's
>>sound can sometimes be cleaner because it has a low-pass cut-off
>>filter that filters out the ultrasonic frequencies.  He said that
>>the GS lacked such a feature.
>
>    The Ensoniq has no low-pass filter, but there _is_ one on the IIgs
>motherboard (everything coming out of the chip has to go through it).
>This lowers the quality of the Ensoniq when your playing higher
>frequency sound (roughly anything above 26 kHz). I have heard that
>the Amiga 1000 had a low-pass filter as well, but you could modify
>the board to by-pass it (I think the Amiga 500/2000 also had one,
>but you could disable it through software?).
>
>    Article? Well I do have an e-mail message from David Empson who
>was explaining the technical details behind the low-pass filter, but
>I don't want to post it without his permission. It certainly does
>exist though, I've found references to the low-pass filter in some
>Addison Wesley books and it should be covered in the Apple II technotes.
>It definitely is not an advantage to have one. :/

Wrong.  Sampling theory is quite clear on this point.  When any signal
is recreated from discrete samples, there is no useful output above
one-half the sampling frequency, the so-called Nyquist frequency.

If energy above the Nyquist frequency is not filtered out, it will only
increase the noise in the re-created waveform.  Think of it this way:
the unfiltered output is a stairstep approximation to the sound being
produced, while the desired sound is a continuous waveform (with
maximum frequency not exceeding Sampling Freq./2).

The same filtering must be applied prior to sampling an input waveform,
since any energy above the Nyquist frequency would simply be "folded
back" (aliased) into the frequency range from 0 to Nyquist freq.

Since "brick wall" lowpass filters are hard to make and have undesirable
phase-shift characteristics near the cutoff frequency, gentler lowpass
filters are used, which will be 3db down somewhat before the Nyquist
frequency, so that they can be >30db down at the Nyquist frequency
(and beyond).

Note that much lower lowpass cutoffs should be used when sampling
at lower frequencies.  For example, if sampling at 11KHz, then the
lowpass should cutoff lower than 5.5KHz, the Nyquist frequency.

There ain't no such thing as a free lunch--and you can't do better
than the Nyquist frequency!  A lowpass simply acknowledges
reality and makes for a proper design.



-michael

 Email:  mjmahon@aol.com
 Home page:  http://members.aol.com/MJMahon/