Conference cookie::notes$archive:cd_v1

These are some good questions.  I'd like to know more about the field too.

Are you talking about digital filters?  The ones that ring BEFORE the
transient as well as afterwards?

I'm finally getting some minimum phase speakers.  There arn't many on
the market.  One common example is a KEF speaker that has 900 degrees
of phase shift at 20khz - all in the pursuit of the Julian Hirsh nirvana
of flat response.  Some of the minimum phase speakers are:  Theil, Vanderstein,
Acoustat, Apogee, Magnapan, VMPS, JSE, Spica, Proac, and Clements.  The
claimed benefit is better imaging.

What does the filter ringing sound like?

Well, I found a reference to phase and ripple sensitivites in a Philips
technical article.  They cite:

Lagadec, R. and Stockham, T., "Dispersive models for A/D and D/A conversion
systems", Paper presented at the 75th Convention of the Audio Engineering
Society, March 1984, Paris.

The results indicate a "discerning" listener can detect a ripple of
+/- 0.2 db in the pass band (20-20khz).

"Most people can hear phase distortion (group delay variation) in an audio
system as a change in the stereo 'picture'.  This effect is worse when any
phase distortion is accompanied by ripple in the pass-band.  It is notoriously
difficult, however, to correlate ripple and phase response to audibility
thresholds.  Experiments ([ref above]) suggest that any audible effects are
due, not to the ripple amplitude itself (many high-quality audio systems
used to have ripple figures exceeding +/- 0.5 db), but to the large number
of ripples in the pass-band of a filter with steep roll-off."  ( p. 220,
Electronic Components and Applications, Vol.6, No. 4, 1984)

"The audible manifestation of steep roll-off, pass-band ripple and non-linear
phase response are pre-echoes and post-echoes which can be readily observed
on test burst sine waves (400HZ test signal acc. to Institut fuer Rundfunk-
technik, Munich, Germany)"

Anybody have access to the AES Proceedings?

Re questions in .1 and .2

Actually, I'm addressing the general question of what may be important, or
even perceptable, in filters.  Linear phase filters are the ones that
have ringing on both sides, and are commonly done in players as digital
finite impulse response (transversal) oversampling filters.  However,
there have been dubious claims that add-on all/pass networks which achieve
the same sort of effect greatly clarify the sound.  I say dubious since
the authors (such as John Meyer) are selling such devices and predisposed
to hear improvements.  The devices may in fact be perceputally null, or 
harmful.

There may be something to Lagadec/Stockham's paper and I'm going to
obtain a copy (it's only in preprint form, but as an AES member it
will be easy for me to obtain).  In particular, it turns out that
elliptic function lowpass filters of higher order have better, not
worse, group delay characteristics.  This is because such filters
have more, lower Q poles contributing to the approximation, and these
poles don't ring as badly.  Interestingly, merely adding a single
extra stage onto a .5 or so dB ripple filter, knocks the ripple down
to the .005 dB range, so in lumped circuit form it doesn't cost much
for overkill.

At low frequencies, a liner phase impulse sounds like a reverse time
buildup followed by a normal BONG - a strange sonar-ish sound.  FFT
analyzers, such as the HP (forget the number) produce this signal; its
worth hearing to get a feel for the actual sound at various frequency
ranges.  Unfortunately, a minimum phase pulse with the same spectrum is
not available for comparison.

Actually, its only with analog tape decks, and multi-driver loudspeakers,
that nonminimum phase response sounds have become commonplace.

- Jim

Hmm..  It has been over five years since I took my only DSP course, and I
have forgotten much of what I didn't learn too well in the first
place.

I assume we are talking about CD players using over-sampling since otherwise
we won't be using any digital filters.  It may be that a non-casual
filter may produce funny sound because it rings before the impulse hits.
But I don't see how we can here it in real live.  Assuming the recording
people did a good job of bandlimiting the music so that aliasing never
occurs, there will never be any transient fast enough to cause a
properly designed digital filter to ring.  If there is enough impulse
to cause a digital filter to ring the way we see it in stereo magazines,
we are already way into the aliasing domain and nothing is guaranteed
to sound right.

Linear phase filters (constructed typically with FIR filters??) have
their phase shift varying linearly with frequency, so in fact they
behave nothing more than a constant time delay, which we should be able
to hear.

Given a choice, I prefer linear phase filter on technical merits
even though I don't think I can ever hear the difference.

/tom

I've done a few calculations related to this question with some very
interesting results.

I wrote a program to design elliptic antialias filters and then plot their
group delay, impulse response, and energy time curves, and then compared them
to some responses for a few simple transversal filters.

Lowering the ripple of elliptic filters does reduce the ripple on the
group delay and the energy time curve, but below .01 dB the curves
smoothed out very well.  Only with rediculous 1 dB ripples did I see any
spurious late energy arrivals; with normal ripples the curve was essentially
monotonic.  So, high order filters are better, not worse as is often claimed.
The ringing dies away within 500 usec or so...

The interesting thing is that transversal filters have quite a bit less
close in ringing, even though there is a 'preshoot'... apparently the
dispersion of minimum phase filters extends the duration of the ringing
as well as putting it all after the main arrival.

So, it seems that if ringing itself is a problem, going to oversampling
(from the A-D converters all the way thru to the player) is best.  The
preshoot of the linear phase filters is short enough, that no dulling of
transient attacks should be perceptable.

However, there are claims that group delay variations over an order of
magnitude greater than what these filters cause are barely detectable,
so this may not be a problem at all.  In addition, I think nearly all
recordings to date have passed thru at least one uncompensated elliptic
filter, so the type of filter in a player probably doesn't matter as far
as group delay is concerned.

- Jim