"adamdea" wrote in message
news:31632483.1067.1316541102365.JavaMail.geo-discussion-forums@yqnk41...

I have been trying to understand the problem of ringing caused by AA and AI
filters.

AFAIK there is no such problem providing that the sample rate is 44KHz or
greater, and provided that reasonable care is used in the design of the
digtial filter.

This is based on a number of DBTs performed a number of different folks over
the years.

I have run into one problem which seems to be assumed in all the
literature I have read. I am rather hoping that there may be a simple
answer to it.
It seems to be stated that filters (non gaussian) will all embed ringing at
their transition frequency.

In general, avoiding ringing in a low pass filter requires gentle slopes and
plenty of damping.

This leads to the point in the argument that provided that the ringing may
be inaudible since it is at inaudible frequencies.

How about that! ;-)

I appreciate Jim's contribution "Ringing in the Ears" (free from Audiomisc
or available for £3.50 for people like me who were too dumb to look first)
suggests that it may be audible due to non-linearities in our hearing
perception.

Jim seems to like being agnostic or otherwise ignore the outcomes of
reliable listening tests. One problem with good reliable subjective tests
is that they can be time sinks. Theory can often be expounded upon with far
less effort.

However before I can wrestle with that I would be grateful if someone
(Jim?) could confirm what is the relationship between the filter transtion
band and slope on the one hand and ringing frequency and amplitude on the
other. can the latter be readily quantified from the impulse response of
the filter, or even better all the filters in the chain put together.

AFAIK the most common tool for designing digital filters for digital audio
is Matlab, which can apparently synthesize and simulate a wide variety of
options. It is possible to control the ringing at the expense of filter
slopes, and it is possible to position the ringing either before or after
the main peak of the stimulus at the expense of desirable attributes such as
being minimum phase or linear phase.

I have become confused as to whether the ringing is at one frequency (eg
the -6db point) or whether it is in the entire range of the transition
band down to the negligible -100db or so.

AFAIK, the ringing is generally at the Nyquist frequency.

On Thursday, 22 September 2011 13:18:19 UTC+1, Jim Lesurf wrote:


Afraid not. I've put this together in my own mind from thinking about the
topic and looking at various sources over the years. [1] Some texts do look
at aspects, but I'm not sure which would be helpful, or even available
these days. That said, for a general but thorough into to digital audio I
would recommend 'The Art of Digital Audio' by Watkinson. (Focal Press)

Apart from that, one day my round tuit supply may allow me to put up more
on such topics, and even get to making my own book on the IT side of this
up on the web. However the book is aimed at undergrad level, so isn't
necessarily an easy read as an into.




I may have a look at Watkinson. I have Pohlmann which was recommended to me by John Atkinson, and which I have found useful on a number of topics.


Which is your book which you refer to? I do have a copy of your Information and Measurement (2nd ed), but could not see where this issue is considered.. I am tempted to say that i had not found this book intimidatingly difficult (although I have not read it all from cover the cover yet), but I suspect that if i did so I would receive the reply that I had plainly not understood a word.


Just holding onto the point that
"The problem in analysing the effect of a filter (or the filter chain)
seems to be in working out which part of the wiggling simply represents
the reconstruction of signal between sample values and which bit
represents an artifact not in the signal."


....and returning to reference to http://www.cirlinca.com/include/aes97ny.pdf
the reference in that article to energy dispersion of the transient response (energy against time) is somewhat misleading as an indicator of "time smear" because it does not represent the time smear as such but the energy of the impulse response wiggle some of which *might* be time smear.

At least this would be the case if we were looking at the impulse response of the reconstruction filter.

Can I just clarify- are you saying that one equally can't deduce the extent to which wiggling in an anti alias filter's impulse response would represent time smear because there is no defined pre AA filter signal which that impulse might represent and the result would vary within the set of signals which that impulse might represent to yield a range from 0-100% of the wiggling.

