Shifting audio signal 180 degrees out of phase
@attbi.com carl.reynolds@attbi.com
got my EE degree, and 25 years since I've used what I learned....
I need to take an audio signal and shift it 180 degrees out of phase so I
can re-combine it with the source signal and theoretically cancel each other
out. This is probably a matter of hooking an inductor in series, or
something rediculously simple as that, but I have no idea where to start
(what values to use for the inductor, etc..) I also need the phase shift to
be variable, so I can fine tune it.
Please be specific, I am old and brain-dead, and my math is almost as rusty
as my electronics.
joe
Kalman Rubinson
The passive inductor/capacitor network is frequency-sensitive. Easy
way today is to use an inverting opamp stage.
Kal
Gary Glaenzer
that would imply a time delay of 180 degrees, which is fine for a
fixed-frequency signal, but for all practical purposes impossible if you're
dealing with a mix of frequencies
you want to 'create a signal of opposite polarity ' as the original
and as has been suggested an inverting op-amp is cheap and easy
Regards,
G
"Kalman Rubinson" <k...@nyu.edu> wrote in message
news:cuomaucrst3j4g56a...@4ax.com...
Randy Yates
>
> you don't actually want to 'shift it 180 degrees out of phase'
Yes, you do.
> that would imply a time delay of 180 degrees,
No, it wouldn't, since time delay is in seconds and degrees is not
a unit compatible with seconds.
> which is fine for a
> fixed-frequency signal, but for all practical purposes impossible if you're
> dealing with a mix of frequencies
No, it's not impossible. It simply means that the time delay through the circuit
would not be constant with respect to frequency.
--
% Randy Yates % "...the answer lies within your soul
%% Fuquay-Varina, NC % 'cause no one knows which side
%%% 919-577-9882 % the coin will fall."
%%%% <ya...@ieee.org> % 'Big Wheels', *Out of the Blue*, ELO
http://personal.rdu.bellsouth.net/~yatesc
Gary Glaenzer
"Randy Yates" <ya...@ieee.org> wrote in message
news:3CAB8DEA...@ieee.org...
> Gary Glaenzer wrote:
> >
> > you don't actually want to 'shift it 180 degrees out of phase'
>
> Yes, you do.
no he doesn't.......he wants a signal to cancel the original.....which would
require one of opposite polarity
>
> > that would imply a time delay of 180 degrees,
>
> No, it wouldn't, since time delay is in seconds and degrees is not
> a unit compatible with seconds.
note the word 'shift' in his original post
BTW, delaying a sine wave a half-cycle (0.0005 seconds at 1 kHz) is the same
as shifting it 180 degrees, and as I go on to say:
> > which is fine for a
> > fixed-frequency signal, but for all practical purposes impossible if
you're
> > dealing with a mix of frequencies
>
> No, it's not impossible. It simply means that the time delay through the
circuit
> would not be constant with respect to frequency.
which is another way of saying that it would NOT be 180 degrees at all
frequencies...good grief, man, this isn't nuclear reactor design, its common
sense
Please, please re-read what his original intent was, then consider this:
Take as an example a wave consisting of a positive-going square wave
followed by negative-going half of a sine wave.
If you 'shift it 180 degrees', you get a negative-going half-sine wave
followed by a positive-going square wave.
adding this to the original will NOT cancel the original
on the other hand, if you take the original, run it thru an inverter ('of
opposite polarity' at the output), then add it to the original, it WILL
cancel the original, because it's a mirror image of the original.
The problem stems with the way he worded his intentions:
> >I need to take an audio signal and shift it 180 degrees out of phase so I
> >can re-combine it with the source signal and theoretically cancel each
other
> >out.
What he wants to do is to create a mirror image of the original.
The problem is that '180 degrees out of phase' may be the same as 'of
opposite polarity' for a sine wave, but what he is trying to cancel is a
complex mix of frequencies, where a 180 dgree shift in phase will NOT be the
same as an opposite-polarity signal.
--
What happens if a big asteroid hits Earth? Judging from realistic
simulations involving a sledge hammer and a common laboratory frog,
we can assume it will be pretty bad. --- Dave Barry
@attbi.com carl.reynolds@attbi.com
for a moment longer (I'm totally embarased at having forgotten so much of my
training - I truely did have an ASEE in 1975) - Once my output is inverted
(by an inverting opamp), it will be combined with the original signal, which
will have traveled out to my stereo, through my speakers, and then across
about 30 feet of airspace to a microphone. After all of that travel, it
would no doubt be a few degrees behind the inverted signal (I would assume
less than 90 degrees) - can anyone give me a simple L&R or C&R combination
that I could use to "fine tune" the resulting lag in phase?
Thanks for your patience...
joe
"Gary Glaenzer" <nobul...@mchsi.com> wrote in message
news:sAMq8.166781$af7.81537@rwcrnsc53...
Richard D Pierce
carl.r...@attbi.com <carl.r...@attbi.com @attbi.com> wrote:
>Great insight, and many thanks, guys - If y'all don't mind holding my hand
>for a moment longer (I'm totally embarased at having forgotten so much of my
>training - I truely did have an ASEE in 1975) - Once my output is inverted
>(by an inverting opamp), it will be combined with the original signal, which
>will have traveled out to my stereo, through my speakers, and then across
>about 30 feet of airspace to a microphone. After all of that travel, it
>would no doubt be a few degrees behind the inverted signal (I would assume
>less than 90 degrees) - can anyone give me a simple L&R or C&R combination
>that I could use to "fine tune" the resulting lag in phase?
Uh, no, not even close.
Depending upon the frequency and such, it may well be several
THOUSAND degree out of phase. Further, the "delay" itself will
hardly be a single delay, but will have ALL of the various and
sundry added delays due to reverberation.
The problem you are trying to solve is essentially an
intractable one, most especially given the way you are trying to
solve it.
--
| Dick Pierce |
| Professional Audio Development |
| 1-781/826-4953 Voice and FAX |
| DPi...@world.std.com |
Kalman Rubinson
>Great insight, and many thanks, guys - If y'all don't mind holding my hand
>for a moment longer (I'm totally embarased at having forgotten so much of my
>training - I truely did have an ASEE in 1975) - Once my output is inverted
>(by an inverting opamp), it will be combined with the original signal, which
>will have traveled out to my stereo, through my speakers, and then across
>about 30 feet of airspace to a microphone. After all of that travel, it
>would no doubt be a few degrees behind the inverted signal (I would assume
>less than 90 degrees) - can anyone give me a simple L&R or C&R combination
>that I could use to "fine tune" the resulting lag in phase?
What are you trying to accomplish?
Kal
@attbi.com carl.reynolds@attbi.com
I am a computer programmer, and have written a voice-controlled MP3 player
(plays music files on the computer). When it is not playing an MP3, it
accepts voice commands without any problem. However, when an MP3 file is
playing, the program can't distinguish between my voice and the music that
is being played, and therefore it is difficult, if not impossible to get it
to respond to the voice commands. What I am hoping to do is to cancel out
the music, leaving only my voice being "heard" by the sound card, therefore
eliminating the sound card's "confusion".
My idea was to take the output of the sound card, before it is sent to the
speakers (or stereo), invert the signal, and then "mix" it with the
microphone input. That way, the music produced by the sound card would be
heterodyned with all of the sound picked up by the mike, thereby canceling
out the music.
"Kalman Rubinson" <k...@nyu.edu> wrote in message
news:4tbnauofmf4simrsd...@4ax.com...
@attbi.com carl.reynolds@attbi.com
The 30 feet of airspace would actually be more like 10 or 12 feet, and that
would be the wost-case schenario.
In the best case, and more important schenario, the un-inverted signal would
only travel through about 3 feet of cable, to a speaker that is only a foot
from the mike. I also wouldn't have to COMPLETELY cancel out the signal - I
have a lot of leaway concerning how much of the signal can be allowed to get
through...
"carl.r...@attbi.com" <carl.r...@attbi.com @attbi.com> wrote in
message news:uIQq8.162504$Yv2.51105@rwcrnsc54...
Richard D Pierce
carl.r...@attbi.com <carl.r...@attbi.com @attbi.com> wrote:
>I guess I should be more specific.
>
>I am a computer programmer, and have written a voice-controlled MP3 player
>(plays music files on the computer). When it is not playing an MP3, it
>accepts voice commands without any problem. However, when an MP3 file is
>playing, the program can't distinguish between my voice and the music that
>is being played, and therefore it is difficult, if not impossible to get it
>to respond to the voice commands. What I am hoping to do is to cancel out
>the music, leaving only my voice being "heard" by the sound card, therefore
>eliminating the sound card's "confusion".
>
>My idea was to take the output of the sound card, before it is sent to the
>speakers (or stereo), invert the signal, and then "mix" it with the
>microphone input. That way, the music produced by the sound card would be
>heterodyned with all of the sound picked up by the mike, thereby canceling
>out the music.
Welll, practically, you can't get there.
First, you have to compensate for the acoustic delay between the
speaker and the microphone. Second, you have to account for the
electrical delay due to the response of the speaker. This is
VERY difficult to even characterize, much less compensate for.
Third, you have to account for the room reverberation. And so on
and so forth.
The problem you're trying to solve is essentially the same as
real-time room/loudspeaker correction (since it deals with all
the same factors), and this is a problem which, even given VERY
expensive methods, is solved partially and poorky, at best.
You're FAR better off directly increasing the signal to noise
between voice and music to begin with.
Richard D Pierce
carl.r...@attbi.com <carl.r...@attbi.com @attbi.com> wrote:
>To clarify further:
>
>The 30 feet of airspace would actually be more like 10 or 12 feet, and that
>would be the wost-case schenario.
Okay, is it 10 or is it 12? Or is it, in fact, variable? If so,
you've got a HIGE problem to solve, because you have a direct
delay of 1 mS/foot to deal with.
>In the best case, and more important schenario, the un-inverted signal would
>only travel through about 3 feet of cable, to a speaker that is only a foot
>from the mike. I also wouldn't have to COMPLETELY cancel out the signal - I
>have a lot of leaway concerning how much of the signal can be allowed to get
>through...
The best case is merely very difficult to solve, ALL opther
cases are much harder,
I have a sneaking suspicion you don't really appreciate the
magnitude of the problem you're dealing with. Your simple
inversion simply will not work at all, because of all the
factors I have mentioned.
@attbi.com carl.reynolds@attbi.com
thirds, and wrapped it around the end of the mike, forming a sort of
funnel - increased my recognition acuracy DRASTICALLY! I guess all I really
need is a more uni-directional mike. I had considered that early on, but got
off on a tangent...
Anyway, thanks for all the info - I had been putting off this problem for
several months, not spending enough real effort on it in lue of other
issues. R U interested in a cool, voice-controlled MP3 player? Let me know
and I'll send you a copy when I get the bugs worked out...
Keepin' it Simple,
joe
"Richard D Pierce" <DPi...@TheWorld.com> wrote in message
news:Gu1Mz...@world.std.com...
Richard D Pierce
carl.r...@attbi.com <carl.r...@attbi.com @attbi.com> wrote:
>"Richard D Pierce" <DPi...@TheWorld.com> wrote in message
>news:Gu1Mz...@world.std.com...
>>
>> real-time room/loudspeaker correction (since it deals with all
>> the same factors), and this is a problem which, even given VERY
>> expensive methods, is solved partially and poorky, at best.
>>
>> You're FAR better off directly increasing the signal to noise
>> between voice and music to begin with.
>
I can.
>I took a Post-It Note, folded it in
>thirds, and wrapped it around the end of the mike, forming a sort of
>funnel - increased my recognition acuracy DRASTICALLY!
Yup, you increased the ratio of the voice signal to the music
("noise").
>I guess all I really
>need is a more uni-directional mike.
And if you maximize the ratio of the distance between the music
and the mike vs you and the mike, you'll increase it further.
Kalman Rubinson
>I can't believe what just happened - I took a Post-It Note, folded it in
>thirds, and wrapped it around the end of the mike, forming a sort of
>funnel - increased my recognition acuracy DRASTICALLY! I guess all I really
>need is a more uni-directional mike. I had considered that early on, but got
>off on a tangent...
Mebbe a contact mike on you would do even better.
Kal
Arny Krueger
"carl.r...@attbi.com" <carl.r...@attbi.com @attbi.com> wrote
in message news:uIQq8.162504$Yv2.51105@rwcrnsc54...
> I am a computer programmer, and have written a voice-controlled MP3
player
> (plays music files on the computer). When it is not playing an MP3,
it
> accepts voice commands without any problem. However, when an MP3
file is
> playing, the program can't distinguish between my voice and the
music that
> is being played, and therefore it is difficult, if not impossible
to get it
> to respond to the voice commands. What I am hoping to do is to
cancel out
> the music, leaving only my voice being "heard" by the sound card,
therefore
> eliminating the sound card's "confusion".
> My idea was to take the output of the sound card, before it is sent
to the
> speakers (or stereo), invert the signal, and then "mix" it with the
> microphone input. That way, the music produced by the sound card
would be
> heterodyned with all of the sound picked up by the mike, thereby
canceling
> out the music.
There are two important parameters to consider in a product like
this:
(1) How much cancellation do you want to obtain in dB?
(2) What is the highest frequency at which you wish to obtain
cancellation?
Basically you are proposing a system composed two paths
Path one is composed of an amplifier, wires, loudspeaker, some space,
and a microphone by which the undesired signal is sent. Let's say
that all this stuff taken together has a transfer function we will
call H1(t).
Path two is composed of an electrical network of fairly great
complexity that simulates H1(t). Let's call it H2(t).
What you want to do is match H2(t) to H1(t) in such a way that when
you mix the output of H2(t) with the output of your microphone, the
music is cancelled out to the desired degree.
This can work as long as:
(1) H1(t) is stable and consistent.
(2) H2(t) is a really good match for H1(t).
Obviously, H2(t) is going to have to include some delay. It's going
to have to simulate the in-room response of the amplifier and speaker
with pretty good amplitude and phase accuracy. The higher the
frequency you make this system operate at, the harder its going to be
to actually implement.
You can find information about the current state of this kind of
technology at
http://www.google.com/search?hl=en&q=active+noise+control .
Randy Yates
>[...]
> The problem you are trying to solve is essentially an
> intractable one,
Nah - it's done all the time in your POTS. It's called
an "echo canceller."
--
Randy Yates
DSP Engineer, Sony Ericsson Mobile Communications
Research Triangle Park, NC, USA
randy...@ericsson.com, 919-472-1124
Randy Yates
>
> "Randy Yates" <ya...@ieee.org> wrote in message
> news:3CAB8DEA...@ieee.org...
> > Gary Glaenzer wrote:
> > >
> > > you don't actually want to 'shift it 180 degrees out of phase'
> >
> > Yes, you do.
>
> no he doesn't.......he wants a signal to cancel the original.....which would
> require one of opposite polarity
"opposite polarity" and "180 degrees out of phase at all frequencies"
is the same thing.
> > > that would imply a time delay of 180 degrees,
> >
> > No, it wouldn't, since time delay is in seconds and degrees is not
> > a unit compatible with seconds.
>
> note the word 'shift' in his original post
Irrelevent. I'm addressing your statement, which makes no sense in the
same way that answering "5 feet" to your friend on the telephone
would make no sense had he asked you how long it would take for you
to arrive.
