Regarding DC removal filter.

On Aug 3, 9:09 pm, R.Lyons@_BOGUS_ieee.org (Rick Lyons) wrote:
>
> Hi Marcel,
>    I don't understand your DC-removal
> scheme.  As far as I know, there's no
> low-computation linear-phase way to
> remove (attenuate) DC.  If you have
> some sort of "neat" way to do so,
> we'd sure like to hear more about your
> method.

i don't think it's necessarily "neat", but using a very long CIC
moving-average filter to determine a DC component, and subtracting
that from a sample that is delayed precisely 1/2 the length of the
moving average filter is both linear-phase and low-computation.  no?

long delay lines and long CIC filters are low compuation and if memory
is cheap and long throughput delay is acceptable, this is cheep and
effective.  (but not really worthy of an IEEE SP mag article.)

r b-j

On Sat, 04 Aug 2007 18:18:11 -0000, robert bristow-johnson
<rbj@audioimagination.com> wrote:

>On Aug 3, 9:09 pm, R.Lyons@_BOGUS_ieee.org (Rick Lyons) wrote:
>>
>> Hi Marcel,
>>    I don't understand your DC-removal
>> scheme.  As far as I know, there's no
>> low-computation linear-phase way to
>> remove (attenuate) DC.  If you have
>> some sort of "neat" way to do so,
>> we'd sure like to hear more about your
>> method.
>
>i don't think it's necessarily "neat", but using a very long CIC
>moving-average filter to determine a DC component, and subtracting
>that from a sample that is delayed precisely 1/2 the length of the
>moving average filter is both linear-phase and low-computation.  no?
>
>long delay lines and long CIC filters are low compuation and if memory
>is cheap and long throughput delay is acceptable, this is cheep and
>effective.  (but not really worthy of an IEEE SP mag article.)
>
>r b-j

Hi Robert,
  Ah ha.  OK, I see what you mean.

My guess is that such a process wouldn't 
work too well at all.  It seems to me that 
those nasty "sidelobes" of a CIC filter's 
freq response (the first one being -13 db 
relative to the mainlobe value) will cause 
very unpleasant passband ripple in the 
spectrum of the signal at the output of 
the subtractor.

I'll have to experiment with this idea.

Thanks Robert,
[-Rick-]

On Aug 4, 4:16 pm, R.Lyons@_BOGUS_ieee.org (Rick Lyons) wrote:
> On Sat, 04 Aug 2007 18:18:11 -0000, robert bristow-johnson
>
> <r...@audioimagination.com> wrote:
> >On Aug 3, 9:09 pm, R.Lyons@_BOGUS_ieee.org (Rick Lyons) wrote:
>
> >>    I don't understand your DC-removal
> >> scheme.  As far as I know, there's no
> >> low-computation linear-phase way to
> >> remove (attenuate) DC.  If you have
> >> some sort of "neat" way to do so,
> >> we'd sure like to hear more about your
> >> method.
>
> >i don't think it's necessarily "neat", but using a very long CIC
> >moving-average filter to determine a DC component, and subtracting
> >that from a sample that is delayed precisely 1/2 the length of the
> >moving average filter is both linear-phase and low-computation.  no?
>
> >long delay lines and long CIC filters are low compuation and if memory
> >is cheap and long throughput delay is acceptable, this is cheep and
> >effective.  (but not really worthy of an IEEE SP mag article.)
>
..
>   Ah ha.  OK, I see what you mean.
>
> My guess is that such a process wouldn't
> work too well at all.

it's approximately the same as my guess.  assuming no non-linear
tricky things, a DC blocker is an HPF (with at least one zero at z=1
that is not cancelled by a pole at the same location).  i was just
addressing the design specs:

1. blocks DC,
2. phase-linear,
3. less than a zillion multipy-accumulates.

nothing was said about throughput latency or the frequency-domain
behavior around (but not equal to) DC.

if the CIC filter is really long, it will have a frequency response of
nearly zero at any frequency except those around zero (which is where
all the nasty sidelobe action is gonna be, the sidelobes higher up
will be so close to zero, i don't think when you subtract them from 1,
nobuddy cares).  and the CIC will have a constant delay (group or
phase delay, doesn't matter), relative to frequency; match that with
the delay of the unity gain tap.

