Goto page Previous 1, 2, 3, 4, 5, 6, 7 Next
tm
Guest
Thu Sep 02, 2010 7:56 pm
"John Larkin" <jjlarkin_at_highNOTlandTHIStechnologyPART.com> wrote in message
news:g7sv76lom9pbpja13eii48vaanut9oeckl_at_4ax.com...
Quote:
On Thu, 2 Sep 2010 12:18:08 -0400, "tm" <the_obamunist_at_whitehouse.gov
wrote:
"John Larkin" <jjlarkin_at_highNOTlandTHIStechnologyPART.com> wrote in
message
news:rakr76tpoiph198ldaiq5uvjpneebuvlbt_at_4ax.com...
I'm expecting to get some photodiode pulses that are just a bit too
fast to handle with cheapish amps and comparators and such. It would
be nice to have an analog filter that would accept a roughly gaussian
pulse, maybe 2 ns wide, and stretch it to, say, 5 or 6 ns wide,
substantially flat on top if possible. Rep-rate might go up to 40 MHz
maybe.
An LC phase-linear lowpass filter with a reasonable number of poles
would make a slower sorta gaussian blip, not very flat, with a
substantial tail, which would limit my rep-rate to some extent.
If I run the pulse through a tapped analog delay line, maybe five 1 ns
taps, and sum the signals that appears at each tap, I can get a pretty
flat pulse. That amounts to a FIR/transversal filter with all
coefficients = 1, tweakable a little maybe. That's OK if I can get and
afford such a delay line and can sum the tap signals without great
hassles.
We were playing around with using a 3 or maybe 5 pole LC lowpass
filter, but summing the signals from intermediate nodes, instead of
just taking the last one. This looks promising but mathematically
messy to do really well, a "lost in space" situation maybe. A filter
that makes a beautiful output pulse can have some ghastly intermediate
waveforms.
Any ideas? What sort of filter has a rectangular-pulse impulse
response?
John
How about a short piece of transmission line open on the far end. It will
reflect
back with twice the delay and add to the original pulse.
That would double the pulse length.
tm
That almost works.
I think this works:
input----50r-----+-------integrator----out
|
|
|
|
| 50r transmission
| line
|
|
gnd
It has the ideal transfer function, impulse in and rectangular pulse
out.
John
But I think you want the line open so the reflected pulse will be positive.
Shouldn't take much to find out.
tm
---
news://freenews.netfront.net/ - complaints: news_at_netfront.net ---
Richard Henry
Guest
Fri Sep 03, 2010 1:27 am
On Sep 2, 11:52 am, John Larkin
<jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
Quote:
On Thu, 2 Sep 2010 09:32:53 -0700 (PDT), Richard Henry
pomer...@hotmail.com> wrote:
On Sep 1, 4:15 am, John Larkin
jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
On Tue, 31 Aug 2010 21:44:07 -0700 (PDT), "m...@sushi.com"
m...@sushi.com> wrote:
On Aug 31, 9:29 pm, John Larkin
jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
I'm expecting to get some photodiode pulses that are just a bit too
fast to handle with cheapish amps and comparators and such. It would
be nice to have an analog filter that would accept a roughly gaussian
pulse, maybe 2 ns wide, and stretch it to, say, 5 or 6 ns wide,
substantially flat on top if possible. Rep-rate might go up to 40 MHz
maybe.
An LC phase-linear lowpass filter with a reasonable number of poles
would make a slower sorta gaussian blip, not very flat, with a
substantial tail, which would limit my rep-rate to some extent.
If I run the pulse through a tapped analog delay line, maybe five 1 ns
taps, and sum the signals that appears at each tap, I can get a pretty
flat pulse. That amounts to a FIR/transversal filter with all
coefficients = 1, tweakable a little maybe. That's OK if I can get and
afford such a delay line and can sum the tap signals without great
hassles.
We were playing around with using a 3 or maybe 5 pole LC lowpass
filter, but summing the signals from intermediate nodes, instead of
just taking the last one. This looks promising but mathematically
messy to do really well, a "lost in space" situation maybe. A filter
that makes a beautiful output pulse can have some ghastly intermediate
waveforms.
Any ideas? What sort of filter has a rectangular-pulse impulse
response?
John
Why wouldn't you just stretch the pulses with logic circuits? Your
delay element would be a string of inverters.
It has to be analog and linear. Downstream will be amplifiers and
comparators, as noted.
Another possibility is two cascaded Bessel filters. The first shapes
the 2 ns pulse into, say, a 6 ns gaussian pulse, and the second pretty
much just delays that. The sum of the filter outputs will be pretty
much flat and will settle out fast.
John
John
If the pulses are always the same height and polarity, there is no
need for such complexity.
If the pulses were always the same height, I wouldn't have to measure
them.
With all the signal torture you have suggested, you won't be measuring
anything like the original pulses.
Sample the pulse source at a rate fast enough to detect the shortest
one and store the samples in a ring buffer of sufficient size to store
the longest one . When a pulse is detected, store the contents of the
buffer in a more permanent location and analyze them at your leisure.