> BTW, delaying a sine wave a half-cycle (0.0005 seconds at 1 kHz) is the same
> as shifting it 180 degrees, and as I go on to say:
>
> > > which is fine for a
> > > fixed-frequency signal, but for all practical purposes impossible if
> you're
> > > dealing with a mix of frequencies
> >
> > No, it's not impossible. It simply means that the time delay through the
> circuit
> > would not be constant with respect to frequency.
>
> which is another way of saying that it would NOT be 180 degrees at all
> frequencies
No, it is not another way of saying such. It is saying just the opposite.
>...good grief, man, this isn't nuclear reactor design, its common
> sense
Right. It seems to be hard for some people to grasp anyway.
> The problem is that '180 degrees out of phase' may be the same as 'of
> opposite polarity' for a sine wave, but what he is trying to cancel is a
> complex mix of frequencies, where a 180 dgree shift in phase will NOT be the
> same as an opposite-polarity signal.
That is true. But, who said that "180 degress out of phase" HAD to mean
that there is only one frequency involved? You can shift EACH frequency 180 degrees out
of phase if you allow the time delay through the circuit to vary
as a function of frequency.
--
Randy Yates
DSP Engineer, Sony Ericsson Mobile Communications
Research Triangle Park, NC, USA
randy...@ericsson.com, 919-472-1124
.
Gary Glaenzer
opposed to 'of opposite polarity' you obviously cannot refute the example.
therefore, this conversation is over
enjoy your delusions of what '180 degrees out of phase' really means.
goodbye
"Randy Yates" <eus...@rtp.ericsson.com> wrote in message
news:3CACC8F8...@rtp.ericsson.com...
Randy Yates
>
> well, since you saw fit to delete my example of 180 degrees out of phase as
> opposed to 'of opposite polarity' you obviously cannot refute the example.
Oh, you mean this:
> Take as an example a wave consisting of a positive-going square wave
> followed by negative-going half of a sine wave.
>
> If you 'shift it 180 degrees', you get a negative-going half-sine wave
> followed by a positive-going square wave.
>
> adding this to the original will NOT cancel the original
???
Right you are. What you would be doing in this case is shifting all
frequencies in the input by the same time delay, namely, 1/2 of the
period of this composite signal.
However, there is another interpretation of "shifting by 180 degrees,"
and that is the one I used. In this interpretation, what is being
shifted by 180 degrees is not the composite signal in-total, but
rather each of the frequency components of the composite signal. Since
180 degrees at one frequency corresponds to a different amount of time
than 180 degrees at another frequency, what you would have is a circuit
(or "system") which passes information with different delays at
different frequencies. And that is, effectively, what a polarity
reversal does (believe it or not)!
Gary Glaenzer
> Gary Glaenzer wrote:
> >
> > well, since you saw fit to delete my example of 180 degrees out of phase
as
> > opposed to 'of opposite polarity' you obviously cannot refute the
example.
>
> Oh, you mean this:
>
> > Take as an example a wave consisting of a positive-going square wave
> > followed by negative-going half of a sine wave.
> >
> > If you 'shift it 180 degrees', you get a negative-going half-sine wave
> > followed by a positive-going square wave.
> >
> > adding this to the original will NOT cancel the original
>
> ???
what's with the '???'...........is it not obvious that shifting that
hypothetica waveform and then adding it to the original will not achieve
what the original poster desired, namely 'cancellation' ?
>
> Right you are. What you would be doing in this case is shifting all
> frequencies in the input by the same time delay, namely, 1/2 of the
> period of this composite signal.
>
> However, there is another interpretation of "shifting by 180 degrees,"
> and that is the one I used. In this interpretation, what is being
> shifted by 180 degrees is not the composite signal in-total, but
> rather each of the frequency components of the composite signal.
a nice trick if you can pull it off, and not what the original poster had in
mind at all
> Since
> 180 degrees at one frequency corresponds to a different amount of time
> than 180 degrees at another frequency, what you would have is a circuit
> (or "system") which passes information with different delays at
> different frequencies. And that is, effectively, what a polarity
> reversal does (believe it or not).
what we have here, obviously, is you have decided that the only way to
accomplish this is to 'shift all frequencies individually' by 180 degrees;
and I have elected to 'do a polarity reversal'.
you can agrue all you want that they are identical, but you'll never
convince me that they are.
so you keep telling yourself that it's simpler to simultaneously 'shift 180
degrees' a multitude of separate frequencies, and I'll settle for an
inverter, OK ?
Randy Yates
>
> "Randy Yates" <eus...@rtp.ericsson.com> wrote in message
> news:3CACE434...@rtp.ericsson.com...
> > Gary Glaenzer wrote:
> > >
> > > well, since you saw fit to delete my example of 180 degrees out of phase
> as
> > > opposed to 'of opposite polarity' you obviously cannot refute the
> example.
> >
> > Oh, you mean this:
> >
> > > Take as an example a wave consisting of a positive-going square wave
> > > followed by negative-going half of a sine wave.
> > >
> > > If you 'shift it 180 degrees', you get a negative-going half-sine wave
> > > followed by a positive-going square wave.
> > >
> > > adding this to the original will NOT cancel the original
> >
> > ???
>
> what's with the '???'
In English, this symbol used used to denote a question. It is the
terminator to the "Oh, you mean this:" question fragment that I placed
before the quoted material.
> > However, there is another interpretation of "shifting by 180 degrees,"
> > and that is the one I used. In this interpretation, what is being
> > shifted by 180 degrees is not the composite signal in-total, but
> > rather each of the frequency components of the composite signal.
>
> a nice trick if you can pull it off, and not what the original poster had in
> mind at all
You're pulling it off when you do a simple inversion. The thing I was
hopelessly trying to show you is that they are one and the same. Apparently
you're too boneheaded to do the analysis so I won't reveal to you the
key property of Fourier transforms that makes this so.
--
Randy Yates
DSP Engineer, Sony Ericsson Mobile Communications
Research Triangle Park, NC, USA
randy...@ericsson.com, 919-472-1124
.
Gary Glaenzer
>
> You're pulling it off when you do a simple inversion. The thing I was
> hopelessly trying to show you is that they are one and the same.
Apparently
> you're too boneheaded to do the analysis so I won't reveal to you the
> key property of Fourier transforms that makes this so.
> --
ah, the old ad hominem attack
you lose
goodbye
Chris Welsh
Polarity, as they pertain to Audio systems on <http://www.prosoundweb.com>
that should clear the issue...
Chris W
Jamie Hanrahan
wrote:
If the "attack" had been Randy's idea, you'd have something there. But it
wasn't, not at all. You said yourself:
> what we have here, obviously, is you have decided that the only way to
> accomplish this is to 'shift all frequencies individually' by 180 degrees;
> and I have elected to 'do a polarity reversal'.
>
> you can agrue all you want that they are identical, but you'll never
> convince me that they are.
i.e., you're either unable to understand, or unwilling to consider, the
proof that they're identical. If it's the former, well, "bonehead" is
perhaps a bit of a harsh word, but descriptive nonetheless.
If the latter, then you're not really a bonehead, but you've made a
conscious decision to act like one... which in my book is worse.
Either way, you've said that it's pointless to waste time presenting you
with the facts. So Randy said he wouldn't bother.
>you lose.
No, YOU lose. You're the one who has closed your mind to the possibility
of learning something you did not previously know. You know the old line,
"those who argue for their prejudices get to keep them"? Congratulations!
--- jeh
Gary Glaenzer
near the bottom
an excellent summation
goodbye
"Jamie Hanrahan" <j...@cmkrnl.com> wrote in message
news:pitpau0dqqfpcttub...@4ax.com...
amoli...@visi-dot-com.com
>> and that is the one I used. In this interpretation, what is being
>> shifted by 180 degrees is not the composite signal in-total, but
>> rather each of the frequency components of the composite signal.
>
>a nice trick if you can pull it off, and not what the original poster had in
>mind at all
It is exactly what the original poster had in mind.
>> Since
>> 180 degrees at one frequency corresponds to a different amount of time
>> than 180 degrees at another frequency, what you would have is a circuit
>> (or "system") which passes information with different delays at
>> different frequencies. And that is, effectively, what a polarity
>> reversal does (believe it or not).
>
>what we have here, obviously, is you have decided that the only way to
>accomplish this is to 'shift all frequencies individually' by 180 degrees;
>and I have elected to 'do a polarity reversal'.
>
>you can agrue all you want that they are identical, but you'll never
>convince me that they are.
Then you'll continue to be wrong.
Any sinusoid, shifted 180 degrees, is identical to
the negative of the original sinusoid, right?
Randy's point is that if you view the signal as a sum of
sinusoids, then shifting each sinusoid in the sum by 180 degrees
is exactly the same thing as negating ("inverting") each, which
is the same thing as inverting the original signal. This is basically
just arithmetic.
>so you keep telling yourself that it's simpler to simultaneously 'shift 180
>degrees' a multitude of separate frequencies, and I'll settle for an
>inverter, OK ?
Neither one is simpler, since they are identical operations.
For a signal that was a finite sum of sinusoids I suppose you could
IMPLEMENT inversion by carving the signal up into component
sinusoids, and shifting each by 180 degrees, and then adding, but
you'd be an idiot to do it. It's so much easier to multiply everything
in sight by -1.
Jamie Hanrahan
wrote:
>http://www.prosoundweb.com/live/labbest/speakercabs/speakercabs3.shtml
>
>near the bottom
>
>an excellent summation
No, it's wrong. A phase shift of 180 degrees IS the same as polarity
inversion, even for a complex waveform. That they're somehow different is
a widely held belief, but that doesn't make it right.
--- jeh
Michael R. Kesti
> That is true. But, who said that "180 degress out of phase" HAD to mean
> that there is only one frequency involved? You can shift EACH frequency 180 degrees out
> of phase if you allow the time delay through the circuit to vary
> as a function of frequency.
What happens when there is more than one frequency present?
--
========================================================================
Michael Kesti | "And like, one and one don't make
| two, one and one make one."
mke...@gv.net | - The Who, Bargain
Frank Johnson
ie. Two identical inserts with cardiod response, physically back to back
and electrically wired in antiphase.
Not perfect by any means but a recognised method used in noisy environments.
Frank.
r.crowley
> I can't believe what just happened - I took a Post-It Note, folded it in
> thirds, and wrapped it around the end of the mike, forming a sort of
> funnel - increased my recognition acuracy DRASTICALLY! I guess all I
really
> need is a more uni-directional mike. I had considered that early on, but
got
> off on a tangent...
Now you're on the right track!
People with resources infinitely greater than you posses have never
accomplished what you are attempting. (Ambient mitigation in uncontrolled
conditions) IMHO Mr. Pierce is correct, yours is an intractable problem.
Someone else suggested that modern telephone technology (DSP echo canceling)
was a possibility. But note that telephones are never used in acoustic
full-duplex as you are attempting. Telephones are either half-duplex (mic
and earphone isolated by your skull), or simplex (as in auto-VOX
speakerphones).
The echos produced by the contemporary telephone system with its mix of
analog and digital links is complex enough. The echo problem would be
intractable if acoustic echoes were thrown into the mix. Perhaps someday
we'll have a solution for it, but not likely this year (or next). And even
when it IS solved, it will likely be several years before the technology
(whatever it is) is available to us for experimentation.
Note that the commercial voice-command products use close-up (head-worn)
microphones for reliable pickup of the talker. Many use wireless links for
portability.
Remember that the crew talking to "Computer" (actually Majel's voice) on
StarTrek was science fiction!
StuWelwood
>was a possibility. But note that telephones are never used in acoustic
>full-duplex as you are attempting. Telephones are either half-duplex (mic
>and earphone isolated by your skull), or simplex (as in auto-VOX
>speakerphones).
>
>The echos produced by the contemporary telephone system with its mix of
>analog and digital links is complex enough. The echo problem would be
>intractable if acoustic echoes were thrown into the mix. Perhaps someday
>we'll have a solution for it, but not likely this year (or next). And even
>when it IS solved, it will likely be several years before the technology
>(whatever it is) is available to us for experimentation.
>
>Note that the commercial voice-command products use close-up (head-worn)
>microphones for reliable pickup of the talker. Many use wireless links for
>portability.
As someone who works at applying speech recognition technology to telephone
system applications, I can tell you that full duplex operation with echo
cancellation is done every day. This technology is far beyond that of
inexpensive voice command software. No special equipment is required at the
user end. I write applications that perform speaker-independent, continuous
speech, speech recognition applications every day that make use of (in fact,
require) full duplex capability. These use mature, commercially available
products that are fully developed.
As an example, see http://www.nuance.com/.
Stuart Welwood
http://members.aol.com/StuWelwood
Randy Yates
>
> "carl.r...@attbi.com" wrote ...
> > I can't believe what just happened - I took a Post-It Note, folded it in
> > thirds, and wrapped it around the end of the mike, forming a sort of
> > funnel - increased my recognition acuracy DRASTICALLY! I guess all I
> really
> > need is a more uni-directional mike. I had considered that early on, but
> got
> > off on a tangent...
>
> Now you're on the right track!
>
> People with resources infinitely greater than you posses have never
> accomplished what you are attempting. (Ambient mitigation in uncontrolled
> conditions) IMHO Mr. Pierce is correct, yours is an intractable problem.
>
> Someone else suggested that modern telephone technology (DSP echo canceling)
> was a possibility. But note that telephones are never used in acoustic
> full-duplex as you are attempting. Telephones are either half-duplex (mic
> and earphone isolated by your skull), or simplex (as in auto-VOX
> speakerphones).
>
> The echos produced by the contemporary telephone system with its mix of
> analog and digital links is complex enough. The echo problem would be
> intractable if acoustic echoes were thrown into the mix. Perhaps someday
> we'll have a solution for it, but not likely this year (or next). And even
> when it IS solved, it will likely be several years before the technology
> (whatever it is) is available to us for experimentation.
Bullshit. As a DSP developer for mobile phones (formerly for Ericsson and
now for Sony Ericsson), I can tell you this is EXACTLY what happens in a
mobile phone. The downlink signal is acoustically coupled from the loudspeaker
to the microphone. The result, if uncorrected, is echo. We and every other
mobile manufacturer must include echo canceller technology to remove this
echo. This technology is here now and has been for awhile.
There is NOTHING inherently difficult about this task. It is simply
a textbook problem of adaptive filtering.
--
% Randy Yates % "...the answer lies within your soul
%% Fuquay-Varina, NC % 'cause no one knows which side
%%% 919-577-9882 % the coin will fall."
%%%% <ya...@ieee.org> % 'Big Wheels', *Out of the Blue*, ELO
http://personal.rdu.bellsouth.net/~yatesc
Brad Griffis
will help clear a few of your issues up.
1. e^(j*pi) = e^(-j*pi) = -1
This should make it clear that inverting something (i.e. multiplying by -1)
is equivalent to a phase shift of pi radians or 180 degrees.
2. f(t-t0) <-------> F(w)e^(-jwt0)
This shows that for a CONSTANT time delay t0 you get a phase shift that is
PROPORTIONAL to w. Therefore in order to get all frequencies pi radians out
of phase you will need to have different time delays for each frequency.
3. f(t) = inf sum of sinusoids (i.e. Fourier series eqn)
Any periodic waveform can be represented by a Fourier series. An aperiodic
signal can be represented using Fourier transforms. Hence you can think of
any waveform as being the sum of a bunch of individual sine waves.
So if f(t) = inf sum of sinusoids then -f(t) = -(inf sum of sinusoids). By
distributing the negative to each sinusoid it should be clear that the
negative of a signal is just the sum of the component (minus) sine waves.
