Conference napalm::commusic_v1

    
    Yes, it has a great deal of phase shift around the
    steep cutoff.
    
    This 'smears' signals going thru the filter as
     a function of their component frequencies,
    (the highs are delayed more). This results in
    oh, things like ringing on a square wave input.
    
    Not a big deal for elec music, maybe, but a real
    minus in the hi-fi music reproduction area (read:
    compact disk)
    
    ron
    

    I don't believe ringing is a function of phase shift (the ringing
    is at a frequency that the filter won't pass).  I think ringing
    has more to do with the cutoff slope.  Tom, please feel free to
    correct me.  I.e., a low pass filter with a sufficiently low cutoff
    frequency will convert a square wave into a sine wave, regardless
    of how much phase shift it introduces, with no apparent ringing.
    It is possible that a square wave whose fundamental is way below
    the cutoff frequency will acquire some ringing.
    
    Also, it remains a subject of debate as to how audible phase shifts
    are.  The ear is notoriously insensitive to phase differences unless
    they manifest themselves in obvious ways (e.g., cancellation due
    to 180 degree phase shifts).  For an asymmetric waveform, a 180
    phase shift is not the same as inversion, so you don't get the obvious
    kind of cancellation here, and most interesting musical waveforms
    are asymmetric.
                                                   
    However, your basic point is correct - you don't get steep cutoffs
    for free, and the coin of the realm is phase shift.
    
    len.
    

Geez, going back and reading, I see that my first reply did strange things.

Here's what I wrote originally:

Yes, an anti-aliasing filter is just an analog LP filter at the front end of
a digital sampling system.  It's job is to cut off all of the frequencies 
higher than half the sampling frequency of the digital sampling system.
If signals that were N Hz greater in frequency than half the sampling freq
were allowed to be sampled, they would appear within the digital system 
as signals that are N Hz *less* than half the sampling frequency.  When
converted back into the analog domain (ie: playback of a sample from a 
Mirage), the signal would not sound the same as the original...

Todd.

    
    Half right. There also must be a filter on the reconstructed
    output since the output will contain signal at the sampling
    (or playback) rate. Thats gotta go.
    
    Also, no reason these guys have to be analog. Some 5th order
    eliptical filter chips are on the market (and cheep) and want
    *surprise* a clock input. output filter cutoff freq is some
    fraction of the clock rate.
    
    The reason you want a steep slope is so you can get the most
    bandwith out of the lowest possible sampling rate (less memory
    needed for samples). Nyquist theory says 'thou shalt not sample
    less than 2 times the highest frequency in the input signal".
    But thats ideal. requiring an ideal LP filter....ultimate steep.
    
    Len, remember bode plops, er plots? The phase shift due to this
    filter (which can also be in terms of 'delay') is a function of
    freq. A signal coming in other than a sine wave, composed of
    many frequencies, will have each of those freqs. delayed by a
    different amount. After this complex delay the signal is now
    changed in terms of waveform. You have a point about not hearing
    phase differences in signals, but I'm not sure that works here.
    (Not sure it doesnt either...) Two entirely different looking
    waveforms may sound the same. Fer sure.
    
    Like totally critically damped,
    ron
    

rep : -.1 
but you still need an analogue filter up front because the discrete filters,
though analogue, are samplers, like the MF10, and need anti-aliasing filters
in front of them.

Also, outputs of digital systems need filters to suppress the logic edges
and fishing around the D/A did, which is a good reason to put a sample/hold
after the dac, and then a filter to suppress the transitions the s/h
makes.
Tom

Right.  You have to clean up the signal in the analog domain before you
convert it into the digital domain to prevent aliasing (unless you have
an infinitely fast A->D converter).  If the digital filter chip is fast
enough to sample audio freqs w/o an analog front end, then the digital 
filtering provided by that chip isn't necessary anyways - you've already
digitized it!

And yes you need an analog filter on the D->A end to smooth the discrete
logic edges, but that is not called an anti-aliasing filter.  Aliasing 
occurs in the conversion from the analog domain to the digital domain...

Todd.

    
    First, you hear aliasing in the analog "domain".
    
    Traditionally, the input filter is the anti-aliasing
    filter and the filter AFTER a D to A is usually called
    a 'smoothing' filter.
    
    Second, you havent 'already digitized' a signal
    going through a digital filter since the output
    of that chip is reconstructed low-passed analog
    that will go into your more expensive A to D .
    The chip suggested is something like the AMI S3528.
    Yes, it does require an anti-aliasing filter! (youre
    right) But that becomes a simple 6db/octave (read:
    a resistor and a capacitor) About $5 in 100's. 7th
    order (!!!) eliptical low pass. Signal out is down 51db
    at f=1.3(f_sub_cutoff). Great! That allows you to
    approach sampling at the nyquist limit, again, the
     point of which is to have a longer sample time 
    for a finite amount of memory.
    
    Enuf. Anyone want the spec sheet/ app notes?

    ron
    

    Tell me more!  I confess, I haven't been following analog function
    blocks in a while.  Are these filters easily available?  Cheap?
    Sources?  How 'bout a synopsis of the specs -- max operating
    freq., input referred noise, analog range for common supply levels,
    etc.  Are they quiet enough for use in 12-bit systems?  16-bit?
    
    (Yes, I will also refer to the ANALOG conference.)
    Thanks