[Frank Stearns]

Are we assured that a typical digital high pass filter, say of the type found in the
EQ of Protools 8 and later, removes DC offset?

There's a lengthy back story here which for brevity sake I'll not get into, but it
seems that a digital HPF is not removing DC offset -- at least in PT.

I'm in the process of setting up some tests to prove it one way or the
other for sure, but am swamped at the moment.

Curious if any who are well-versed in DSP might say one of the following:

- that's true, a typical digital HPF will not remove DC offset and here's why....

- only a poorly designed digital HPF fails to remove DC offset.

- you're out of your mind, of course by definition ANY digital HPF removes DC
offsets.

- (something else entirely.)


Curious folks want to know.....

Thanks in advance,
Frank
Mobile Audio
--

[Arny Krueger[_4_]]

"Frank Stearns" wrote in message
tion...
Are we assured that a typical digital high pass filter, say of the type
found in the
EQ of Protools 8 and later, removes DC offset?

There's a lengthy back story here which for brevity sake I'll not get
into, but it
seems that a digital HPF is not removing DC offset -- at least in PT.

PT has historically had some of the dumbest technical errors that I've ever
heard of, but to the best of my knowlege, they are generally fixed.

I'm in the process of setting up some tests to prove it one way or the
other for sure, but am swamped at the moment.

Save time and forget about PT?

Curious if any who are well-versed in DSP might say one of the following:

- that's true, a typical digital HPF will not remove DC offset and here's
why....

False.

- only a poorly designed digital HPF fails to remove DC offset.

True.

- you're out of your mind, of course by definition ANY digital HPF removes
DC
offsets.

Mostly true, but never underestimate the human ability to screw up.

In the digital domain, there are at least two ways to remove DC offsets, and
a HPF is just one of them. The second method is to calculate the average
value of the wave and subtract it from each and every sample. Good DAW
software offers both alternatives.

[Scott Dorsey]

Frank Stearns wrote:
Are we assured that a typical digital high pass filter, say of the type found in the
EQ of Protools 8 and later, removes DC offset?

I don't know about Protools. But a real high pass filter will remove DC
offset. There's no obvious error I can think of in a dsp filter that would
cause it not to.

There's a lengthy back story here which for brevity sake I'll not get into, but it
seems that a digital HPF is not removing DC offset -- at least in PT.

What makes you think so? The waveform not be centered around zero all the
time, sometimes that's just the way the waveform is.

Take every sample, sum them all together... and you won't necessarily get
zero either.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Neil Gould]

Scott Dorsey wrote:
Frank Stearns wrote:
Are we assured that a typical digital high pass filter, say of the
type found in the EQ of Protools 8 and later, removes DC offset?

I don't know about Protools. But a real high pass filter will remove
DC offset. There's no obvious error I can think of in a dsp filter
that would cause it not to.

There's a lengthy back story here which for brevity sake I'll not
get into, but it seems that a digital HPF is not removing DC offset
-- at least in PT.

What makes you think so? The waveform not be centered around zero
all the time, sometimes that's just the way the waveform is.

Take every sample, sum them all together... and you won't necessarily
get zero either.

Well... this is an interesting thought exercise, if nothing else. I would
think DC offset would be represented either as a steady-state level above or
below zero when no AM is present or as an asymmetrical waveform. If so,
perhaps DAW software "removes" DC offset in the absence of a steady-state
level by re-centering the waveform based on an average of the waveform
values, and that creates an opportunity for error, especially for asymmetric
waveforms lacking DC offsets.

--
Neil

[Peter Larsen[_3_]]

Neil Gould wrote:

Scott Dorsey wrote:

What makes you think so? The waveform not be centered around zero
all the time, sometimes that's just the way the waveform is.

Take every sample, sum them all together... and you won't necessarily
get zero either.