Throwing equations 1 and 2 into the mix you should know that each individual
sine wave being shifted 180 degrees corresponds to shifting each sine wave
180 degrees which corresponds to different time delays at each frequency.
I hope this clears confusion for somebody.
Brad Griffis
MSEE Student
University of Illinois at Urbana-Champaign
Randy Yates
>
> Randy Yates wrote:
>
> > That is true. But, who said that "180 degress out of phase" HAD to mean
> > that there is only one frequency involved? You can shift EACH frequency 180 degrees out
> > of phase if you allow the time delay through the circuit to vary
> > as a function of frequency.
>
> What happens when there is more than one frequency present?
Then each of those frequencies is shifted by an amount of time that causes
the frequency to be 180 degrees out of phase, so that, for the entire
group of frequencies, H(f) = e^{j*pi} * G(f), where H(f) is the output
and G(f) is the input. Since e^{j*pi} = -1, you get the required
polarity reversal.
Bob_Stanton
> "Michael R. Kesti" wrote:
> >
> > Randy Yates wrote:
> >
> > > That is true. But, who said that "180 degress out of phase" HAD to mean
> > > that there is only one frequency involved? You can shift EACH frequency 180 degrees out
> > > of phase if you allow the time delay through the circuit to vary
> > > as a function of frequency.
> >
> > What happens when there is more than one frequency present?
>
> Then each of those frequencies is shifted by an amount of time that causes
> the frequency to be 180 degrees out of phase, so that, for the entire
> group of frequencies, H(f) = e^{j*pi} * G(f), where H(f) is the output
> and G(f) is the input. Since e^{j*pi} = -1, you get the required
> polarity reversal.
Bob writes:
An autotransformer with the center tap grounded, inverts the output
signal 180 deg.
Ideal autotransformer.
__________
In >-----------()()()()()--------< out
|
Gnd
The autotransformer does this with either sinewaves, or complex waves,
or sinusoids. There is no time delay (with an ideal transformer) for a
signal passing through. Yet the signal still has 180 deg phase shift.
Bob Stanton
Michael R. Kesti
>"Michael R. Kesti" wrote:
>>
>> Randy Yates wrote:
>>
>> > That is true. But, who said that "180 degress out of phase" HAD to mean
>> > that there is only one frequency involved? You can shift EACH frequency 180 degrees out
>> > of phase if you allow the time delay through the circuit to vary
>> > as a function of frequency.
>>
>> What happens when there is more than one frequency present?
>
>Then each of those frequencies is shifted by an amount of time that causes
>the frequency to be 180 degrees out of phase, so that, for the entire
>group of frequencies, H(f) = e^{j*pi} * G(f), where H(f) is the output
>and G(f) is the input. Since e^{j*pi} = -1, you get the required
>polarity reversal.
So, in order to reverse the polarity of a complex audio signal, such as
music, you would propose using a time delay circuit for each possible
frequency that might occur? How many discrete delays would that be?
Michael R. Kesti
>An autotransformer with the center tap grounded, inverts the output
>signal 180 deg.
>
> Ideal autotransformer.
> __________
> In >-----------()()()()()--------< out
> |
> Gnd
>
>The autotransformer does this with either sinewaves, or complex waves,
>or sinusoids. There is no time delay (with an ideal transformer) for a
>signal passing through. Yet the signal still has 180 deg phase shift.
No, the signal is not shifted by an autotransformer. Its polarity is
inverted.
amoli...@visi-dot-com.com
>No, the signal is not shifted by an autotransformer. Its polarity is
>inverted.
You say tomayto, I say tomahto. Polarity inversion and
a 180-degree phase shift at all frequencies are arguably the same
thing. Of course, the 180-degree phase shift you and I have in mind
has zero group delay.
Richard D Pierce
Geez, I CANNOT believe how utterly obtuse this thread got.
It's VERY simple, as you both said. For ANY arbitrary signal
function producing a time variant voltage F(t), polarity
inversion is NOTHING more complicated than:
F'(t) = -1 * F(t)
That's it. Nothing more.
Take a signal being transmitted over a balanced line:
o-------------------o
o-------------------o
gnd-------------------gnd
And apply the following "physical" transform:
o------- ------o
\ /
X
/ \
o------ ------o
gnd-----------------gnd
That will give you 180 degrees of phase shift at EVERY frequency
with NO additional delay at ANY frequency.
Oy!
--
| Dick Pierce |
| Professional Audio Development |
| 1-781/826-4953 Voice and FAX |
| DPi...@world.std.com |
Brad Griffis
> >Then each of those frequencies is shifted by an amount of time that
causes
> >the frequency to be 180 degrees out of phase, so that, for the entire
> >group of frequencies, H(f) = e^{j*pi} * G(f), where H(f) is the output
> >and G(f) is the input. Since e^{j*pi} = -1, you get the required
> >polarity reversal.
>
> So, in order to reverse the polarity of a complex audio signal, such as
> music, you would propose using a time delay circuit for each possible
> frequency that might occur? How many discrete delays would that be?
>
You obviously have never taken a signal processing course. You would never
BUILD something to delay each and every frequency by separate amounts. The
differing delay of each frequency is a direct consequence of multiplying the
signal by -1. By multiplying the signal by -1 you are multiplying all the
sinusoids comprising the Fourier series of that signal by -1. Since -1
corresponds to a phase shift of 180 degrees and each sinusoid has a
different frequency, each frequency will essentially be delayed by a
different amount of time to give you a 180 degree shift in phase.
Michael R. Kesti
>> >Then each of those frequencies is shifted by an amount of time that causes
>> >the frequency to be 180 degrees out of phase, so that, for the entire
>> >group of frequencies, H(f) = e^{j*pi} * G(f), where H(f) is the output
>> >and G(f) is the input. Since e^{j*pi} = -1, you get the required
>> >polarity reversal.
>>
>> So, in order to reverse the polarity of a complex audio signal, such as
>> music, you would propose using a time delay circuit for each possible
>> frequency that might occur? How many discrete delays would that be?
>>
>You obviously have never taken a signal processing course.
Well, yes, I have, actually. Have you?
> You would never
>BUILD something to delay each and every frequency by separate amounts.
No shit? Then it seems that phase shifting and polarity inversion ARE
different things!
> The
>differing delay of each frequency is a direct consequence of multiplying the
>signal by -1.
No. The terms "delay" and "shift" specify changes in the time rather than
multiplication.
> By multiplying the signal by -1 you are multiplying all the
>sinusoids comprising the Fourier series of that signal by -1.
Correct.
> Since -1
>corresponds to a phase shift of 180 degrees
No. "-1" and "multiplying by -1" do not correspond to a phase shift of
180 degrees. Multiplying by -1 means that the output voltage at any
instant in time is of the polarity opposite of the input voltage at that
time. Phase shifting means that the output voltage an any given instant
is a function of the input voltage at some earlier time.
> and each sinusoid has a
>different frequency, each frequency will essentially be delayed by a
>different amount of time to give you a 180 degree shift in phase.
No. There is no delay inherent in multiplying by -1. That is why polarity
inversion and phase shifting are different things. They appear in some
cases to be the same thing, but that is only because of the repeating
nature of sinusoids. There is no aspect of polarity inversion that results
in a spectrum of delays that are magically applied to a corresponding
spectrum of frequencies.
"Polarity inversion" and "phase shifting" are therefore very different
things and it is in no way, despite appearances, correct to equate them.
Bob_Stanton
> ....
> arbitrary signal
> function producing a time variant voltage F(t), polarity
> inversion is NOTHING more complicated than:
>
> F'(t) = -1 * F(t)
>
> That's it. Nothing more.
>
> Take a signal being transmitted over a balanced line:
>
>
> B o-------------------o B'
>
> gnd-------------------gnd
>
> And apply the following "physical" transform:
>
>
> \ /
> X
> / \
> B o------ ------o A'
>
> gnd-----------------gnd
>
> That will give you 180 degrees of phase shift at EVERY frequency
> with NO additional delay at ANY frequency.
>
> Oy!
Bob writes:
Why are you showing a gnd line on your schematic? In a balance system,
gnd is irrelevent.
(I added labels "A" and "B", to your wires.)
What you did was twist the wires, by 180 deg. Physically twisting the
wires does not cause a 180 deg electrical phase shift.
Bob Stanton
Richard D Pierce
Bob_Stanton <rsta...@stny.rr.com> wrote:
>DPi...@TheWorld.com (Richard D Pierce) wrote in message
>> ....
>> arbitrary signal
>> function producing a time variant voltage F(t), polarity
>> inversion is NOTHING more complicated than:
>>
>> F'(t) = -1 * F(t)
>>
>> That's it. Nothing more.
>>
>> Take a signal being transmitted over a balanced line:
>>
>> A o-------------------o A'
>>
>> B o-------------------o B'
>>
>> gnd-------------------gnd
>>
>> And apply the following "physical" transform:
>>
>>
>> \ /
>> X
>> / \
>> B o------- ------o B
>>
>> gnd-----------------gnd
>>
>> That will give you 180 degrees of phase shift at EVERY frequency
>> with NO additional delay at ANY frequency.
>>
>> Oy!
>
>
>Bob writes:
>
>Why are you showing a gnd line on your schematic? In a balance system,
>gnd is irrelevent.
So? If it's irrelevant, it does not bear commenting.
>(I added labels "A" and "B", to your wires.)
And you did it wrong, thank you.
>What you did was twist the wires, by 180 deg.
No, I introduced essentially a DPDT switch and threw it,
connection A and B from the input to B' and A' to the output.
>Physically twisting the
>wires does not cause a 180 deg electrical phase shift.
Well, YOU twisted the wire when you incorrectly labeled the
outputs. So, it's not surprising it didn't work for you. :-)
Funny how since I posted the above, I got 7 emails replies, all
thanking me for making it very clear and understandable, and
only 1 post that was confused about it. Guess which one THAT
was! :-)
David Collins
rsta...@stny.rr.com (Bob_Stanton) wrote:
> What you did was twist the wires, by 180 deg. Physically twisting the
> wires does not cause a 180 deg electrical phase shift.
>
> Bob Stanton
He's swapping pins 2 and 3 to invert the phase!
DC
Bob_Stanton
> snip ..
> Well, YOU twisted the wire when you incorrectly labeled the
> outputs. So, it's not surprising it didn't work for you. :-)
>
> Funny how since I posted the above, I got 7 emails replies, all
> thanking me for making it very clear and understandable,
That *is* strange! :-0 ?
> and only 1 post that was confused about it. Guess which one THAT was! :-)
Was is George Middius? :-)
Bob Stanton
Brad Griffis
"Michael R. Kesti" <mke...@gv.net> wrote in message
news:3CB491D7...@gv.net...
> Brad Griffis wrote:
>
> > You would
never
> >BUILD something to delay each and every frequency by separate amounts.
>
> No shit? Then it seems that phase shifting and polarity inversion ARE
> different things!
>
Reading back to your earlier post now I see what you were trying to say all
along. Perhaps next time you could just plainly say what you're thinking
rather than being sarcastic. The humor doesn't transfer so well through
text.
>
> "Polarity inversion" and "phase shifting" are therefore very different
> things and it is in no way, despite appearances, correct to equate them.
>
Agreed - polarity inversion only affects the reference point of the signal
but does nothing to change the signal itself. Therefore all you see is the
sign change but no phase shift or anything else.
Amazing how many posts came out of such a simple question!
Michael R. Kesti
>Reading back to your earlier post now I see what you were trying to say all
>along. Perhaps next time you could just plainly say what you're thinking
>rather than being sarcastic. The humor doesn't transfer so well through
>text.
Having participated in Usenet for some 15 years, I am very much aware of
this. Thanks for the reminder, though. Perhaps you might do well to
avoid the ad hominem stuff, such as questioning others' educational
qualifications. This transfers no better than does sarcasm.
>> "Polarity inversion" and "phase shifting" are therefore very different
>> things and it is in no way, despite appearances, correct to equate them.
>>
>
>Agreed - polarity inversion only affects the reference point of the signal
>but does nothing to change the signal itself.
No. Changing the reference point is DC level shifting rather than
inversion. (Uh, oh. ;-)
> Therefore all you see is the
>sign change but no phase shift or anything else.
YES!
>Amazing how many posts came out of such a simple question!
Not so amazing. Many Usenet participants insist that their partial
understandings are complete and correct, and still more are willing to
point out otherwise!
Brian Nichols
> Brad Griffis wrote:
>
>
> >corresponds to a phase shift of 180 degrees
>
> No. "-1" and "multiplying by -1" do not correspond to a phase shift of
> 180 degrees. Multiplying by -1 means that the output voltage at any
> instant in time is of the polarity opposite of the input voltage at that
> time. Phase shifting means that the output voltage an any given instant
> is a function of the input voltage at some earlier time.
>
> > and each sinusoid has a
> >different frequency, each frequency will essentially be delayed by a
> >different amount of time to give you a 180 degree shift in phase.
>
> No. There is no delay inherent in multiplying by -1. That is why polarity
> inversion and phase shifting are different things. They appear in some
> cases to be the same thing, but that is only because of the repeating
> nature of sinusoids. There is no aspect of polarity inversion that results
> in a spectrum of delays that are magically applied to a corresponding
> spectrum of frequencies.
>
> "Polarity inversion" and "phase shifting" are therefore very different
> things and it is in no way, despite appearances, correct to equate them.
I just finished a DSP course so I'm far from an expert, but I'm going
to take a stab at this problem.
I learned that a phase shift and time delay are the same thing, just
expressed in different ways.
Say you have a signal that can be written a sum of sinusoids:
x(t) = Sum(A(i) * cos(2*pi*f(i)*t + phi(i))
where Sum is over i from 0 to N - 1, A(i) is the amplitude of the i'th
sinusoid, f(i) is the i'th sinusoid's frequency, and phi(i) is i'th
sinusoid's phase.
Say that you run x(t) through a system whose output, y(t), is:
y(t) = -x(t) = Sum(-A(i) * cos(2*pi*f(i)*t + phi(i))
So y(t) is x(t), inverted.
Now let's rewrite this equation. Each sinusoid is multiplied by -1
and, according to Euler, e^(j*pi) = -1. Also, cos(theta) =
1/2*(e^(j*theta) + e^(-j*theta). For readibility, let angle(i) =
2*pi*f(i)*t + phi(i).
So we get:
y(t) = Sum(e^(j*pi) * A(i) * 1/2 * [e^(j*angle(i)) +
e^(-j*angle(i))])
Multiply through and use the fact that e^(j*pi) = e^(-j*pi) to get:
y(t) = Sum(A(i) * 1/2 * [e^(j*pi) * e^(j*angle(i)) + e^(-j*pi) *
e^(-j*angle(i))])
y(t) = Sum(A(i) * 1/2 * [e^(j*(angle(i) + pi)) + e^(-j*(angle(i)
+ pi))])
Combine the complex conjugates to get sinusoids and we get:
y(t) = Sum(A(i) * cos(2*pi*f(i)*t + phi(i) + pi))
= -x(t) <--- By the original definition of y(t)
From this derivation, we see that multiplying a signal by -1 (i.e.,
inverting it) is equivalent to shifting the phase of each sinusoidal
component of the signal by 180 degrees.
Bob_Stanton
Bob writes:
Yes, a very interesting slight of hand trick. Basically what Dick did
was relabel the output wires. What had been wire output "A", he
relabeled "B". Then what had been output wire "B", he relabeled "A".