>  It seems to me that
> those nasty "sidelobes" of a CIC filter's
> freq response (the first one being -13 db
> relative to the mainlobe value) will cause
> very unpleasant passband ripple in the
> spectrum of the signal at the output of
> the subtractor.

but, if you're DC blocking in the first place, you must expect some
behavior around DC that you are willing to tolerate.  it's a matter of
how close to DC you cannot tolerate attenuation of some degree, which
is a tradeoff with the delay of the filter, whether it is phase-linear
or not.

> I'll have to experiment with this idea.

it's just a dumb thing.  but it's just a spin off the concept of a DC
blocker being a device that subtracts DC (however it is estimated,
maybe a median filter would be better, who know?) from the signal.  a
CIC moving average filter, that is really the fucking-hell long, gives
you an estimate of the DC value for the signal around the middle of
the CIC delay line.  maybe not such a good estimate.

> Thanks Robert,

yer welcome, Rick.  things like this are cheap and i can forget about
it tomorrow.  i wouldn't wanna take on the responsibilities you do
with IEEE articles.  sorry if i crapped out on the other thing.

r b-j

On Sun, 05 Aug 2007 00:07:55 -0000, robert bristow-johnson
<rbj@audioimagination.com> wrote:

  (all of R B-J's good post snipped by Lyons)
>
>r b-j

Hi Robert,
   I agree with *everything* that you wrote about 
the DC-blocker scheme using a CIC filter.

I think this DC blocking technique, which is new to 
me, deserves further investigation.  Here's why:

Because a standard CIC filter has a 1st sidelobe 
level (in the freq domain) that's 13 dB below the 
mainlobe level, I naively thought that there'd be 
at least 13 dB ripple in the passband of the 
of the DC blocking "highpass" filter.  But I 
was wrong.  The DC blocker's passband ripple is much 
less than 13 dB.  So I believe this "CIC subtraction" 
DC blocking scheme deserves more investigation.

Robert, know that most of the times I run across 
some DSP technique that I've never seen before, 
and I think is kinda clever, I later find out that 
18 tech papers have been published on the method.
Ha ha. 

Too bad Marcel doesn't "pipe in" on our 
conversation here.

See Ya',
[-Rick-]

On Aug 5, 10:50 pm, R.Lyons@_BOGUS_ieee.org (Rick Lyons) wrote:
>
> Hi Robert,
>    I agree with *everything* that you wrote about
> the DC-blocker scheme using a CIC filter.

my goodness!

but remember, i did not "recommend it".  it just was there to satisfy
the 3 specs.  the delay (like any phase-linear filter that does some
serious filtering) might kill you.

> Robert, know that most of the times I run across
> some DSP technique that I've never seen before,
> and I think is kinda clever, I later find out that
> 18 tech papers have been published on the method.
> Ha ha.

Bob Adams (do you know him? or about him, Rick?) said something about
like this when he was presenting stuff, long ago at some AES thingie,
about sigma-delta conversion and noise shaping.  i think he suggested
that one look in those old Bell System Technical Journals from the 50s
and 60s.  you'll almost always find that your "novel idea" was novel,
but about 40 years ago and the novelty wasn't original to you.

but that doesn't stop someone like IVL from patenting it.

r b-j

r b-j wrote:
> Rick Lyons wrote:
>
>
>
> > Hi Robert,
> >    I agree with *everything* that you wrote about
> > the DC-blocker scheme using a CIC filter.
>
> my goodness!
>
> but remember, i did not "recommend it".  it just was there to satisfy
> the 3 specs.  the delay (like any phase-linear filter that does some
> serious filtering) might kill you.

Instead of using spectral inversion on the boxcar averaging filter to
subtract DC, you could use Avins' approach to convert the boxcar
filter into a DC blocker with no delay.

The trick is to assume that DC is constant (is this really a trick?),
and therefore it doesn't matter when you subtract it. You estimate the
DC with the running average filter h = 1/N [1 1 ... 1] (N ones), and
then subtract this estimate from the _current_ input value, ie.

y[n] = x[n] - (h[n] * x[n])
= x[n] * (1 - h[n]).

In effect, one applies the FIR filter h'[n] = 1/N [ (N-1)/N -1 -1 ...
-1]. But an efficient implementation of course uses the recursive
computation of the boxcar averageing filter, ie.

z[n] = z[n-1] - 1/N (x[n-N] + x[n])
y[n] = x[n] - z[n].

This is a minimum-phase filter. In a fixed-point machine, with N=2^n
for some integer n, this can be computed in a breeze. All you need is
the delay.

