Conference napalm::commusic_v1

    Aliasing is a frequency introduced into a sample by attempting to
    sample a sound containing a frequency that is higher than 50% of
    the sampling rate.
    
    Got that?
    
    F'rinstance, if you use a sampling rate of 10Khz, the highest frequency
    you could sample would be 5Khz. Anything higher than .5(Sample_Rate)
    shows up as aliasing, usually as a *lower* frequency sound. In order
    to alleviate aliasing you have to either (a) increase your sampling
    rate (which will propotionally decrease your sample time) or (b)
    filter out all frequencies above the Nyquist Limit. (Nyquist is
    the democrat from Arkansas who passed this stupid law...) Seriously,
    the Nyquist limit is equal to .5(Sample Rate). I just bought the
    anti-aliasing filter for my Mirage which allows me to filter out
    anything up to a Nyquist of 25Khz (50Khz sample rate) at a 150db
    per octave rate. At 50Khz rate, I get a VERY short, VERY clean
    sample.
    
    The hiss you desribed sounds more like quantization noise passed
    through your envelope generator. You may be able to lose some of
    it by closing your filter down a little. If you have a dynamically
    controlled filter, you may want to increase the release rate so
    the filter closes faster as the note decays.
    
    Aunty Alias.

    the 1/2 freq. sample rate assumes a perfect, unrealizable filter.
    
    also the frequency here is any harmonic (ie, sinewave) that is 
    > 1/2 the sample rate. clarification.
    

    Does this hiss you hear always sound the same regardless of where
    on the keyboard you play? Does it change 'pitch'? Proportionally
    or inversely to the pitch you're playing?
    
    Aliasing will change pitch proportionally to the keyboard.
    Quantization noise doesn't.
    
    Edd
    

RE:-1
	If I remember correctly the hiss sounds the same regardless of
	where I play on the keyboard.  The pitch seems to follow the
	sound as it decays.

	You're all more than welcome to stop by, hear it and diddle.
	Of course that's if you are near Marlboro MA.  Which I believe
	you, Edd, are.  Send me mail if so.

Dan

    Edd:
    
    	Really 150 dB/octave?  Wow.  That's the equivalent of 25 analog
    poles.  Is it digital, or analog?  How much was it?
    

    Yep. No typo. 150db/octave...
    
    I'll bring in the specs tomorrow.
    
    $149.95
    
    Edd

    I checked. There is no indication as to the topology of the circuit.
    The literature claims 'a high-speed A/D converter' but doesn't specify
    if the ADC comes before or after the filter.
    
    My gut feeling is it's a digital filter.
    
    Signal is down 53db at 1.3 times the cutoff frequency.
    
    Edd

If it is an anti-aliasing filter, it is *analog*.  You wanna filter out
all the harmonics greater than the nyquist frequency *before* you sample.
It is sometimes also wise to filter going the other way (D to A) with a 
smoothing filter to reduce quantization noise.

The steeper the filter, the more phase error is introduced.  Ya don't get
nothin' for nothin'...

Sounds to me like the D-50 uses 8-bit or 10-bit samples, and that the noise
is quantization noise.  Does anyone know the actual number of bits of
resolution for the D-50?

Todd.

    
    Not only phase error but frequency response ripple. But you're probably
    not too concerned about that in a sampler.
    
    CdH
    

RE: -2
	I can't remember how many bits and I called Shane and he's gonna
call and find out that and where my RAM card is.

Dan

RE: -1
	I stopped by Dan Eaton's office and he had the latest issue of
KEYBOARD which has the Roland D-50 add which advertises 16 bit PCM
sampled sounds.

	Hmmm.

Dan

    16 bits make quantization noise a less likely candidate, donit?
    
    Edd

    I don't know, but it sounds digital to me.  I just can't imagine
    a 150dB/octave analog filter.  The reason is that the analog noise
    introduced in such a beast would probably be unbearable.  I mean, you have
    to amplify after every few stages (or use a bunch of active filter
    stages, which is the same thing) so that your signal doesn't disappear.
    
    By the same token, your converters would really have to crank in
    order to keep the anti-aliasing filter from aliasing if it were
    digital.  I'm really curious now.
    
    Steph
    

It doesn't make sense to me why it would be a digital filter.  After everything
is digitized, why filter it?????  If there is aliasing noise present, it
will be folded down into the meat of the audio spectrum and would be
unfilterable in the digital domain.

It's gotta be analog.  Perhaps it has a zero that helps steepen the rolloff???

Todd.

    By the by, the filter cut-off point is software selectable.
    