Example:
Before: A >-------------------< A'
B >-------------------< B'
After:
A >-------------------< B'
B >-------------------< A'
Final: Change the way the wires are drawn. People don't notice all
that has changed is the labels!
A >----------- -----------< A'
\ /
X
B ____________/ \____________ B'
(Relabeling alone does not cause an actual 180 deg phase shift.)
Bob Stanton
Michael R. Kesti
>"Michael R. Kesti" <mke...@gv.net> wrote in message news:<3CB491D7...@gv.net>...
>>
>> "Polarity inversion" and "phase shifting" are therefore very different
>> things and it is in no way, despite appearances, correct to equate them.
>
>I just finished a DSP course so I'm far from an expert, but I'm going
>to take a stab at this problem.
>
>I learned that a phase shift and time delay are the same thing, just
>expressed in different ways.
That is correct. My point is that both phase shift and its friend time
delay are different from polarity inversion.
>Say you have a signal that can be written a sum of sinusoids:
>
> x(t) = Sum(A(i) * cos(2*pi*f(i)*t + phi(i))
>
>where Sum is over i from 0 to N - 1, A(i) is the amplitude of the i'th
>sinusoid, f(i) is the i'th sinusoid's frequency, and phi(i) is i'th
>sinusoid's phase.
>
>Say that you run x(t) through a system whose output, y(t), is:
>
> y(t) = -x(t) = Sum(-A(i) * cos(2*pi*f(i)*t + phi(i))
>
>So y(t) is x(t), inverted.
>
>Now let's rewrite this equation. Each sinusoid is multiplied by -1
>and, according to Euler, e^(j*pi) = -1. Also, cos(theta) =
>1/2*(e^(j*theta) + e^(-j*theta). For readibility, let angle(i) =
>2*pi*f(i)*t + phi(i).
>
>So we get:
>
> y(t) = Sum(e^(j*pi) * A(i) * 1/2 * [e^(j*angle(i)) + >e^(-j*angle(i))])
>
>Multiply through and use the fact that e^(j*pi) = e^(-j*pi) to get:
>
> y(t) = Sum(A(i) * 1/2 * [e^(j*pi) * e^(j*angle(i)) + e^(-j*pi) * e^(-j*angle(i))])
>
> y(t) = Sum(A(i) * 1/2 * [e^(j*(angle(i) + pi)) + e^(-j*(angle(i) + pi))])
>
>Combine the complex conjugates to get sinusoids and we get:
>
> y(t) = Sum(A(i) * cos(2*pi*f(i)*t + phi(i) + pi))
> = -x(t) <--- By the original definition of y(t)
>
>From this derivation, we see that multiplying a signal by -1 (i.e.,
>inverting it) is equivalent to shifting the phase of each sinusoidal
>component of the signal by 180 degrees.
Well done, and I admit that this derivation shows some validity for
considering phase shifting and inversion to be equivalent. The trouble,
though, is that "each sinusoidal component" part. In the real world,
analog and digital, it is impractical to implement this function by
extracting out each possible frequency and shifting them each by the
required time. Instead, we simply invert (Or is it "reverse"?) the
polarity using an elegantly simple circuits/algorithms.
Larry Brasfield
Brian Nichols (grandfl...@yahoo.com) says...
...
> I just finished a DSP course so I'm far from an expert, but I'm going
> to take a stab at this problem.
>
> I learned that a phase shift and time delay are the same thing, just
> expressed in different ways.
Good, so far.
> Say you have a signal that can be written a sum of sinusoids:
>
> x(t) = Sum(A(i) * cos(2*pi*f(i)*t + phi(i))
>
> where Sum is over i from 0 to N - 1, A(i) is the amplitude of the i'th
> sinusoid, f(i) is the i'th sinusoid's frequency, and phi(i) is i'th
> sinusoid's phase.
I've never listened to any music that could
be described that way. Of course, that's
because I'm not old enough (and neither is
the Universe). The closest anything I've
heard has come to being representable that
way has been some 1.8181818181818... Hertz
(and harmonics thereof) material that was
annoying enough to make me give my turntable
a little nudge.
[Reasonable looking analysis of Fourier series
equivalences cut for space, without dispute.]
> From this derivation, we see that multiplying a signal by -1 (i.e.,
> inverting it) is equivalent to shifting the phase of each sinusoidal
> component of the signal by 180 degrees.
Here, you err. Your analysis applies only to
signals that repeat at f Hz across all time,
and not to "a signal" not so qualified.
If you're interested in an analysis that more
closely meets the issues in this little fray,
I suggest starting with a Fourier transform.
In that frequency domain, the equivalent of
a 180 degree phase shift is multiplication by
the function exp(j * w * pi) which, as anyone
with eyes can see, does not equal -1.
I hope that's the end of my part in this
silly thread.
--
-Larry Brasfield
(address munged)
Brian Nichols
> Brian Nichols wrote:
>
> >"Michael R. Kesti" <mke...@gv.net> wrote in message news:<3CB491D7...@gv.net>...
> >
> >I learned that a phase shift and time delay are the same thing, just
> >expressed in different ways.
>
> That is correct. My point is that both phase shift and its friend time
> delay are different from polarity inversion.
>
Actually, I have no idea why I even bothered saying that. Something
else must have been on my mind.
Actually, the identity above shows that a 180 degree phase shift and a
polarity reversal are *exactly* equivalent. So if system A inverts
the signal and system B shifts the phase of all frequency components
by 180 degrees, A and B will give exactly the same output given the
same input.
> though, is that "each sinusoidal component" part. In the real world,
> analog and digital, it is impractical to implement this function by
> extracting out each possible frequency and shifting them each by the
> required time. Instead, we simply invert (Or is it "reverse"?) the
> polarity using an elegantly simple circuits/algorithms.
I definitely agree with you on that point. For one thing, the
equation for inverting is much easier to write. :-)
Brian Nichols
comments below and let me know where I'm off.
Larry Brasfield <larry_b...@snotmail.com> wrote in message news:<MPG.171ff1d8a...@news.qwest.net>...
> In article <e02065c3.02041...@posting.google.com>,
> Brian Nichols (grandfl...@yahoo.com) says...
> ...
>
> > Say you have a signal that can be written a sum of sinusoids:
> >
> > x(t) = Sum(A(i) * cos(2*pi*f(i)*t + phi(i))
> >
> > where Sum is over i from 0 to N - 1, A(i) is the amplitude of the i'th
> > sinusoid, f(i) is the i'th sinusoid's frequency, and phi(i) is i'th
> > sinusoid's phase.
>
> I've never listened to any music that could
> be described that way. Of course, that's
> because I'm not old enough (and neither is
> the Universe). The closest anything I've
> heard has come to being representable that
> way has been some 1.8181818181818... Hertz
> (and harmonics thereof) material that was
> annoying enough to make me give my turntable
> a little nudge.
>
I find that strange. I just finished an MP3 codec project and I
frequently would load digital audio clips in Matlab and take the DFT,
effectively giving me the sum of sinusoids that would give me the same
discrete time samples. I assume that I could accomplish the same
thing by taking a continuous Fourier transform of a continuous time
signal.
Note that in the definition of x(t), I never specified how A(i), f(i),
and phi(i) are obtained. They could be from a Fourier series, a
Fourier Transform, or just random, non-harmonically related sinusoids
pulled at random from a hat. I also didn't specify the domain of t,
so t may only be defined over the interval[0,T], which could
correspond to a audio clip T seconds long. And, of course, because
x(t) could be any signal, the sum of sinusoids could be infinite.
Perhaps I should have been more specific.
In any event, it's my understanding that:
1) Any periodic function can be written as a sum of sinusoids
from the Fourier series.
2) (Almost) any non-periodic function can be written as a sum of
sinusoids from its Fourier transform.
Please let me know if you think I'm wrong on any of these points
because my basic understanding of DSP is built upon this foundation.
> [Reasonable looking analysis of Fourier series
> equivalences cut for space, without dispute.]
> > From this derivation, we see that multiplying a signal by -1 (i.e.,
> > inverting it) is equivalent to shifting the phase of each sinusoidal
> > component of the signal by 180 degrees.
>
> Here, you err. Your analysis applies only to
> signals that repeat at f Hz across all time,
> and not to "a signal" not so qualified.
>
Once again, the domain of t could be finite.
Your statement really bothers me because a lot of what I have learned
about LTI systems in DSP (e.g. frequency response, z-transforms)
relies upon the fact that signals can be written as a sum of complex
sinusoids.
> If you're interested in an analysis that more
> closely meets the issues in this little fray,
> I suggest starting with a Fourier transform.
> In that frequency domain, the equivalent of
> a 180 degree phase shift is multiplication by
> the function exp(j * w * pi) which, as anyone
> with eyes can see, does not equal -1.
>
Can you please explain this point further.
Looking at e^*(j*w*pi), I see a linear phase response that results in
output signals being advanced by pi seconds or samples. I don't see
how that results in a 180 degree phase shift.
I think that multiplying the Fourier Transform coefficients by
e^(j*pi) would result in a 180 degree phase shift and (the equivalent)
signal inversion.
As a test, I ran the following code in Matlab:
y = loadwave(<some-audio-clip>);
Y = fft(y);
Y = Y .* (cos(pi) + j * sin(pi)); <-- equivalent to e^(j*pi) =
-1
v = ifft(Y);
s = y + v;
The result, s, has all values set almost to zero (I assume they aren't
exactly zero because of rounding errors). In any event, it shows that
v is the polarity inverted version of y, which makes sense because all
FT coefficients have their magnitude reversed.
> I hope that's the end of my part in this
> silly thread.
I hope, before you finally make your exit, you can share a little more
info with me. :-)
Larry Brasfield
> comments below and let me know where I'm off.
You're not far off. I made a mistake in that post,
noticed it a few minutes after I posted it, and
cancelled the post. I considered also posting a
correction, but since it appeared the cancel had
taken effect, I figure the correction would waste
more people's time than any stray, non-canceled
copies of my first post would waste.
I apologize now for being mistaken in that
judgment as well. (I would love to see some
good statistics on cancel effectiveness.)
However, since you've taken the time and
taken me seriously, I will respond below.
> Larry Brasfield <larry_b...@snotmail.com> wrote in message news:<MPG.171ff1d8a...@news.qwest.net>...
> > In article <e02065c3.02041...@posting.google.com>,
> > Brian Nichols (grandfl...@yahoo.com) says...
> > ...
> >
> > > Say you have a signal that can be written a sum of sinusoids:
> > >
> > > x(t) = Sum(A(i) * cos(2*pi*f(i)*t + phi(i))
> > >
> > > where Sum is over i from 0 to N - 1, A(i) is the amplitude of the i'th
> > > sinusoid, f(i) is the i'th sinusoid's frequency, and phi(i) is i'th
> > > sinusoid's phase.
> >
> > I've never listened to any music that could
> > be described that way. Of course, that's
> > because I'm not old enough (and neither is
> > the Universe). The closest anything I've
> > heard has come to being representable that
> > way has been some 1.8181818181818... Hertz
> > (and harmonics thereof) material that was
> > annoying enough to make me give my turntable
> > a little nudge.
> >
>
> I find that strange. I just finished an MP3 codec project and I
> frequently would load digital audio clips in Matlab and take the DFT,
> effectively giving me the sum of sinusoids that would give me the same
> discrete time samples. I assume that I could accomplish the same
> thing by taking a continuous Fourier transform of a continuous time
> signal.
The DFT is a discrete form of the continous
Fourier transform. My only point above was
that your original analysis was based on the
Fourier series, which only applies to signals
that repeat, endlessly. So, I mentioned that
not many signals do that, and that even a
stuck record only comes close.
> Note that in the definition of x(t), I never specified how A(i), f(i),
> and phi(i) are obtained. They could be from a Fourier series, a
> Fourier Transform, or just random, non-harmonically related sinusoids
> pulled at random from a hat. I also didn't specify the domain of t,
> so t may only be defined over the interval[0,T], which could
> correspond to a audio clip T seconds long. And, of course, because
> x(t) could be any signal, the sum of sinusoids could be infinite.
> Perhaps I should have been more specific.
Well, as you stated the expression, it was
clearly the Fourier series. The cosine is
an endlessly repeating signal, so what you
wrote is not true for time-limited signals.
To be talking about the Fourier transform,
you would have to mention an integral
rather than a sum.
> In any event, it's my understanding that:
>
> 1) Any periodic function can be written as a sum of sinusoids
> from the Fourier series.
True.
> 2) (Almost) any non-periodic function can be written as a sum of
> sinusoids from its Fourier transform.
Not quite. The problem is that sinusoids
extend over all time and if your signal
does not, then that sum cannot represent
that signal (for non-zero coefficients).
> Please let me know if you think I'm wrong on any of these points
> because my basic understanding of DSP is built upon this foundation.
I am sure you'll adjust your thinking
a little bit.
> > [Reasonable looking analysis of Fourier series
> > equivalences cut for space, without dispute.]
> > > From this derivation, we see that multiplying a signal by -1 (i.e.,
> > > inverting it) is equivalent to shifting the phase of each sinusoidal
> > > component of the signal by 180 degrees.
> >
> > Here, you err. Your analysis applies only to
> > signals that repeat at f Hz across all time,
> > and not to "a signal" not so qualified.
> >
>
> Once again, the domain of t could be finite.
I don't understand what that means.
One would have to describe a weighting
function (of t) or something because t
does not have abrubt limits, at least
not in mathematical analysis related
to Fourier transforms. (The Laplace
transform deals with positive t, but
this requires some explicit work.)
> Your statement really bothers me because a lot of what I have learned
> about LTI systems in DSP (e.g. frequency response, z-transforms)
> relies upon the fact that signals can be written as a sum of complex
> sinusoids.
That understanding may be good enough for
doing some DSP work, but if you have to
do the math, it will let you down.
> > If you're interested in an analysis that more
> > closely meets the issues in this little fray,
> > I suggest starting with a Fourier transform.
> > In that frequency domain, the equivalent of
> > a 180 degree phase shift is multiplication by
> > the function exp(j * w * pi) which, as anyone
> > with eyes can see, does not equal -1.
> >
>
> Can you please explain this point further.
Yes. I should have written "exp(j * pi)"
rather than "exp(j * w * pi)". I flubbed
the cancellation of w when I wrote out
an expression for the transform of delay
versus frequency. My only, sad excuse
is that it was past my bedtime.
> Looking at e^*(j*w*pi), I see a linear phase response that results in
> output signals being advanced by pi seconds or samples. I don't see
> how that results in a 180 degree phase shift.
Your perception is acute and accurate.
> I think that multiplying the Fourier Transform coefficients by
> e^(j*pi) would result in a 180 degree phase shift and (the equivalent)
> signal inversion.
I agree.
> As a test, I ran the following code in Matlab:
[Cut without dispute because it looks
sensible and gives the right answer.]
> > I hope that's the end of my part in this
> > silly thread.
>
> I hope, before you finally make your exit, you can share a little more
> info with me. :-)
I hope this does it. And I repeat my
apology for seeming to confound the
issue when you had it nearly right.
Brian Nichols
Thanks for the information. It was most informative. Just one last
question (and sorry to the newsgroup denizens, as I'm going way off
topic)
>
> > In any event, it's my understanding that:
> >
> > 1) Any periodic function can be written as a sum of sinusoids
> > from the Fourier series.
>
> True.
>
> > 2) (Almost) any non-periodic function can be written as a sum of
> > sinusoids from its Fourier transform.