Well... this is an interesting thought exercise, if nothing else. I
would think DC offset would be represented either as a steady-state
level above or below zero when no AM is present or as an asymmetrical
waveform. If so, perhaps DAW software "removes" DC offset in the
absence of a steady-state level by re-centering the waveform based on
an average of the waveform values, and that creates an opportunity
for error, especially for asymmetric waveforms lacking DC offsets.

Re-centering would be an error. The mechanism is that the unlinearity of air
causes 2 harmonic distortion and second harmonic distortion causes
asymmetry. The asymmetry is in itself audible as an ""intonation or
chord-weight change"": take a recording of asymmetric vox humana or trumpet
and invert it and do something untraditional: listen! - or take a recording
of a speaker with varying asymmetry and listen for the difference in
perceived voice tonality with difference in asymmetry.

The asymmetry is "known and described in the literature" and it is an
established practice, at least for those that run AM receivers, to insert a
contraption in the audio chain that watches signal asymmetry and inverts
signal polarity to maintain positive asymmetry for optimum transmitter
modulation. The Elton John and Kiki Dee duet was bad news for those systems
.... one singer on each side on a fig 8, my recollection of Studio Sound is
that it was an AKG "the everchanging model".

What is less easily agreed on is that some, myself included, feel that
getting the overall polarity right is important for imaging and perspective.
Interestingly however it appears that just as some types of microphones are
more likely to deliver asymmetric audio than others also the audibility of
it varies with amount of second harmonic distortion added by the playback
transducers.

My observations - I do NOT want to call them "findings" - are that high
quality omni microphones tend to deliver recordings that are more symmetric
than some cardioids, applause does however always seem to have positive
asymmetry, also high pass filtering in post and/or the use of multiband
processing does tend to de-naturalize the waveform and can cause the
polarity to appear inverted.

Kind regards

Peter Larsen

[Neil Gould]

Peter Larsen wrote:
Neil Gould wrote:

Scott Dorsey wrote:

What makes you think so? The waveform not be centered around zero
all the time, sometimes that's just the way the waveform is.

Take every sample, sum them all together... and you won't
necessarily get zero either.

Well... this is an interesting thought exercise, if nothing else. I
would think DC offset would be represented either as a steady-state
level above or below zero when no AM is present or as an asymmetrical
waveform. If so, perhaps DAW software "removes" DC offset in the
absence of a steady-state level by re-centering the waveform based on
an average of the waveform values, and that creates an opportunity
for error, especially for asymmetric waveforms lacking DC offsets.

Re-centering would be an error. The mechanism is that the unlinearity
of air causes 2 harmonic distortion and second harmonic distortion
causes asymmetry. The asymmetry is in itself audible as an
""intonation or chord-weight change"": take a recording of asymmetric
vox humana or trumpet and invert it and do something untraditional:
listen! - or take a recording of a speaker with varying asymmetry and
listen for the difference in perceived voice tonality with difference
in asymmetry.

The asymmetry is "known and described in the literature" and it is an
established practice, at least for those that run AM receivers, to
insert a contraption in the audio chain that watches signal asymmetry
and inverts signal polarity to maintain positive asymmetry for
optimum transmitter modulation. The Elton John and Kiki Dee duet was
bad news for those systems ... one singer on each side on a fig 8, my
recollection of Studio Sound is that it was an AKG "the everchanging
model".

What is less easily agreed on is that some, myself included, feel that
getting the overall polarity right is important for imaging and
perspective. Interestingly however it appears that just as some types
of microphones are more likely to deliver asymmetric audio than
others also the audibility of it varies with amount of second
harmonic distortion added by the playback transducers.

My observations - I do NOT want to call them "findings" - are that
high quality omni microphones tend to deliver recordings that are
more symmetric than some cardioids, applause does however always seem
to have positive asymmetry, also high pass filtering in post and/or
the use of multiband processing does tend to de-naturalize the
waveform and can cause the polarity to appear inverted.