    Do that be any help?
    
    Edd

Must be a digitally controlled analog filter (DCF).

Todd.

Edd stopped by and twiddled with the D-50 and felt that it was quantitization,
hmmm did I spell that right?  A fellow D-50 owner from the USENET concurs...

From USENET, which is still down for newsgroups but not mail...

Newsgroups: rec.music.synth
Organization: Motorola Inc. Austin, Tx

Hi,
My D-50 makes some noise during the decay when the digital reverb is on..
I presume what we are hearing is quantization noise.  When you mix the
D-50 with other instruments ..like on a multi-track tape deck...you can't
hear the quanitzation noise.  Even my old Mirage sounds ok when you mix
it with the rest of the band.... oh well ...nothing is perfect.

Regards,
Charlie Thompson
Motorola Microprocessor Operation
Austin, TX

    Todd. It does not have to be an analog filter.
    
    There are 7th order elliptical filters using CMOS switched capacitor
    technology made by Gould for reasonable price that could easily
    perform as this filter. Specs say "...greater than 51db of rejection
    at 1.3 f_cutoff". Aliasing MUST still be considered tho. There
    is a 'cosine prefilter' in the chip which compensates partially.
    You may need 1 or 2 pole input lowpass used with the input op amp
    There will be some components of f_clock +/- f_in aliasing on the
    output of all sampling devices. There is an on chip sinx/x sample and
    hold that reduces these components to -30db. A single pole analog
    smoothing filter (one cap and resistor)will reduce that again to -50db. 
    
    Oh, and the cutoff frequency is selectable over 64 'preset' values.

    I'd hate to design a quiet 7th order analog filter.

Look.  Edd stated that he purchased an anti-aliasing filter with 150db/octave 
rolloff.  In my eyes, an anti-aliasing filter stage must be in place prior
to signal digitization to clean up the harmonics higher than the nyquist
limit, otherwise aliasing occurs which inserts signal components in the
frequency bounds (0 to fnyquest) of the digitized signal.  

Either I'm missing a method of extracting this aliasing distortion in the
digital domain once it's there (smoothing filter may help), or we're talking
about a two stage filter that removes aliasing frequencies in the analog
domain and then performs some digital filtering for some reason beside
aliasing...

Todd who_is_trying_to_avoid_aliasing_rather_than_fixing_it_after_it's_there.

    Todd, the digital scheme would (probably) work something like this:
    
    
    
    ---->|analog filter|---->|digital filter|----->|analog filter|---->
    
    Now, let's say the analog filter rolls of at F=20KHz, but provides only
    two poles of slope.  The digital filter must run faster than this. That
    is, the clock should be greater than 40KHz (by a reasonably large
    amount).  Then with the digital filter, you make a noiseless, large
    slope (20 million poles?) low pass filter which attenuates 20KHz to
    whatever half it's sampling frequency is.  Then the analog filter on
    the output smooths out the quantisation noise (remember quantisation
    noise?  This is a song about quantisation noise...).
    
    The kicker is that the frequency at which the digital filter samples
    is high enough so that even though the SLOPE of the analog filter
    is pretty wimpy, it has still virtually eliminated harmonics at
    this frequency, simply because it is far away from the cut-off
    frequency.  Et voila, you get minimal aliasing, and a precipitous filter
    cut off characteristic.
    
    Steph
    

Yep.  You definitely need to clean the signal up in the analog domain first
(as you showed).  I understand the benefits of the digital filtering, the
only problem being that you have to sample much faster than twice the nyquist
frequency.  So for Edd's Mirage mod to be a digital filter, it must in fact
boost the sampling rate of the Mirage, no?

Todd.

    It do.
    
    Edd

    hmmm. A side effect.
    
    Now we get it! You CAN use a digital filter. Yes, it DOES need
    a high sample rate for 20Khz cuttoff. I thought we knew that.
    It may be used as a low pass filter with such a steep cutoff
    that a lower sample rate may be used since the component frequencies
    above f_samp/2 are so strongly attenuated.
    
    The chip was quoted me $5 in 100's, but that was a year ago...
    
    rr
    

Hey, if we sample at a sampling rate of infinity samples per second, we
don't have to worry about aliasing anymore!  The nyquist frequency would
be half of infinity, which of course is infinity.  In other words, lets
just stay in the analog domain, and use analog memory chips.

Todd.

    
    letsee.....
    
    then if sampling at infin/2 yields bandwith of infin, does infin/4...
    
    ad absurdum.

    buckets? no. they in fact are samplers too sort of...