>
> Not quite. The problem is that sinusoids
> extend over all time and if your signal
> does not, then that sum cannot represent
> that signal (for non-zero coefficients).
>
So how does one interpret the results of a DFT?
Here's my current understanding... Let's say you have a real,
discrete-time signal, x[n], with 20 samples. In that case, a DFT will
give 20 coefficients, X[n]. So from the results, you could determine
10 sinusoids and a DC component that, when summed together, gives you
the original signal. So I could write:
x[n] = Sum( abs(X[i]) * cos(2*pi*(i/20)*n + angle(X[i])) )
for i = 0 to 10 (I think this sum will off by, at most, a constant
multiplier) and get the same samples as the input signal.
Because I only have 20 inputs samples(x[0],...,x[19]), I also assume
that the domain of x[n] is n = 0 to 19, inclusive, and is something
undefined outside of that interval, both in the case of the original
samples and in the expansion above.
So if this interpretation is incorrect, how should I interpret a DFT?
Larry Brasfield
[quoting Brian]
> > > 2) (Almost) any non-periodic function can be written as a sum of
> > > sinusoids from its Fourier transform.
> >
> > Not quite. The problem is that sinusoids
> > extend over all time and if your signal
> > does not, then that sum cannot represent
> > that signal (for non-zero coefficients).
> >
>
> So how does one interpret the results of a DFT?
>
> Here's my current understanding... Let's say you have a real,
> discrete-time signal, x[n], with 20 samples. In that case, a DFT will
> give 20 coefficients, X[n]. So from the results, you could determine
> 10 sinusoids and a DC component that, when summed together, gives you
> the original signal. So I could write:
>
> x[n] = Sum( abs(X[i]) * cos(2*pi*(i/20)*n + angle(X[i])) )
>
> for i = 0 to 10 (I think this sum will off by, at most, a constant
> multiplier) and get the same samples as the input signal.
>
> Because I only have 20 inputs samples(x[0],...,x[19]), I also assume
> that the domain of x[n] is n = 0 to 19, inclusive, and is something
> undefined outside of that interval, both in the case of the original
> samples and in the expansion above.
>
> So if this interpretation is incorrect, how should I interpret a DFT?
I don't think you need to change your
conceptualization of the DFT.
My points about applicability related to
your initial post where you wrote
Say you have a signal that can be written a sum of sinusoids:
x(t) = Sum(A(i) * cos(2*pi*f(i)*t + phi(i))
where there is no bound put on t.
I believe our chief difference here is
that when you write "sinusoid" you mean
a sine function truncated at both ends
at some (unspecified) time. But what I
quoted above does not have such bounds.
If you look at the definition of the
discrete-time Fourier transform, one
form of which is
X(exp(j w))
= Sum[over n = -infinity to +infinity](x[n] exp(-j w n)))
and its inverse,
x[n]
= Integral[from -pi to +pi](X(exp(j w) exp(j w n T) d(w)] / (2 pi)
you see that the signal reconstruction
(the inverse) is effected with bounded
sinusoids. Notice also that the domain
of n is unbounded for the DFT. (This
goes to show that a bounded signal is by
no means to be assumed when one mentions
DFT's or FT's in a general way.)
Of course, in real applications of the DFT,
n is bounded and nobody uses sinusoids of
any kind to do inverse DFT's, (preferring
the FFT in everything I have seen).
I just think you have transferred your
more-or-less accurate knowledge of the
theoretical basis for the DFT to showing
a hypothetical signal reconstruction and
left out a little detail that nit-picky
folks like me love to pick nits about.
You mentioned earlier that you had not
specified the domain of t and that the
domain of t could be finite. And if you
had specified a finite domain for t, I
would have had no nits to pick.
I almost feel like apologizing for having
picked that nit, but that's the Usenet way.
So I hope it has not been too obnoxious.
Brad Griffis
"Brian Nichols" <grandfl...@yahoo.com> wrote in message
> So how does one interpret the results of a DFT?
Interpretation of the DFT:
There are two main ways of conceptualizing the DFT. The first method is how
I learned it in my undergrad DSP courses and the second way was taught in my
graduate DSP course.
1. The DFT can be thought of as a sampled version of the DTFT (discrete
time Fourier transform). Let's say you had a single rectangular pulse in
time. A rectangle in time transforms to a sinc in frequency. That is, if
you took the DTFT of that rectangle in time you would get a continuous sinc
in frequency. If you took the DFT of that rectangle you would get samples
of that sinc. The number of DFT samples is equivalent to the number of time
samples. Also, for some reason many people don't seem to realize that the
FFT and the DFT are the same thing. The FFT is just the fast algorithm
created by Tukey and Cooley way back before my day.
2. The DFT is really the discrete Fourier series. Let's first look back to
continuous time signals to understand this. For a continuous signal that is
periodic, you can represent it with a Fourier series. If it has period T
and fundamental frequency w0 = 2pi / T then the Fourier series will contain
only frequencies at w0 and its harmonics. Now let's say you have a periodic
signal in discrete time. You can write a Fourier series for this signal
which, like its continuous time counterpart, will also only have components
at multiples of the fundamental. The fundamental maps to n=1 and then you
get frequencies at all the multiples of this (i.e. 2, 3, ... N-1).
When deciding which interpretation to use you need to consider the nature of
the signal. If you had a perfectly periodic signal then you would think of
the FFT of a period of that signal to show exactly the frequency content of
that signal. If you had a signal that was perfectly aperiodic such as a
single rectangle, then you would think of the FFT of this rectangle as being
a sampled version of the true frequency content. For most signals though
you will be somewhere in between. A signal that comes close to being
periodic is usually referred to as being quasi periodic. A television
signal, for example, is quasi periodic. The reason for this is that from
one scan line to the next, there is usually VERY little difference. This
makes the signal look almost as if it is periodic. This causes its spectrum
to have single spikes in it at the rate of the scan lines. Many audio
signals can also be thought of as quasi periodic. For example, if you look
at speech in the time domain it looks sort of like an EKG. It repeats
basically the same thing.
Sorry for the HUGE post, but I hope that some of you may benefit.
Brad
Randy Yates
> [...]
> The trouble,
> though, is that "each sinusoidal component" part. In the real world,
> analog and digital, it is impractical to implement this function by
> extracting out each possible frequency and shifting them each by the
> required time.
Sorry, Michael, but it is not. That is, it *is* practical to build such
a function.
This is exactly what a practical Hilbert transformer does, only in this
case you cannot simplify the process into one involving a multiply by
any real constant since the phase shift at each frequency is 90 degrees.
Granted the practical Hilbert transformer is not *perfect*, but many
implementations do a good job.
Michael R. Kesti
>"Michael R. Kesti" wrote:
>> [...]
>> The trouble,
>> though, is that "each sinusoidal component" part. In the real world,
>> analog and digital, it is impractical to implement this function by
>> extracting out each possible frequency and shifting them each by the
>> required time.
>
>Sorry, Michael, but it is not. That is, it *is* practical to build such
>a function.
>
>This is exactly what a practical Hilbert transformer does, only in this
>case you cannot simplify the process into one involving a multiply by
>any real constant since the phase shift at each frequency is 90 degrees.
>Granted the practical Hilbert transformer is not *perfect*, but many
>implementations do a good job.
Can we agree that a reasonable test of practicality of a technique is
whether that technique is used in products one can purchase? Can you
name any audio products that use Hilbert transformers to provide a 180
degree phase shift of all in-band frequencies?
Randy Yates
>
> Randy Yates wrote:
>
> >"Michael R. Kesti" wrote:
> >> [...]
> >> The trouble,
> >> though, is that "each sinusoidal component" part. In the real world,
> >> analog and digital, it is impractical to implement this function by
> >> extracting out each possible frequency and shifting them each by the
> >> required time.
> >
> >Sorry, Michael, but it is not. That is, it *is* practical to build such
> >a function.
> >
> >This is exactly what a practical Hilbert transformer does, only in this
> >case you cannot simplify the process into one involving a multiply by
> >any real constant since the phase shift at each frequency is 90 degrees.
> >Granted the practical Hilbert transformer is not *perfect*, but many
> >implementations do a good job.
>
> Can we agree that a reasonable test of practicality of a technique is
> whether that technique is used in products one can purchase?
Yes, that seems very reasonable.
> Can you
> name any audio products that use Hilbert transformers to provide a 180
> degree phase shift of all in-band frequencies?
As I said above, Hilbert transformers shift 90 degrees, not 180 degrees. Therefore
the application of an HT is not to polarity inversion. Typical applications include
the generation of the analytic signal, quadrature downconversion, computation of
minimum phase, etc. I know this doesn't make much sense, but it would take us
fairly far afield of the current topic to explain it all.
Are there any audio products you can purchase that use practical Hilbert transformers?
Yes. I'm not sure if it's the DRA Labs MLSSA or the Audio Precision System One, but
John Atkinson used it to measure the excess phase of the Thiele CS6 speakers in
the following article:
http://www.stereophile.com/fullarchives.cgi?218
There are probably dozens of other designs that utilize a Hilbert transformer, but
such an algorithm, especially when implemented in a processor or DSP, is not
usually addressed in the overall functioning or features of the product.
Michael R. Kesti
<snip>
>> Can you
>> name any audio products that use Hilbert transformers to provide a 180
>> degree phase shift of all in-band frequencies?
>
>As I said above, Hilbert transformers shift 90 degrees, not 180 degrees.
I was thinking that you might be suggesting running two, in series.
> Therefore
>the application of an HT is not to polarity inversion. Typical applications include
>the generation of the analytic signal, quadrature downconversion, computation of
>minimum phase, etc. I know this doesn't make much sense, but it would take us
>fairly far afield of the current topic to explain it all.
It seems, then, that the HT does not qualify as a practical way to
implement polarity inversion.
<more deletia>
Randy Yates
> [...]
> implement polarity inversion.
I think you misunderstand the situation. As Brian Nichols so beautifully
pointed out, multiplying by -1 is indeed the same thing as shifting
the phase by 180 degrees at each frequency. In fact, what you are doing
when you multiply by -1 is "filtering" (i.e., convolving) with a perfect
180 degree phase shifter. I put "filtering" in quotes because this is a
one-tap FIR filter, which means convolution simplifies into a single
multiplication. Not much of a filter, but in this case it performs the
desired function perfectly.
Now two posts ago, you wrote this:
In the real world,
analog and digital, it is impractical to implement this function by
extracting out each possible frequency and shifting them each by the
required time.
That is false because the practical one-tap filter does exactly this. Because
the amount of phase shift is the special value 180 degrees, we get this
especially simple form of the filter.
What I was attempting to further show is that your statement is not
true even if we don't use the value of 180 degrees. Enter the Hilbert
transformer. Here we have a filter (in this case, a non-trivial one)
that practically shifts the phase at each possible frequency by
90 degrees.
I'm sorry if I wasn't clear - I did skip from your discussion of
polarity inversion (a 180 degree phase shifter) to a 90 degree
phase shifter without being perfectly clear about the point.
Randy Yates
> [...]
Are you absolutely sure about this? Would you like to bet some
cash?
Try this gedanken. Put a 1000 Hz sine wave into a "polarity
inverter." Does the output match the input? No, of course not.
It appears to be delayed by exactly 500 microseconds. No, it
IS delayed by 500 microseconds.
OK, put a 2000 Hz sine wave into a "polarity
inverter." Does the output match the input? No, of course not.
It appears to be delayed by exactly 250 microseconds. No, it
IS delayed by 250 microseconds.
Etc.
> That is why polarity
> inversion and phase shifting are different things. They appear in some
> cases to be the same thing, but that is only because of the repeating
> nature of sinusoids. There is no aspect of polarity inversion that results
> in a spectrum of delays that are magically applied to a corresponding
> spectrum of frequencies.
Yes, there is, and Brian Nichols proved it mathematically.
> "Polarity inversion" and "phase shifting" are therefore very different
> things and it is in no way, despite appearances, correct to equate them.
You have taken every post in this thread that has shown, indeed proven,
otherwise and contradicted it with absolutely no hard mathematics or logical
reasoning. In effect, you are sticking your head in the sand.
John Woodgate
<3CBA5AD3...@ieee.org>) about 'Shifting audio signal 180 degrees
out of phase', on Mon, 15 Apr 2002:
>Put a 1000 Hz sine wave into a "polarity
>inverter." Does the output match the input? No, of course not.
>It appears to be delayed by exactly 500 microseconds. No, it
>IS delayed by 500 microseconds.
Where is the delay in a unity-gain inverting op-amp circuit? If I put 1
Hz in, will I have to wait half a second for the output to appear?
--
Regards, John Woodgate, OOO - Own Opinions Only. http://www.jmwa.demon.co.uk
Interested in professional sound reinforcement and distribution? Then go to
http://www.isce.org.uk
PLEASE do NOT copy news posts to me by E-MAIL!
Michael R. Kesti
>"Michael R. Kesti" wrote:
>> [...]
>> No. There is no delay inherent in multiplying by -1.
>
>Are you absolutely sure about this? Would you like to bet some
>cash?
Oh, pulleeze. See the last line of this article.
>Try this gedanken. Put a 1000 Hz sine wave into a "polarity
>inverter." Does the output match the input? No, of course not.
>It appears to be delayed by exactly 500 microseconds. No, it
>IS delayed by 500 microseconds.
>
>OK, put a 2000 Hz sine wave into a "polarity
>inverter." Does the output match the input? No, of course not.
>It appears to be delayed by exactly 250 microseconds. No, it
>IS delayed by 250 microseconds.
You were correct the first time. It only appears to be delayed because of
the repeating nature of sinus waves.
>Etc.
What happens when the sinus wave is no longer repeating? Connect a sine
wave generator, switch, invertor, and an oscilloscope thus:
+---------+
/ |\ | |
+---------o o-----| >o-----| o'scope |
| |/ | |
+------+ +---------+
| sine | |
| gen | ---
+------+ -
|
---
-
Set the generator for 1 KHz and close the switch. At what time, relative
to closing the switch, will the o'scope begin to receive voltage from the
invertor? If you open the switch, when will the o'scope no longer receive
the inverted sinusoid? Will anything in these regards change if the
generator is set to 2 KHz?
>You have taken every post in this thread that has shown, indeed proven,
>otherwise and contradicted it with absolutely no hard mathematics or logical
>reasoning. In effect, you are sticking your head in the sand.
Any more of this crap and you will have the honor of being the latest entry
in my killfile.
Randy Yates
> What happens when the sinus wave is no longer repeating?
Then it is no longer a sinus wave. A sinus wave is not
"a sinus wave that is no longer repeating," it is a sinus wave.
> Connect a sine
> wave generator, switch, invertor, and an oscilloscope thus:
>
> +---------+
> / |\ | |
> +---------o o-----| >o-----| o'scope |
> | |/ | |
> +------+ +---------+
> | sine | |
> | gen | ---
> +------+ -
> |
> ---
> -
>
> Set the generator for 1 KHz and close the switch. At what time, relative
> to closing the switch, will the o'scope begin to receive voltage from the
> invertor?
What has this got to do with disproving that a 1 kHz sine wave is not
delayed by 500 microseconds by an inverter? The input is not a sine
wave, but some beast composed of many different sine waves.
> Any more of this crap and you will have the honor of being the latest entry
> in my killfile.