>
> > Robert, know that most of the times I run across
> > some DSP technique that I've never seen before,
> > and I think is kinda clever, I later find out that
> > 18 tech papers have been published on the method.
> > Ha ha.
>
> Bob Adams (do you know him? or about him, Rick?)

Of course Rick knows him. Remember the discussion about Adams' method
to interpolate every M-th sample?

> said something about
> like this when he was presenting stuff, long ago at some AES thingie,
> about sigma-delta conversion and noise shaping.  i think he suggested
> that one look in those old Bell System Technical Journals from the 50s
> and 60s.  you'll almost always find that your "novel idea" was novel,
> but about 40 years ago and the novelty wasn't original to you.
>
> but that doesn't stop someone like IVL from patenting it.

Who is IVL?

Regards,
Andor

On Aug 7, 2:57 am, Andor <andor.bari...@gmail.com> wrote:
> r b-j wrote:
..
> > Bob Adams (do you know him? or about him, Rick?)
>
> Of course Rick knows him. Remember the discussion about Adams' method
> to interpolate every M-th sample?

ooops.  to many dead brain cells.

> > said something about
> > like this when he was presenting stuff, long ago at some AES thingie,
> > about sigma-delta conversion and noise shaping.  i think he suggested
> > that one look in those old Bell System Technical Journals from the 50s
> > and 60s.  you'll almost always find that your "novel idea" was novel,
> > but about 40 years ago and the novelty wasn't original to you.
>
> > but that doesn't stop someone like IVL from patenting it.
>
> Who is IVL?

http://www.ivl.com/

http://www.patentstorm.us/patents/5301259.html
http://www.patentstorm.us/patents/5567901.html
http://www.patentstorm.us/patents/6336092.html

r b-j

robert bristow-johnson wrote:
> On Aug 7, 2:57 am, Andor <andor.bari...@gmail.com> wrote:
>> r b-j wrote:
> ..
>>> Bob Adams (do you know him? or about him, Rick?)
>> Of course Rick knows him. Remember the discussion about Adams' method
>> to interpolate every M-th sample?
> 
> ooops.  to many dead brain cells.
> 
>>> said something about
>>> like this when he was presenting stuff, long ago at some AES thingie,
>>> about sigma-delta conversion and noise shaping.  i think he suggested
>>> that one look in those old Bell System Technical Journals from the 50s
>>> and 60s.  you'll almost always find that your "novel idea" was novel,
>>> but about 40 years ago and the novelty wasn't original to you.
>>> but that doesn't stop someone like IVL from patenting it.
>> Who is IVL?
> 
> http://www.ivl.com/
> 
> http://www.patentstorm.us/patents/5301259.html
> http://www.patentstorm.us/patents/5567901.html
> http://www.patentstorm.us/patents/6336092.html

Those people are just plan evil. They are in the karaoke business. :-)

Steve

>
> > > said something about
> > > like this when he was presenting stuff, long ago at some AES thingie,
> > > about sigma-delta conversion and noise shaping.  i think he suggested
> > > that one look in those old Bell System Technical Journals from the 50s
> > > and 60s.  you'll almost always find that your "novel idea" was novel,
> > > but about 40 years ago and the novelty wasn't original to you.
>
> > > but that doesn't stop someone like IVL from patenting it.
>

Hello Robert, et. al.,

If the idea can be found in an old published journal (which predates
the patent), then the patent is effectively null and void.
A lot of application patents can be nullified if one does the right
research. A big part of a patent's being valid is its being novel and
not being obvious to someone skilled in the art. I don't know about
IVL's patents - do they have any novel content?

Clay

Clay wrote:
>>>> said something about
>>>> like this when he was presenting stuff, long ago at some AES thingie,
>>>> about sigma-delta conversion and noise shaping.  i think he suggested
>>>> that one look in those old Bell System Technical Journals from the 50s
>>>> and 60s.  you'll almost always find that your "novel idea" was novel,
>>>> but about 40 years ago and the novelty wasn't original to you.
>>>> but that doesn't stop someone like IVL from patenting it.
> 
> Hello Robert, et. al.,
> 
> If the idea can be found in an old published journal (which predates
> the patent), then the patent is effectively null and void.
> A lot of application patents can be nullified if one does the right
> research. A big part of a patent's being valid is its being novel and
> not being obvious to someone skilled in the art. I don't know about
> IVL's patents - do they have any novel content?

This is the theory. The high cost of litigation, and the quirky outcome 
of presenting even simple technical concepts to a court, makes the 
reality a whole different story.

Steve