Thanks very much Arnie.


Could you point me to the DBTs which have been performed.


AS you will see from the above thread, I have become confused as to what the expression "ringing at" a particular frequency means.


From what Jim says it seems that he is suggesting that it doesn't mean anything other than that this is the frequency the cutting off above which generates the ringing.

I have come across the Sox manual on the net. Since i use a squeezebox for music and the AFAIK Sox program comes with it or at least works with it. It seems that i should be able to introduce pretty well any filtration I like including Minimum, linear and intermediate phase and filter slopes getting very close to vertical.

In fact Sox is currently used for sample rate conversion of everything up to 24/96 prior to application of DRC. I think this measn i have filters (minimum phase i think), at lots of points in the audible range.

"adamdea" wrote in message
news:28850714.1859.1316703150761.JavaMail.geo-discussion-forums@yqaa12...

Could you point me to the DBTs which have been performed.

Check the archives of the Hydrogen Audio Froum.

AS you will see from the above thread, I have become confused as to what
the expression "ringing at" a particular frequency means.

Seems simple enough. If a waveform has ringing on it, you isolate a segment
of it that has the strongest ringing and measure the frequency of the
ringing, say with a FFT analyzer.

From what Jim says it seems that he is suggesting that it doesn't mean
anything other than that this is the frequency the cutting off above which
generates the ringing.

Since the design frequency of filters like these is just below the Nyquist
frequency, its all the same.

I have come across the Sox manual on the net. Since i use a squeezebox for
music and the AFAIK Sox program comes with it or at least works with it. It
seems that i should be able to introduce pretty well any filtration I like
including Minimum, linear and intermediate phase and filter slopes getting
very close to vertical.

It is what it is.

In fact Sox is currently used for sample rate conversion of everything up
to 24/96 prior to application of DRC. I think this measn i have filters
(minimum phase i think), at lots of points in the audible range.

I think that any well-deisgned program like Sox is going to put the filters
for 44 KHz sampling just above the audible range.

In article
4989401.987.1316702605902.JavaMail.geo-discussion-forums@yqjw35,
adamdea
wrote:
On Thursday, 22 September 2011 13:18:19 UTC+1, Jim Lesurf wrote:

Which is your book which you refer to? I do have a copy of your
Information and Measurement (2nd ed), but could not see where this issue
is considered.

That is the book. It doesn't explicitly deal with the details of ringing.
But should be useful for the IT basics, and approachs the topic from the
kind of viewpoint I've been outlining. e.g. that defining how the stream of
samples was created matters and that what you see from reconstruction may
reflect that.

Also, if you have the 2nd edition (paperback) it may be useful to look at
page 257. The figure and surrounding discussion. This deals with
interferometry and imposed coherence. But it has underpinning similarities
to the relationship between time and frequency responses in systems like
the audio ones we are discussing.

Note that a wideband signal and system generates an interferogram
(equivalent to an impulse response) that is like a spike of a given
scale width. The wider the bandwidth, the narrower this peak. It is
actually just another manifestation of the same basic signal theory.
And the response depends on all the items in the 'signal chain' inc
how the input was generated. Note also that no sampling need be
involved. This isn't only a 'digital sampling' behaviour. It is
far more general.


I am tempted to say that i had not found this book
intimidatingly difficult (although I have not read it all from cover the
cover yet), but I suspect that if i did so I would receive the reply
that I had plainly not understood a word.

I'm happy to accept that anyone who buys my books must be very perceptive
and have good taste. :-)


Just holding onto the point that "The problem in analysing the effect
of a filter (or the filter chain) seems to be in working out which part
of the wiggling simply represents the reconstruction of signal between
sample values and which bit represents an artifact not in the signal."


...and returning to reference to
http://www.cirlinca.com/include/aes97ny.pdf the reference in that
article to energy dispersion of the transient response (energy against
time) is somewhat misleading as an indicator of "time smear" because it
does not represent the time smear as such but the energy of the impulse
response wiggle some of which *might* be time smear.