A killfile darkens your world much like a sandy piece of ground can.
amoli...@visi-dot-com.com
John Woodgate <j...@jmwa.demon.co.uk> wrote:
>I read in rec.audio.tech that Randy Yates <ya...@ieee.org> wrote (in
><3CBA5AD3...@ieee.org>) about 'Shifting audio signal 180 degrees
>out of phase', on Mon, 15 Apr 2002:
>>Put a 1000 Hz sine wave into a "polarity
>>inverter." Does the output match the input? No, of course not.
>>It appears to be delayed by exactly 500 microseconds. No, it
>>IS delayed by 500 microseconds.
>
>Where is the delay in a unity-gain inverting op-amp circuit? If I put 1
>Hz in, will I have to wait half a second for the output to appear?
No, that would be GROUP delay, which for that circuit is
both very small and nearly constant (I assume -- what I know of
op-amps can be written on a very small object in a very large
font.)
Randy's talking about the other kind of delay.
Michael R. Kesti
>"Michael R. Kesti" wrote:
>> What happens when the sinus wave is no longer repeating?
>
>Then it is no longer a sinus wave. A sinus wave is not
>"a sinus wave that is no longer repeating," it is a sinus wave.
>
>> Connect a sine
>> wave generator, switch, invertor, and an oscilloscope thus:
>>
>> +---------+
>> / |\ | |
>> +---------o o-----| >o-----| o'scope |
>> | |/ | |
>> +------+ +---------+
>> | sine | |
>> | gen | ---
>> +------+ -
>> |
>> ---
>> -
>>
>> Set the generator for 1 KHz and close the switch. At what time, relative
>> to closing the switch, will the o'scope begin to receive voltage from the
>> invertor?
>
>What has this got to do with disproving that a 1 kHz sine wave is not
>delayed by 500 microseconds by an inverter? The input is not a sine
>wave, but some beast composed of many different sine waves.
Fine. But the real world voice and music signals we deal with in audio
practice are not sines, either. You can't have it both ways, Randy!
>> Any more of this crap and you will have the honor of being the latest entry
>> in my killfile.
>
>A killfile darkens your world much like a sandy piece of ground can.
And ad hominem attacks do nothing to support your side of this discussion.
I will not respond if you again resort to them.
Randy Yates
Hello Michael,
Forgive my ad-hominem attacks. You are absolutely right - they do nothing
to further understanding.
I think you have a healthy degree of skepticism. I also think it may serve
us well if we back up a bit and let me ask you this: Do you believe that
any signal (well, most signals we see in the real world anyway) can be
represented as a combination of sine waves? I'm not just talking about
periodic signals either, but both periodic and aperiodic signals. As
an example close to this discussion, consider a 1 kHz sine wave that
has been "gated" by a pulse of amplitude 1 that comes on at t=0 and
goes off at t=1 [second]. Do you believe this signal can be described
by a sum of sine waves? Joe says it can (you know Joe, right? Joseph Fourier?).
But wait - how can a signal which is zero for most of the time be represented
by a bunch of signals that are zero nowhere (except at isolated points)?
Randy Yates
Right, Andrew, although as you know from the long thread on group
delay in comp.dsp awhile back, I would not in general say group
delay is the time for a signal to go through a circuit. But
it is in this case. In fact, for this circuit, the group
delay is 0 since GD = -d phi/dt and phi = pi/2, a constant.
--
Randy Yates
DSP Engineer, Sony Ericsson Mobile Communications
Research Triangle Park, NC, USA
randy...@ericsson.com, 919-472-1124
Michael R. Kesti
<snip>
>Forgive my ad-hominem attacks. You are absolutely right - they do nothing
>to further understanding.
Thank-you.
>I think you have a healthy degree of skepticism. I also think it may serve
>us well if we back up a bit and let me ask you this: Do you believe that
>any signal (well, most signals we see in the real world anyway) can be
>represented as a combination of sine waves? I'm not just talking about
>periodic signals either, but both periodic and aperiodic signals. As
>an example close to this discussion, consider a 1 kHz sine wave that
>has been "gated" by a pulse of amplitude 1 that comes on at t=0 and
>goes off at t=1 [second]. Do you believe this signal can be described
>by a sum of sine waves? Joe says it can (you know Joe, right? Joseph Fourier?).
>But wait - how can a signal which is zero for most of the time be represented
>by a bunch of signals that are zero nowhere (except at isolated points)?
What's this, now? A trick question? What I believe is that Fourier's
fine work has little or nothing to do with whether a circuit or algorithm
that multiplies its input by -1 can be properly considered to apply a
spectrum of delays in order to invert the polarity of a spectrum of
frequencies and whether it is practical to attempt to implement a
circuit or algorithm that does so.
Bob_Stanton
Bob writes
John Woodgate was correct, there is *no* delay. There is no group
delay, and there is no time delay.
If someone puts an impulse into a unity-gain inverting op-amp (or a
phase inverting ideal transformer), an inverse pulse will
instantaneously appear on the output.
Randy believes phase shift = time delay, as fervently as ancient
people believed the earth was flat.
Bob Stanton
Brother
Bob_Stanton wrote:
> John Woodgate was correct, there is *no* delay. There is no group
> delay, and there is no time delay.
>
> If someone puts an impulse into a unity-gain inverting op-amp (or a
> phase inverting ideal transformer), an inverse pulse will
> instantaneously appear on the output.
>
> Randy believes phase shift = time delay, as fervently as ancient
> people believed the earth was flat.
But unlike the flat-earthers, he is correct. And what
evidence is there that ancient people believed the
earth to be flat? Some fictional event created by
Washington Irving, maybe? Eratosthenes came up with
a fairly good estimate of the diameter of the Earth
well before the Christian era -- so at least some
ancients knew the world to be round.
Phase shift DOES involve time delay. Polarity inversion
does not. An inverting transformer inverts POLARITY,
not PHASE. A 180-degree phase shift is not the same
as a polarity inversion. Swapping the + and - wires
at a speaker's terminals, for instance, provides no
phase shift at all. It creates a polarity inversion.
A lot of the confusion and controversy on this thread
results from terminology applied sloppily or just
plain incorrectly. Although a polarity inversion is
often loosely known as a phase inversion or a 180-degree
phase shift, it is not. A polarity inversion has nothing
to to with frequency, time, or angular measurement.
No phase, no shift, no delay, no degrees, no radians.
A phase shift involves time. Use the terms correctly,
and much of the hoo-ha vanishes.
Phil Allison
"Brother" <nos...@last.com> wrote in message
news:3CBB9080...@last.com...
>
> Phase shift DOES involve time delay. Polarity inversion
> does not. An inverting transformer inverts POLARITY,
> not PHASE. A 180-degree phase shift is not the same
> as a polarity inversion. Swapping the + and - wires
> at a speaker's terminals, for instance, provides no
> phase shift at all. It creates a polarity inversion.
>
> A lot of the confusion and controversy on this thread
> results from terminology applied sloppily or just
> plain incorrectly. Although a polarity inversion is
> often loosely known as a phase inversion or a 180-degree
> phase shift, it is not. A polarity inversion has nothing
> to to with frequency, time, or angular measurement.
> No phase, no shift, no delay, no degrees, no radians.
> A phase shift involves time. Use the terms correctly,
> and much of the hoo-ha vanishes.
** This, IMHO, is the best post so far. Those who are immersed in all
the math associated with digital FFT are sometimes dazzeled by the light and
fail to see fundamental physical reality.
The world we all (well most anyway) live in IS the time domain and
NOT the arteficial frequency domain. In the real world one event either
precedes another, is at the same time or is later.
In the frequency domain, time does not exist - how "real" is that?
Regards, Phil
Brad Griffis
"Phil Allison" <bi...@bigpond.com> wrote in message
news:pmUu8.41200$uR5....@newsfeeds.bigpond.com...
Phil,
I agree with you that this is the best post so far. I severely disagree
with you, however, on the importance of the frequency domain. This is an
_audio_ newsgroup, and _frequencies_ are quite important!
The frequency domain is very much real. It is simply a different coordinate
system. For this particular problem I totally agree that to get a 180
degree phase shift it is much better to swap wires rather than build some
kind of system or inverter to do it. Don't act all arrogant as if you can
just toss all signal processing out the window and replace it with wire
swapping or something though.
Brad
Randy Yates
>
> Randy believes phase shift = time delay, as fervently as ancient
> people believed the earth was flat.
The problem you and others have here is that you can't accept
what the mathematics is telling you. I know your anguish. It's
not that I can't relate. My first "degree" was from a hands-on
school - DeVry Institute of Technology - so I value my practical and
experiential knowledge, and I am constantly comparing my theoretical
interpretations with these. I agree that it does seem very odd to
think that all this complicated mess of phase delaying is going on
in a simple old polarity inverter, but this interpretation is a
direct consequence of Fourier theory and linear system theory, and
you effectively reject one when you reject the other.
John Atkinson
> Are there any audio products you can purchase that use practical Hilbert
> transformers? Yes. I'm not sure if it's the DRA Labs MLSSA or the Audio
> Precision System One, but John Atkinson used it to measure the excess
> phase of the Thiele CS6 speakers in the following article:
>
> http://www.stereophile.com/fullarchives.cgi?218
for measuring loudspeakers.
John Atkinson
Editor, Stereophile
Richard D Pierce
Any system that uses MLS stimulus and derives the impulse
response uses a Hilbert transform. DRA MLSSA is one example,
AudioMatica Clio is another.
--
| Dick Pierce |
| Professional Audio Development |
| 1-781/826-4953 Voice and FAX |
| DPi...@world.std.com |
Randy Yates
>
> Randy Yates wrote:
>
> <snip>
>
> >Forgive my ad-hominem attacks. You are absolutely right - they do nothing
> >to further understanding.
>
> Thank-you.
>
> >I think you have a healthy degree of skepticism. I also think it may serve
> >us well if we back up a bit and let me ask you this: Do you believe that
> >any signal (well, most signals we see in the real world anyway) can be
> >represented as a combination of sine waves? I'm not just talking about
> >periodic signals either, but both periodic and aperiodic signals. As
> >an example close to this discussion, consider a 1 kHz sine wave that
> >has been "gated" by a pulse of amplitude 1 that comes on at t=0 and
> >goes off at t=1 [second]. Do you believe this signal can be described
> >by a sum of sine waves? Joe says it can (you know Joe, right? Joseph Fourier?).
> >But wait - how can a signal which is zero for most of the time be represented
> >by a bunch of signals that are zero nowhere (except at isolated points)?
>
> What's this, now? A trick question?
No, Michael, not at all. I am perfectly serious. We have to back up to]
a point on which we can agree in order to move forward. That is the
reason for my question.
> What I believe is that Fourier's
> fine work has little or nothing to do with whether a circuit or algorithm
> that multiplies its input by -1 can be properly considered to apply a
> spectrum of delays in order to invert the polarity of a spectrum of
> frequencies and whether it is practical to attempt to implement a
> circuit or algorithm that does so.
Let me back up even further. Are you willing to consider the
possibility that your belief is flawed?
--
Randy Yates
DSP Engineer, Sony Ericsson Mobile Communications
Research Triangle Park, NC, USA
randy...@ericsson.com, 919-472-1124
.
Phil Allison
"Brad Griffis" <bgri...@uiuc.edu> wrote in message
news:utVu8.41210$tg4.4...@vixen.cso.uiuc.edu...
> >
> > ** This, IMHO, is the best post so far. Those who are immersed in
all the math associated with digital FFT are sometimes dazzeled by the light
and fail to see fundamental physical reality.
The world we all (well most anyway) live in IS the time domain and
> > NOT the arteficial frequency domain. In the real world one event either
precedes another, is at the same time or is later.
> >
> > In the frequency domain, time does not exist - how "real" is
that?
> >
> > Regards, Phil
> >
>
> Phil,
>
> I agree with you that this is the best post so far. I severely disagree
> with you, however, on the importance of the frequency domain. This is an
> _audio_ newsgroup, and _frequencies_ are quite important!
>
> The frequency domain is very much real. It is simply a different
coordinate
> system. For this particular problem I totally agree that to get a 180
> degree phase shift it is much better to swap wires rather than build some
> kind of system or inverter to do it. Don't act all arrogant as if you can
> just toss all signal processing out the window and replace it with wire
> swapping or something though.
>
> Brad
** Frequency IS a very important concept in electronics - I never said
it was not.
The frequency domain is, however, a convenient simplification of
reality - sometimes an oversimplification. It must be applied wisely and
only where appropriate.
If a time domain analysis produces the simplest correct view of what
is happening then that is the one to go for. (Occam's razor applies)
Regards, Phil
Richard D Pierce
Phil Allison <bi...@bigpond.com> wrote:
> ** Frequency IS a very important concept in electronics - I never said
>it was not.
>
> The frequency domain is, however, a convenient simplification of
>reality - sometimes an oversimplification. It must be applied wisely and
>only where appropriate.
No, it is not. It is simply an orthagonal coordinate system of
time. The two are directly interchangeable without loosing any
information whatsoever via the appropriate transform. Once this
is understood, it is as easy to view a signal in the time domain
as it is in the frequency domain.
Beyond that, once must consider the fact that there the
frequency domain itself has DIRECT consequences in the real
world. For example, the physiological response to frequency is
the perception of pitch, and it merely obfuscates the
relationship if we discard frequency in favor of its orthoganl
transform, time.
> If a time domain analysis produces the simplest correct view of what
>is happening then that is the one to go for. (Occam's razor applies)
The time domain produces A view, it is not inherently the
correct view. It might be a more CONVENIENT view, but a view in
the time domain contains precisly the same information as a view
of the same phenomenon in the frequency domain, the relation
being via the appropriate transform.
These concepts are driven solidly home when one finally comes to
term with the classic time-frequency uncertainty principle:
delta F * delta T >= 2
where delta F is the uncertainty in frequency and delta T is the
uncertainty in time. The product of these uncertainties CAN
NEVER be less than a finite number. This has direct and profound
consequences on a practical basis. The classic real-world
example of this is the piano tuner's dilemma. When tuning a
piano by ear, one must listen to the beat or difference
frequency nin order to judge how well the notes are tuned. If
one wants it tuned well (i.e., small uncertainty in frequency),
one must listen for a long time (i.e., large uncertainty in
time) to count the appropriate number of beats. In tuning two
unison strings, if one wanted them at EXACTLY the same pitch (0
uncertainty in frequency), in other words, absolutely no
differences beats at all, one would have to wait forever (i.e.,
infinite uncertainty in time) to prove that no beats existed.
Claiming one orthogonal dimension of a system is more or less
valid than another a priori suggest a lack of understanding of
the domains.
Phil Allison
"Richard D Pierce" <DPi...@TheWorld.com> wrote in message
news:GuoMC...@world.std.com...
>
> No, it is not. It is simply an orthagonal coordinate system of
> time. The two are directly interchangeable without loosing any
> information whatsoever via the appropriate transform.
** And that transform is known as ?
Regards, Phil
Brad Griffis
Phil,
The transform that is able to switch back and forth between time and
frequency with no loss whatsoever is known as the Fourier transform. That's
why you are able to take a signal, take the Fourier transform, then take the
inverse Fourier transform, and end up with exactly what you started with.
There is no loss of information in the process, it is just a different
coordinate system to look at.
Brad
Richard D Pierce
Depends upon what you want to do. In one case, the transform is
nothing more than taking the reciprocal. Take a repetative
signal whose frequency is measured in cycles per second, and the
reciprocal gives the time per cycle, in seconds.
But, the answer I suspect you wanted from me has the word
(fourier" in it, and that is just ONE transform for moving from
one orthogonal coordinate to another.