I completely agree with your observations, which is why I'm curious about
trying to remove DC offset in post-processing if there is no point of
reference such as a steady state level without AM signal. Since an xformer
balanced input won't pass DC, I wondered how likely it would be that DC
offset would be passed through the A/D. So, in either case, HPF is not much
of a solution.

--
best,

Neil

[Arny Krueger[_4_]]

"Neil Gould" wrote in message
...

I completely agree with your observations, which is why I'm curious about
trying to remove DC offset in post-processing if there is no point of
reference such as a steady state level without AM signal. Since an xformer
balanced input won't pass DC, I wondered how likely it would be that DC
offset would be passed through the A/D. So, in either case, HPF is not
much
of a solution.

If you are editing audio, and any of the file(s) involved have a DC offset,
there is a very good chance of introducing a click when you cut and paste.
Not just theory, but a fact that many of us who actually edit audio (and not
just pontificate about it) have unfortunately experienced.

You can minimize clicks and pops due to editing by only cutting on zero
crossings, but that is not a 100% solution. You can also minimize clicks and
pops by merging the file segments with what amounts to be a cross fade or
other smoothing technique, but again these are not 100% solutions. Cross
fades and other smoothing techniques essentially turn the clicks and pops
into thumps. The good news is that a thump that is composed of only very low
frequencies can become very hard to hear.

Having a file whose DC component approaches zero over shorter periods of
time is therefore a good thing, because it helps minimize the possibility of
introducing thumps, clicks and pops when you are editing it.

If you apply a HPF to a file, you are ensuring that the average or DC
component of the file approaches zero over a period of time that decreases
as you increase the cut-off frequency. If you zero out the files average
value, then the DC component of the file is only zero for the exact segment
of the file that you processed.

[Scott Dorsey]

Neil Gould wrote:
I completely agree with your observations, which is why I'm curious about
trying to remove DC offset in post-processing if there is no point of
reference such as a steady state level without AM signal. Since an xformer
balanced input won't pass DC, I wondered how likely it would be that DC
offset would be passed through the A/D. So, in either case, HPF is not much
of a solution.

Most DC offset problems are actually caused by an imbalance in the A/D
itself.

Since DC offset is in fact a shift in the quiescent level, it means that
cutting in and out of silent parts will cause an abrupt change in level,
ie. a pop.

A good fix for this is low-pass filtering, ie. offset removal.

Another issue with offset is that it eats a tiny bit of your dynamic range
up if your waveform is symmetric.... and in the age of trying to squeeze
every bit of level out, making sure your dc offset is eliminated is a way
to get a little extra level for free.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Les Cargill[_4_]]

Peter Larsen wrote:
Neil Gould wrote:

Scott Dorsey wrote:

What makes you think so? The waveform not be centered around zero
all the time, sometimes that's just the way the waveform is.

Take every sample, sum them all together... and you won't necessarily
get zero either.

Well... this is an interesting thought exercise, if nothing else. I
would think DC offset would be represented either as a steady-state
level above or below zero when no AM is present or as an asymmetrical
waveform. If so, perhaps DAW software "removes" DC offset in the
absence of a steady-state level by re-centering the waveform based on
an average of the waveform values, and that creates an opportunity
for error, especially for asymmetric waveforms lacking DC offsets.

Re-centering would be an error. The mechanism is that the unlinearity of air
causes 2 harmonic distortion and second harmonic distortion causes
asymmetry. The asymmetry is in itself audible as an ""intonation or
chord-weight change"": take a recording of asymmetric vox humana or trumpet
and invert it and do something untraditional: listen! - or take a recording
of a speaker with varying asymmetry and listen for the difference in
perceived voice tonality with difference in asymmetry.


One whale of a lot depends on what the recording is to be used for. If
it's a solo thing, and it will neither be spliced or edited - then
maybe.

If it is to be mixed or edited, then no. I think you pretty much have
to zero it out. I have not examined in detail what CoolEdit does to
eliminate DC offset, but if there's much at all, I take it out.
Surprisingly, it almost never comes up.