At least this would be the case if we were looking at the impulse
response of the reconstruction filter.

Again, I'd be wary of terms like "time smear" because the definition the
writer/speaker uses may not be exactly in line with the definitions used by
others.

But the basic point is the the scale-size of the 'width' (in time) of the
filter's impulse response is inversely proportional to the bandwidth of the
filter. Note that this isn't the HF cutoff, but the bandwith. Although for
a LPF the two values are essentially the same. The *shape* of the impulse
response depends on the details of the filter's frequency response (inc
phase effects). But the time-width scales with the bandwidth. They are two
sides of the same coin in signal theory terms.

Can I just clarify- are you saying that one equally can't deduce the
extent to which wiggling in an anti alias filter's impulse response
would represent time smear because there is no defined pre AA filter
signal which that impulse might represent and the result would vary
within the set of signals which that impulse might represent to yield a
range from 0-100% of the wiggling.

Not quite. I'm saying that the impulse response as normally measured for
tests on DACs takes a specific input - usually one non-zero sample value
with all pre and post samples zero. But you would need to examine the
*recording* system to determine what would cause *that* to produce such a
series of sample values. So lacking that knowlege, you can only make
assumptions about how close the 'output' looks like what was recorded.

However *if* you *know* the recordings was made with a perfect sinc filter
that suits Nyquist, you can say that such an output matches what the
recording defined.

The main snag is that you may not know this, and the details may vary from
one recording to another.

The minor snag is that no practical filer can truly give you a perfect
filter of this type since it requires an infinite scope in time! :-)

A sinc function has 'wiggles' all the way to +/- infinite time. So you
can't have such a filter in reality unless you started the recording at the
Big Bang and are willing to wait until Heat Death to form your conclusions.
:-)

So in reality all the FIR filters that are said to be 'sinc like' use a
time-limited impulse response as the basis of the filter. That alters the
behaviour. This means the 'wiggles' do fall to nothing beyond a finite
range only because/when the real filters are *not* 'perfect' in theory.
Again, engineering just works on doing what seems 'good enough' for the
task in hand. But anyone can argue about where that boundary sits.

Slainte,

Jim

--
Please use the address on the audiomisc page if you wish to email me.
Electronics http://www.st-and.ac.uk/~www_pa/Scot...o/electron.htm
Armstrong Audio http://www.audiomisc.co.uk/Armstrong/armstrong.html
Audio Misc http://www.audiomisc.co.uk/index.html

In article
28850714.1859.1316703150761.JavaMail.geo-discussion-forums@yqaa12,
adamdea wrote:

In fact Sox is currently used for sample rate conversion of everything
up to 24/96 prior to application of DRC. I think this measn i have
filters (minimum phase i think), at lots of points in the audible range.


FWIW I use SOX for various purposes, and find it works well. However I've
not tried it as a base for different filter functions.

Slainte,

Jim

--
Please use the address on the audiomisc page if you wish to email me.
Electronics http://www.st-and.ac.uk/~www_pa/Scot...o/electron.htm
Armstrong Audio http://www.audiomisc.co.uk/Armstrong/armstrong.html
Audio Misc http://www.audiomisc.co.uk/index.html

In article , Arny
Krueger
wrote:


In fact Sox is currently used for sample rate conversion of everything
up to 24/96 prior to application of DRC. I think this measn i have
filters (minimum phase i think), at lots of points in the audible
range.

I think that any well-deisgned program like Sox is going to put the
filters for 44 KHz sampling just above the audible range.

I'd need to check man for sox. But IIRC as well as providing some
'standard' filters, sox lets you tweak their bandwidth, etc, or apply
user-defined filters. If so, it should be quite a flexible way to
experiment with the properties and effects of a wide range of filtering
and resampling.