The point which you are SO fearful of, sir, is that the two are
merely orthogonal views of EXACTLY the same thing: one is NOT
more socrosanct than the other.
Pitty you don't want to realize this, because it opens up a HUGE
realm of understanding and insight.
Pity indeed.
Phil Allison
"Richard D Pierce" <DPi...@TheWorld.com> wrote in message
> > ** And that transform is known as ?
>
> Depends upon what you want to do. In one case, the transform is
> nothing more than taking the reciprocal. Take a repetative
> signal whose frequency is measured in cycles per second, and the
> reciprocal gives the time per cycle, in seconds.
>
> But, the answer I suspect you wanted from me has the word
> (fourier" in it, and that is just ONE transform for moving from
> one orthogonal coordinate to another.
>
> The point which you are SO fearful of, sir, is that the two are
> merely orthogonal views of EXACTLY the same thing: one is NOT
> more socrosanct than the other.
>
> Pitty you don't want to realize this, because it opens up a HUGE
> realm of understanding and insight.
>
> Pity indeed.
** You are making toally unwarranted assumptions about my motives and
mindset.
I asked simply because you did not state the name. Without a name I
can do no research of my own. I actually want to know if I am mistaken about
the "frequency domain" since I have long believed and been told that is is
not possible to switch back and forth with the "time domain".
Perhaps I have mixed the sort of graphs one sees on the screens of a
spectrum analyser with something it should not be called.
I will look further into this and contact my private "experts".
Regards, Phil
Randy Yates
> [...]
> I actually want to know if I am mistaken about
> the "frequency domain" since I have long believed and been told that is is
> not possible to switch back and forth with the "time domain".
That's utterly false. If a function f(t) has a Fourier transform
(and most do, barring those pathological beasts that violate the
so-called Dirichlet conditions), then you can usually go back and
forth between the time and frequency domains perfectly. Even
when you can't, i.e., even when the inverse Fourier transform
doesn't completely match the original function f(t), there is
no energy in their difference.
One common (in the theoretical realm) situation in which they
don't match is when f(t) is discontinuous, as e.g. in a square
wave. In that case, the inverse Fourier transform of the Fourier
transform of f(t) will yield a point that is midway between
the discontinuities.
Please try to forget what you have been told in the past about
not being able to switch back and forth between the time domain
and the frequency domain. You can!
Richard D Pierce
> I actually want to know if I am mistaken about
> the "frequency domain" since I have long believed and been told that is is
> not possible to switch back and forth with the "time domain".
This is preposterous. Whoever told you this is clueless.
Say you have a signal that repeats 1000 times per second. It's
frequency is 1000 Hz. It's period is 1/1000 Hz or 0.001 second.
That's just the most simple and obvious DIRECT transformation
between the frequency and time domains. And the transformation
is PERFECTLY lossless.
The Fourier transform can move your from the time domain view to
the frequency domain view and back again with ease. It's
accuracy in doing so is dependent on the quality of the
implementation AND, as are all such transforms, on the
fundamental limitations imposed by the time-frequency
uncertainty relationship (this is an inherent limitation not of
the transform, but on the very representation of the signal
itself).
If you meet up with the guy that told you this, give him a
dope-slap for me, okay? :-)
Bob_Stanton
>
> But unlike the flat-earthers, he is correct. And what
> evidence is there that ancient people believed the
> earth to be flat? Some fictional event created by
> Washington Irving, maybe?
It is not fiction. The "Flat Earth Society" existed up until the 19th
century.
> Eratosthenes came up with
> a fairly good estimate of the diameter of the Earth
> well before the Christian era -- so at least some
> ancients knew the world to be round.
This is true. Eratosthenes was good! Some ancients knew that the earth
was round, however, many (most?) thought it was flat. It was obvious
in my statement, that I was *refering* to those ancients who thought
the earth was flat.
>
> Phase shift DOES involve time delay. Polarity inversion
> does not. An inverting transformer inverts POLARITY,
> not PHASE. A 180-degree phase shift is not the same
> as a polarity inversion.
Not so. A 180 deg phase shift is *not* the same as a polarity
inversion. When a signal goes into a three terminal black box and
comes out 180 deg out of phase, that is phase shift. The input and
output signals exist at the same time! One can put scope leads on the
input and output and see two signals, 180 deg out of phase with each
other. The example below is not a simple polarity inversion, it is a
phase change.
---------
In -----------0 0------- Out
0
----------
|
Gnd
> Swapping the + and - wires
> at a speaker's terminals, for instance, provides no
> phase shift at all. It creates a polarity inversion.
With your "polarity inversion" only one source signal exits. Reversing
the load does not change the phase of the source. *Nothing has
changed*. Wire + is still plus. Wire - is this minus. All you have
done is reverse the load. (Connected the speaker backwards) There is
no input and output signal that you can put your scope leads on, and
observe a phase change.
>
> A lot of the confusion and controversy on this thread
> results from terminology applied sloppily or just
> plain incorrectly.
I don't think your term "polarity inversion" is very good. I tried to
looked it up in "The Illustrated Dictionary of Electronics", *it
wasn't there*. If we are going to avoid confusion and controversy,
perhaps we should all use terms that are in common usage.
The word "polarity" is normally used with DC or magnetics, not with
AC. For example, if two speakers are connected such that their outputs
cancel, we commonly say: "The speakers are out of phase". We don't
say, "I believe there is a polarity inversion in the system". :-)
> Although a polarity inversion is
> often loosely known as a phase inversion or a 180-degree
> phase shift, it is not. A polarity inversion has nothing
> to to with frequency, time, or angular measurement.
> No phase, no shift, no delay, no degrees, no radians.
Right. A simple polarity inversion, such as reversing leads, does not
involve time.
> A phase shift involves time. Use the terms correctly,
> and much of the hoo-ha vanishes.
Are you defining "phase shift" as involving time? Can you prove it?
Bob Stanton
DuBois
> Brother <nos...@last.com> wrote in message news:<3CBB9080...@last.com>...
> > But unlike the flat-earthers, he is correct. And what
> > evidence is there that ancient people believed the
> > earth to be flat? Some fictional event created by
> > Washington Irving, maybe?
> It is not fiction. The "Flat Earth Society" existed up until the 19th
> century.
They still exist today. What they pretend to believe
for entertainment puropses is irrelevent.
> > Eratosthenes came up with
> > a fairly good estimate of the diameter of the Earth
> > well before the Christian era -- so at least some
> > ancients knew the world to be round.
>
> This is true. Eratosthenes was good! Some ancients knew that the earth
> was round, however, many (most?) thought it was flat. It was obvious
> in my statement, that I was *refering* to those ancients who thought
> the earth was flat.
Can you provide any evidence that many or most ancients
believed this?
> > Phase shift DOES involve time delay. Polarity inversion
> > does not. An inverting transformer inverts POLARITY,
> > not PHASE. A 180-degree phase shift is not the same
> > as a polarity inversion.
>
> Not so. A 180 deg phase shift is *not* the same as a polarity
> inversion.
That's what he said. Duh.
> When a signal goes into a three terminal black box and
> comes out 180 deg out of phase, that is phase shift. The input and
> output signals exist at the same time! One can put scope leads on the
> input and output and see two signals, 180 deg out of phase with each
> other. The example below is not a simple polarity inversion, it is a
> phase change.
So how do you tell the difference between phase and polarity
on a scope? How do you know it's a 180 degree phase change and
not a simple inversion? A sine wave inverted can look just like
a sine wave shifted 180 degrees, on a scope.
> ---------
> In -----------0 0------- Out
> 0
> ----------
> |
> Gnd
>
> > Swapping the + and - wires
> > at a speaker's terminals, for instance, provides no
> > phase shift at all. It creates a polarity inversion.
>
> With your "polarity inversion" only one source signal exits. Reversing
> the load does not change the phase of the source. *Nothing has
> changed*. Wire + is still plus. Wire - is this minus. All you have
> done is reverse the load. (Connected the speaker backwards)
So *Something has changed*
> There is
> no input and output signal that you can put your scope leads on, and
> observe a phase change.
>
> > A lot of the confusion and controversy on this thread
> > results from terminology applied sloppily or just
> > plain incorrectly.
>
> I don't think your term "polarity inversion" is very good. I tried to
> looked it up in "The Illustrated Dictionary of Electronics", *it
> wasn't there*.
Try using a real reference instead of a picture book?
Cheap shot, I know, based only on the title. But
your opinion of the term "polarity inversion" is
irrelevant and ill-informed.
> If we are going to avoid confusion and controversy,
> perhaps we should all use terms that are in common usage.
If we are going to avoid confusion and controversy,
we should use correct terminology rather than popular
but technically incorrect terminology. It's the use
of such sloppy and incorrect terminology as you propose
that, in part, leads to the confusion.
> The word "polarity" is normally used with DC or magnetics, not with
> AC. For example, if two speakers are connected such that their outputs
> cancel, we commonly say: "The speakers are out of phase". We don't
> say, "I believe there is a polarity inversion in the system". :-)
We who like to use technical terms correctly do indeed call
it a polarity inversion, and those who care to propogate
sloppy thinking and technically bogus terminology can call
it what they like. The word polarity is used quite often in
audio. Check out Howard Tremain's Audio Cyclopedia (yes, it
has some pictures for you) and see what he says about those
who confuse polarity with phase. I can't quote it exactly
from memory, and my copy is at work.
> > Although a polarity inversion is
> > often loosely known as a phase inversion or a 180-degree
> > phase shift, it is not. A polarity inversion has nothing
> > to to with frequency, time, or angular measurement.
> > No phase, no shift, no delay, no degrees, no radians.
>
> Right. A simple polarity inversion, such as reversing leads, does not
> involve time.
But you just said that you didn't use the term.
So maybe you have no clue what it means.
> > A phase shift involves time. Use the terms correctly,
> > and much of the hoo-ha vanishes.
> Are you defining "phase shift" as involving time? Can you prove it?
From rane.com:
phase
Audio signals are complex AC (alternating current) periodic phenomena
expressed mathematically as phasors, or vectors. Phase refers to a
particular value of t (time) for any periodic function, i.e. it is the
relationship between a reference point and the fractional part of the
period through which the signal has advanced relative to an arbitrary
origin.
phase shift
The fraction of a complete cycle elapsed as measured from a specified
reference point and expressed as an angle. See RaneNote: Exposing
Equalizer Mythology.
polarity
A signal's electromechanical potential with respect to a reference
potential. For example, if a loudspeaker cone moves forward when a
positive voltage is applied between its red and black terminals, then it
is said to have a positive polarity. A microphone has positive polarity
if a positive pressure on its diaphragm results in a positive output
voltage. [Usage Note: polarity vs. phase shift: polarity refers to a
signal's reference NOT to its phase shift. Being 180° out-of-phase and
having inverse polarity are DIFFERENT things. We wrongly say something
is out-of-phase when we mean it is inverted. One takes time; the other
does not.]
recordingeq.com:
Phase - A measurement (expressed in degrees) of the time difference
between two similar waveforms.
Phase Shift - A delay introduced into an audio signal measured in
degrees delayed.
In my work, the difference between phase and polarity is
an important concept, used on a daily basis.
cjt
No, I think the confusion arises from some participants thinking that shifting
the phase of all components of a signal by a specified angle is the same as
shifting all those components by a time corresponding to a specified angle of
some _dominant_ component.
Larry Brasfield
Richard D Pierce (DPi...@TheWorld.com) says...
> Phil Allison wrote:
> > I actually want to know if I am mistaken about
> > the "frequency domain" since I have long believed and been told that is is
> > not possible to switch back and forth with the "time domain".
>
> This is preposterous. Whoever told you this is clueless.
>
> Say you have a signal that repeats 1000 times per second. It's
> frequency is 1000 Hz. It's period is 1/1000 Hz or 0.001 second.
>
> That's just the most simple and obvious DIRECT transformation
> between the frequency and time domains. And the transformation
> is PERFECTLY lossless.
As you've described it, phase was lost.
(And amplitude, too, for that matter.)
Coincidentally, that is the same reason
so many people are led to believe that
the Fourier transform and various DFT's
are somehow lossy. The result of the
conversion to the frequency domain is
often stripped of its phase, and when
that modified result is converted back
to the time domain, a signal differing
from the original is seen.
It is only by keeping the full complex
value of the transform output that it
can be reversed, and then, at least
from a mathematical viewpoint, it can
be perfectly reversed.
> The Fourier transform can move your from the time domain view to
> the frequency domain view and back again with ease. It's
> accuracy in doing so is dependent on the quality of the
> implementation AND, as are all such transforms, on the
> fundamental limitations imposed by the time-frequency
> uncertainty relationship (this is an inherent limitation not of
> the transform, but on the very representation of the signal
> itself).
The Fourier transform, when applied to signals
meeting certain mathematical criteria, (which
all real-world signals meet), is also perfectly
lossless. The accuracy of the transformation
depends on nothing resembling a "time-frequency
uncertainty relationship".
I suppose there are limitations to representing
signals, probably related to quantum limits,
but I don't see how these are really relevant
to the reversability of frequency transforms.
> If you meet up with the guy that told you this, give him a
> dope-slap for me, okay? :-)
I'm tempted to say something impudent here.
Please give me credit for refraining ;-)
--
-Larry Brasfield
(address munged)
John Woodgate
out of phase', on Tue, 16 Apr 2002:
>Please try to forget what you have been told in the past about
>not being able to switch back and forth between the time domain
>and the frequency domain. You can!
This is the sort of confusion that occurs when a mathematically rigorous
statement is introduced without s follow up of 'in practical terms...'.
It starts with pi; the exact value is unknowable but it doesn't matter.
In the same way, as you indicated, rigorously you may not be able to
switch between time and frequency domains but, provided you keep your
wits about you when dealing with discontinuous functions (and one or two
others which, AFAIK, do not arise in electrical signals but can in
fields), you can.
--
Regards, John Woodgate, OOO - Own Opinions Only. http://www.jmwa.demon.co.uk
Interested in professional sound reinforcement and distribution? Then go to
http://www.isce.org.uk
PLEASE do NOT copy news posts to me by E-MAIL!
John Woodgate
(in <67ef4d54.02041...@posting.google.com>) about 'Shifting
audio signal 180 degrees out of phase', on Tue, 16 Apr 2002:
>Brother <nos...@last.com> wrote in message news:<3CBB9080...@last.com>...
>
>>
>> But unlike the flat-earthers, he is correct. And what
>> evidence is there that ancient people believed the
>> earth to be flat? Some fictional event created by
>> Washington Irving, maybe?
>
>It is not fiction. The "Flat Earth Society" existed up until the 19th
>century.
I went to a lecture by a Flat-Earthist in about 1965.
[snip]
>
>>
>> Phase shift DOES involve time delay. Polarity inversion
>> does not. An inverting transformer inverts POLARITY,
>> not PHASE. A 180-degree phase shift is not the same
>> as a polarity inversion.
>
>Not so. A 180 deg phase shift is *not* the same as a polarity
>inversion. When a signal goes into a three terminal black box and
>comes out 180 deg out of phase, that is phase shift. The input and
>output signals exist at the same time! One can put scope leads on the
>input and output and see two signals, 180 deg out of phase with each
>other. The example below is not a simple polarity inversion, it is a
>phase change.
>
> ---------
>In -----------0 0------- Out
> 0
> ----------
> |
> Gnd
>
>
I just don't understand that argument at all. Suppose what is in the box
is an op-amp unity-gain inverter. The input and output signals exist at
the same time. There is no delay.