The asymmetry is "known and described in the literature" and it is an
established practice, at least for those that run AM receivers, to insert a
contraption in the audio chain that watches signal asymmetry and inverts
signal polarity to maintain positive asymmetry for optimum transmitter
modulation. The Elton John and Kiki Dee duet was bad news for those systems
... one singer on each side on a fig 8, my recollection of Studio Sound is
that it was an AKG "the everchanging model".

What is less easily agreed on is that some, myself included, feel that
getting the overall polarity right is important for imaging and perspective.
Interestingly however it appears that just as some types of microphones are
more likely to deliver asymmetric audio than others also the audibility of
it varies with amount of second harmonic distortion added by the playback
transducers.


I can believe this.


When i have played with asymmetric distortion as an effect, I always
remove DC offset. This makes rather unexpected and radical changes in
waveform, so I think the overall premise that DC offset is to the
good holds, when you're doing "documentary" work.

My observations - I do NOT want to call them "findings" - are that high
quality omni microphones tend to deliver recordings that are more symmetric
than some cardioids, applause does however always seem to have positive
asymmetry, also high pass filtering in post and/or the use of multiband
processing does tend to de-naturalize the waveform and can cause the
polarity to appear inverted.

Bizarre.


Kind regards

Peter Larsen

snip
--
Les Cargill

[Peter Larsen[_3_]]

Les Cargill wrote:

When i have played with asymmetric distortion as an effect, I always
remove DC offset.

DC offset and asymmetry are different concepts, you will have asymmetry even
if the "silence" line is on the midline and it will remain even if you
remove the DC offset as Cooledit does it, ie. a "few hz" highpass.

This makes rather unexpected and radical changes in
waveform, so I think the overall premise that DC offset is to the
good holds, when you're doing "documentary" work.

Whatever you do it is - in my opinion - good practice to remove dc offset,
CoolEdit can do it when accepting input, but doesn't if the input is a file,
then remembering setting the toggle once and for all doesn't do.

I had stopped highpass filtering in post, because some clarity is lost for
every single modification of the audio - I have changed workflow and aim for
doing all the post, even for a stereo recording, in the Audition 3
multitrack worksspace, but after have a "thermic air rise" surprise due to
event lights recently it is back in the workflow.

Always remember to wonder whenever the recording meters are more stationary
than suggested by the audible signal envelope! - what made it "an issue"
was:

1 part microphones without internal highpass,
1 part microphones without windscreen and some 12 feet above the floor,
1 part "cardioid compensating bass boost" and finally
1 part a friends transmission line-loudspeakers that - at a guess -
enhanced the VLF by adding second harmonic distortion and then hitting a
room resonance in his listening room and air noise, it is only very
marginally audible on my big un's that just play it back as it is, a very
deep rumble.

High pass filtering is also in most contexts a good practice because it
keeps small loudspeaker membrane excursion low, and thus reduces actual
playback distortion for those that do not have full range loudspeakers.

Listening on extra playback systems is also a good practice.

Kind regards

Peter Larsen

[Richard Webb[_3_]]

On Thu 2011-Nov-10 09:20, Scott Dorsey writes:
I don't know about Protools. But a real high pass filter will
remove DC offset. There's no obvious error I can think of in a dsp
filter that would cause it not to.

Much as I thought as well.

There's a lengthy back story here which for brevity sake I'll not get into, but
it seems that a digital HPF is not removing DC offset -- at least in PT.

What makes you think so? The waveform not be centered around zero
all the time, sometimes that's just the way the waveform is.

Much as I always believed or was taught as well.

Take every sample, sum them all together... and you won't
necessarily get zero either.
--scott

tHIs is true, which begs the question if the problem is
something audible that will negatively impact listenability
or further downstream processing.

Regards,
Richard
--
| Remove .my.foot for email
| via Waldo's Place USA Fidonet-Internet Gateway Site
| Standard disclaimer: The views of this user are strictly his own.