Slainte,

Jim

--
Please use the address on the audiomisc page if you wish to email me.
Electronics http://www.st-and.ac.uk/~www_pa/Scot...o/electron.htm
Armstrong Audio http://www.audiomisc.co.uk/Armstrong/armstrong.html
Audio Misc http://www.audiomisc.co.uk/index.html

On Sep 22, 5:38Â*pm, Jim Lesurf wrote:
In article
28850714.1859.1316703150761.JavaMail.geo-discussion-forums@yqaa12,
Â* Â*adamdea wrote:

In fact Sox is currently used for sample rate conversion of everything
up to 24/96 prior to application of DRC. I think this measn i have
filters (minimum phase i think), at lots of points in the audible range..

FWIW I use SOX for various purposes, and find it works well. However I've
not tried it as a base for different filter functions.

I notice that the manual contained the tantalizing words

"All resamplers use �lters that can sometimes create ‘echo’ (a.k.a.
‘ringing’) artefacts with transient signals such as those that occur
with ‘�nger snaps’ or other highly percussive sounds. Such
artefacts are much more noticeable to the human ear if they occur
before the transient (‘pre-echo’)
than if they occur after it (‘post-echo’). Note that frequency of any
such artefacts is related to the
smaller of the original and new sampling rates but that if this is at
least 44.1kHz, then the artefacts
will lie outside the range of human hearing."

From my inept questioning and faltering understanding of your answers
it appears that this is probably true (ie the artefacts probably can't
be heard) although not because the artefacts are simply composed of
frequencies we can't hear, but because they represent the time domain
effect of changes in frequencies we can't hear (?). Or maybe
something like a beat frequency where the beating is too fast to
hear(?)

Anyway I am going to read chapter 24 of your excellent book.
Carefully.

On Sep 22, 5:36*pm, Jim Lesurf wrote:
In article
4989401.987.1316702605902.JavaMail.geo-discussion-forums@yqjw35,
adamdea

Again, engineering just works on doing what seems 'good enough' for the
task in hand. But anyone can argue about where that boundary sits.

I am absolutely sure this is the nub of the matter. I am trying to
understand however how we can analyse what "good enough" might mean in
the context of ringing behaviour.
Ignoring for the moment empirical experiments to determine whether
people can detect the effect of filters, how do we set about trying to
analyse the phenomenon to establish how much ringing and "at" what
frequency (whatever that means) is good enough? That was really where
I was coming from in my first post.

In article
,
adamdea wrote:
On Sep 22, 5:38 pm, Jim Lesurf wrote:


FWIW I use SOX for various purposes, and find it works well. However
I've not tried it as a base for different filter functions.

I notice that the manual contained the tantalizing words

"All resamplers use filters that can sometimes create 'echo' (a.k.a.
'ringing') artefacts with transient signals such as those that occur
with 'finger snaps' or other highly percussive sounds. Such artefacts are
much more noticeable to the human ear if they occur before the transient
('pre-echo') than if they occur after it ('post-echo'). Note that
frequency of any such artefacts is related to the smaller of the
original and new sampling rates but that if this is at least 44.1kHz,
then the artefacts will lie outside the range of human hearing."

From my inept questioning and faltering understanding of your answers it
appears that this is probably true (ie the artefacts probably can't be
heard)

The problem with the quote from sox is that it examples a common behaviour
in audio. It presents a mix of fact, opinion, and assumed 'fact' without
clear distinctions as as sweeping generalisations. This makes it very hard
for the unwary reader to tease out reliable portions from guesswork or
error. Or assertions that may be correct in some cases, but not others.

Slainte,

Jim

--
Please use the address on the audiomisc page if you wish to email me.
Electronics http://www.st-and.ac.uk/~www_pa/Scot...o/electron.htm
Armstrong Audio http://www.audiomisc.co.uk/Armstrong/armstrong.html
Audio Misc http://www.audiomisc.co.uk/index.html