>
>
>> Swapping the + and - wires
>> at a speaker's terminals, for instance, provides no
>> phase shift at all. It creates a polarity inversion.
>
>
>With your "polarity inversion" only one source signal exits. Reversing
>the load does not change the phase of the source.
Neither does a 3-terminal box that introduces a '180 deg. phase change'.
>*Nothing has
>changed*. Wire + is still plus. Wire - is this minus. All you have
>done is reverse the load. (Connected the speaker backwards) There is
>no input and output signal that you can put your scope leads on, and
>observe a phase change.
Yes, the loudspeaker case is not a very good example. The unity-gain
inverter is a better one IMHO.
>
>>
>> A lot of the confusion and controversy on this thread
>> results from terminology applied sloppily or just
>> plain incorrectly.
>
>I don't think your term "polarity inversion" is very good. I tried to
>looked it up in "The Illustrated Dictionary of Electronics", *it
>wasn't there*. If we are going to avoid confusion and controversy,
>perhaps we should all use terms that are in common usage.
Get a) a better dictionary, b) a copy of AES standard AES-26-2001, 'AES
recommended practice for professional audio - Conservation of the
polarity of audio signals'. AES standards can now be obtained FREE
through the AES web site.
>
>The word "polarity" is normally used with DC or magnetics, not with
>AC. For example, if two speakers are connected such that their outputs
>cancel, we commonly say: "The speakers are out of phase". We don't
>say, "I believe there is a polarity inversion in the system". :-)
No, we don't, but we should. It is the correct term.
>
>> Although a polarity inversion is
>> often loosely known as a phase inversion or a 180-degree
>> phase shift, it is not. A polarity inversion has nothing
>> to to with frequency, time, or angular measurement.
>> No phase, no shift, no delay, no degrees, no radians.
>
>Right. A simple polarity inversion, such as reversing leads, does not
>involve time.
>
>> A phase shift involves time. Use the terms correctly,
>> and much of the hoo-ha vanishes.
>
>Are you defining "phase shift" as involving time? Can you prove it?
>
I can't determine where the basis of the disagreement lies. Perhaps this
is it:
a) If the polarity of a non-sinusoidal waveform is inverted, the shape
is preserved.
b) If *each sinusoidal (or cisoidal, to be very pedantic) component of a
non-sinusoidal waveform is phase-shifted by 180 degrees*, the shape is
NOT preserved.
But when we loosely talk about a '180 deg. phase-shift', we don't mean
b), we mean a).
Arny Krueger
> I actually want to know if I am mistaken about
> the "frequency domain" since I have long believed and been told
that is
> not possible to switch back and forth with the "time domain".
Practical demonstrations of the ability to move real-world signals
between the time and frequency domain with excellent accuracy can
easily be done using FFT-based equalizers such as the one in Cool
Edit (shareware downloadable from www.syntrillium.com).
CoolEdit transforms time-sampled audio signals into the frequency
domain, modifies the real part of the transformed signal based on a
user-supplied frequency response curve, and transforms the results
back into a time-sampled signal. It also performs similar operations
with user-supplied time-sampled impulse response datasets.
Both the forward and inverse transformations still take place when
you apply a flat response curve with a FFT filter.
You should listen to the results you get when you use this FFT-based
of filter to apply a flat frequency response curve to music. The
results should be, and seem to be audibly indistinguishable from the
original files.
A useful tool for doing listening tests comparing the "Before and
after" results of FFT-based transformations of audio signals can be
downloaded from www.pcabx.com .
John Woodgate
<larry_b...@snotmail.com> wrote (in <MPG.1726ada19539648d989713@new
s.qwest.net>) about 'Shifting audio signal 180 degrees out of phase', on
Tue, 16 Apr 2002:
>Please give me credit for refraining ;-)
Of chorus. (;-)
Richard D Pierce
Phil Allison <bi...@bigpond.com> wrote:
>"Richard D Pierce" <DPi...@TheWorld.com> wrote in message
>news:Guouv...@world.std.com...
>> > ** And that transform is known as ?
>> Depends upon what you want to do. In one case, the transform is
>> nothing more than taking the reciprocal. Take a repetative
>> signal whose frequency is measured in cycles per second, and the
>> reciprocal gives the time per cycle, in seconds.
>>
>> But, the answer I suspect you wanted from me has the word
>> (fourier" in it, and that is just ONE transform for moving from
>> one orthogonal coordinate to another.
>>
>> The point which you are SO fearful of, sir, is that the two are
>> merely orthogonal views of EXACTLY the same thing: one is NOT
>> more socrosanct than the other.
>>
>> Pitty you don't want to realize this, because it opens up a HUGE
>> realm of understanding and insight.
>mindset.
I made absolutely no assumptions about your mindset, sir. Your
mindset is something YOU describe in clear and unambiguous terms
when you stated, and I quote:
"I have long believed and been told that is is not possible
to switch back and forth with the "time domain"."
From that statement, I took it ("assumed" if you will) that your
mindset was such that you did not believe you could switch back
and forth with the time domain. If I assumed incorrectly, please
show me how "you do not believe you could switch back and forth
from the time domain" is different from "you do not believe you
could switch back and forth from the time domain."
Frnakly, the subtlety of the distinction is lost on me.
>I asked simply because you did not state the name.
There is no "one" name. There are several such transforms. The
reciprocal example I gave is just one example. A trivial one,
but an example none the less. The common name is the FOurier
transform and it variants, the DFT, the FFT and others.
> Without a name I
>can do no research of my own.
Several others have mentioned the FT. It's a common concept in
signal analysis.
>Perhaps I have mixed the sort of graphs one sees on the screens of a
>spectrum analyser with something it should not be called.
"Spectrum analyzer" is something of a vague term. It can mean a
lot of things to a lot of people. The constant-percentage-
bandwidth analyzer, like so-called 1/3 octave "real time
analyzers" loose a lot of information and cannot perform
reversable transforms. They were never intended to. But many
people call them "spectrum analyzers." Other "spectrum
analyzers" may use FT techniques, but, as a matter of operator
convenience, throw away half the data, thus making them
non-reversable transforms.
But, rest assured, if you are dealing with signals of finite
duration, energy and bandwidth (name a real-world signal that
does NOT meet this criteria), an FT can move you from the time
to the frequency domain and back again at will, NO exceptions.
Perhaps more importantly than trying to put a name on one
specific operation, you'd be better served by looking at it in a
more fundamental view. The implications of the mentioned
time-frequency uncertainty relation is one place to start: it
forces you to confront the fact that the time and frequency
domains are not only related, THEY ARE THE SAME THING, simply
viewed froma different perspective.
At an even more fundamental level, consider the basic inits of
each domain:
Time - fundamental unit is "seconds"
Frequency - fundamental unit is "1/seconds"
This is not a happy coincdidence: it is a DIRECT and INEVITCABLE
consequence of the direct link two ordinate systems.
I fear you are focusing on a specific mathematical operation as
"proof" of either the link or lack thereof, and that's a
fundamental mistake. The link is there even if we didn't have
Monsieur Jean Baptiste Joseph Fourier to give us one tool set to
navigate within the linked realms.
I keep returning to the time-frequency uncertainty relation,
because it provides an example of the fundamental link between
the two views of the domain independent of the FT: If Mr.
Fourier had not come along, piano tuners would still have to
confront the fact that accurate tuning requires a long time:
that's NOT because of the Fourier transform or any other
mathematical artifice: IT'S BECAUSE THE TIME AND FREQUENCY
DOMAIN ARE INEXTRICABLY LINKED AND SIMPLY ORTHOGONAL VIEWS OF
THE SAME PHENOMENON.
Get THAT concept down first. Then go out and look at the
available toolsets, such as the FT, the DFT, the FFT and all the
rest as a means of navigating the waters. But, without an
dunderstanding of the fundamental link, the toolsets will do you
no good.
>I will look further into this and contact my private "experts".
Perhaps, given what you have told us about their expertise, you
might want to consider other sources.
Richard D Pierce
John Woodgate <j...@jmwa.demon.co.uk> wrote:
>I read in rec.audio.tech that Larry Brasfield
><larry_b...@snotmail.com> wrote (in <MPG.1726ada19539648d989713@new
>s.qwest.net>) about 'Shifting audio signal 180 degrees out of phase', on
>Tue, 16 Apr 2002:
>
>>Please give me credit for refraining ;-)
>
>Of chorus. (;-)
I'd hope that you'd not stoop solo and change the entire tenor
of the ddiscussion, since your entire premise is built on a
flimsy bass. I'd rather see you give your alto a better
argument.
Bob_Stanton
...
> > It is not fiction. The "Flat Earth Society" existed up until the 19th
> > century.
>
> They still exist today. What they pretend to believe
> for entertainment puropses is irrelevent.
>
Oh no, it's not irrelevent. Those who believed the earth was flat,
based their belief on the "obvious" fact they could see it was flat.
Those who believe phase shift shows a time delay, base their opinions
on waveforms on an oscilloscope. What appears obvious is not always
correct, whether it is looking at the earth, or looking at an
oscilloscope.
> Can you provide any evidence that many or most ancients
> believed this?
>
Can you provide any evidence that most ancients didn't?
> > > Phase shift DOES involve time delay. Polarity inversion
> > > does not. An inverting transformer inverts POLARITY,
> > > not PHASE. A 180-degree phase shift is not the same
> > > as a polarity inversion.
> >
> > Not so. A 180 deg phase shift is *not* the same as a polarity
> > inversion.
>
> That's what he said. Duh.
>
We all seem to agree that a phase shift is not the same as polarity
inversion. I'm glad you noticed that. You have a remarkable ability to
see the obvious. :-)
> So how do you tell the difference between phase and polarity
> on a scope? How do you know it's a 180 degree phase change and
> not a simple inversion? A sine wave inverted can look just like
> a sine wave shifted 180 degrees, on a scope.
>
I'm saying, if the output signal is 180 deg out of phase with the
input, than it is "PHASE SHIFTED". You want to call it "polarity
inverted". It is not polarity inverted, it is phase shifted.
> > ----------
> > In -----------0 0------- Out
> > 0
> > ----------
> > |
> > Gnd
> >
> > > Swapping the + and - wires
> > > at a speaker's terminals, for instance, provides no
> > > phase shift at all. It creates a polarity inversion.
> >
> > With your "polarity inversion" only one source signal exits. Reversing
> > the load does not change the phase of the source. *Nothing has
> > changed*. Wire + is still plus. Wire - is this minus. All you have
> > done is reverse the load. (Connected the speaker backwards)
>
> So *Something has changed*
Nothing has changed with the source signal. There has been a polarity
inversion, but not a phase shift.
>
> Try using a real reference instead of a picture book?
> Cheap shot, I know, based only on the title.
I see you are able to draw conclusions about a book you have never
seen, knowing only the title. You must have incredible analytic
ability, or psychic powers.
> But
> your opinion of the term "polarity inversion" is
> irrelevant and ill-informed.
>
That's your opinion, about my opinion.
> > If we are going to avoid confusion and controversy,
> > perhaps we should all use terms that are in common usage.
>
>
> We who like to use technical terms correctly do indeed call
> it a polarity inversion, and those who care to propogate
> sloppy thinking and technically bogus terminology can call
> it what they like. The word polarity is used quite often in
> audio. Check out Howard Tremain's Audio Cyclopedia (yes, it
> has some pictures for you) and see what he says about those
> who confuse polarity with phase. I can't quote it exactly
> from memory, and my copy is at work.
>
Maybe if you try real hard, you can get your memory to work and give
us a quote.
> > > Although a polarity inversion is
> > > often loosely known as a phase inversion or a 180-degree
> > > phase shift, it is not. A polarity inversion has nothing
> > > to to with frequency, time, or angular measurement.
> > > No phase, no shift, no delay, no degrees, no radians.
> >
> > Right. A simple polarity inversion, such as reversing leads, does not
> > involve time.
>
> But you just said that you didn't use the term.
You are surprised that I speak your language ? :-)
> So maybe you have no clue what it means.
So maybe *you* have no clue what phase shift means. Let me give you a
quote from my handy-dandy little pictionary.
"Phase Shift
1. A change in the dispalcement, as funtion of time, of a periodic
disturbance having constant frequency. 2. The magnitude of such a
change, measured in fractions of a wavelength or in electrical
degrees."
..
> > Are you defining "phase shift" as involving time? Can you prove it?
>
> From rane.com:
> phase
> Audio signals are complex AC (alternating current) periodic phenomena
> expressed mathematically as phasors, or vectors. Phase refers to a
> particular value of t (time) for any periodic function, i.e. it is the
> relationship between a reference point and the fractional part of the
> period through which the signal has advanced relative to an arbitrary
> origin.
>
Yes, "phase" can be expressed as: time or angle. But not as "delay".
Don't confuse time with delay.
> phase shift
> The fraction of a complete cycle elapsed as measured from a specified
> reference point and expressed as an angle. See RaneNote: Exposing
> Equalizer Mythology.
>
Note that it says, expressed in angle.
> polarity
> A signal's electromechanical potential with respect to a reference
> potential. For example, if a loudspeaker cone moves forward when a
> positive voltage is applied between its red and black terminals, then it
> is said to have a positive polarity. A microphone has positive polarity
> if a positive pressure on its diaphragm results in a positive output
> voltage. [Usage Note: polarity vs. phase shift: polarity refers to a
> signal's reference NOT to its phase shift. Being 180° out-of-phase and
> having inverse polarity are DIFFERENT things. We wrongly say something
> is out-of-phase when we mean it is inverted. One takes time; the other
> does not.]
>
We agree that phase shift and polarity inversion are not the same
thing. Since the subject of this thread is phase shift, your (above)
statement is irrelivant.
> recordingeq.com:
> Phase - A measurement (expressed in degrees) of the time difference
> between two similar waveforms.
That definition I agree with. My little Illustrated Electronics
Pictionary says: phase difference is, "1. The difference (in time,
angle or fractional cycle) between the instants at which two
alternating quantities reach a given value."
By your book's definition of "phase", and also by my book's definition
of "phase difference": the example I gave of (an ideal inverting
transformer), shows a true phase shift.
> Phase Shift - A delay introduced into an audio signal measured in
> degrees delayed.
That is *not* correct. A "phase shift" is a difference expressed in
time (or angle).
Phil Allison
"Richard D Pierce" <DPi...@TheWorld.com> wrote in message
> >> Pitty you don't want to realize this, because it opens up a HUGE
> >> realm of understanding and insight.
> > ** You are making toally unwarranted assumptions about my motives
and mindset.
>
> I made absolutely no assumptions about your mindset, sir. Your
> mindset is something YOU describe in clear and unambiguous terms
> when you stated, and I quote:
>
> "I have long believed and been told that is is not possible
> to switch back and forth with the "time domain"."
>
> From that statement, I took it ("assumed" if you will) that your
> mindset was such that you did not believe you could switch back
> and forth with the time domain. If I assumed incorrectly, please
> show me how "you do not believe you could switch back and forth
> from the time domain" is different from "you do not believe you
> could switch back and forth from the time domain."
>
> Frnakly, the subtlety of the distinction is lost on me.
** I think you should hop back into the time domain yourself. The
statement you quote of mine was made after and in response your insulting
remarks, not before, so cannot be used to justify them.
You have also quoted it out of context and so it has lost its meaning.
The time when events happen is important - you seem to be very
careless about this. That is a worry.
Regards, Phil