John Larkin
Guest
Fri Sep 03, 2010 3:49 am
On Thu, 2 Sep 2010 15:27:47 -0700 (PDT), Richard Henry
<pomerado_at_hotmail.com> wrote:
Quote:
On Sep 2, 11:52 am, John Larkin
jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
On Thu, 2 Sep 2010 09:32:53 -0700 (PDT), Richard Henry
pomer...@hotmail.com> wrote:
On Sep 1, 4:15 am, John Larkin
jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
On Tue, 31 Aug 2010 21:44:07 -0700 (PDT), "m...@sushi.com"
m...@sushi.com> wrote:
On Aug 31, 9:29 pm, John Larkin
jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
I'm expecting to get some photodiode pulses that are just a bit too
fast to handle with cheapish amps and comparators and such. It would
be nice to have an analog filter that would accept a roughly gaussian
pulse, maybe 2 ns wide, and stretch it to, say, 5 or 6 ns wide,
substantially flat on top if possible. Rep-rate might go up to 40 MHz
maybe.
An LC phase-linear lowpass filter with a reasonable number of poles
would make a slower sorta gaussian blip, not very flat, with a
substantial tail, which would limit my rep-rate to some extent.
If I run the pulse through a tapped analog delay line, maybe five 1 ns
taps, and sum the signals that appears at each tap, I can get a pretty
flat pulse. That amounts to a FIR/transversal filter with all
coefficients = 1, tweakable a little maybe. That's OK if I can get and
afford such a delay line and can sum the tap signals without great
hassles.
We were playing around with using a 3 or maybe 5 pole LC lowpass
filter, but summing the signals from intermediate nodes, instead of
just taking the last one. This looks promising but mathematically
messy to do really well, a "lost in space" situation maybe. A filter
that makes a beautiful output pulse can have some ghastly intermediate
waveforms.
Any ideas? What sort of filter has a rectangular-pulse impulse
response?
John
Why wouldn't you just stretch the pulses with logic circuits? Your
delay element would be a string of inverters.
It has to be analog and linear. Downstream will be amplifiers and
comparators, as noted.
Another possibility is two cascaded Bessel filters. The first shapes
the 2 ns pulse into, say, a 6 ns gaussian pulse, and the second pretty
much just delays that. The sum of the filter outputs will be pretty
much flat and will settle out fast.
John
John
If the pulses are always the same height and polarity, there is no
need for such complexity.
If the pulses were always the same height, I wouldn't have to measure
them.
With all the signal torture you have suggested, you won't be measuring
anything like the original pulses.
I'll be measuring the pulse energy, which is what matters. A simple
passive pulse stretcher, if I can find one, greatly reduces the cost
of downstream electronics.
Quote:
Sample the pulse source at a rate fast enough to detect the shortest
one and store the samples in a ring buffer of sufficient size to store
the longest one . When a pulse is detected, store the contents of the
buffer in a more permanent location and analyze them at your leisure.
With 2 ns pulses, the sample rate would be 5 GHz or so, with maybe 10
bits of ADC resolution. That's 50 gigabits of data per second. The
"leisure" I have for analysis is hundreds of nanoseconds.
John
John Larkin
Guest
Fri Sep 03, 2010 3:53 am
On Thu, 2 Sep 2010 19:05:45 -0700 (PDT), Bill Sloman
<bill.sloman_at_ieee.org> wrote:
Quote:
On Sep 3, 2:07 am, John Larkin
jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
On Thu, 2 Sep 2010 06:53:29 -0700 (PDT),BillSloman
bill.slo...@ieee.org> wrote:
On Sep 2, 8:39 pm, brent <buleg...@columbus.rr.com> wrote:
On Sep 1, 1:06 am,BillSloman<bill.slo...@ieee.org> wrote:
An infinitely long FIR filter, for one.
Isn't that an oxymoron?
Absolutely. A rectangular pulse response is equally unattainable.
What about this?
in----+-----delay line------------+
| |
| sum -----integrate----out
| |
+-----------(-1)------------+
Its impulse response is a rectangular pulse and it's linear.
Not with a real delay line. They always degrade the rise and fall
times to some extent.
Yes. That would make the fall time of the impulse-provoked rectangular
pulse slower, but it still winds up at zero.
Since I want a flattish pulse to analyze, and recovery to baseline
before the next one hits, that might do.
John
Bill Sloman
Guest
Fri Sep 03, 2010 5:04 am
On Sep 3, 12:02 am, Bill Sloman <bill.slo...@ieee.org> wrote:
Quote:
On Sep 2, 3:39 pm, whit3rd <whit...@gmail.com> wrote:
On Sep 1, 7:24 pm, John Larkin
jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
The bad news is that few people still make analog delay lines, and
they want 4 weeks to come up with a sample. I don't have 4 weeks. So
either I make my own tapped LC delay line from parts, or try cascading
and summing Bessel filters.
For 5 ns, you need about five feet of twinlead or coax cable to make a
delay line.
Probably there's a way to get it without a four week wait.
The propagation delay of light in air (or vacuum) is about 1nsec per
foot. The propagation delay in the dielelctric of coaxial cable is
slower, at about 1.5nsec per foot. You can get 1.1mm OD teflon
dielecric minature coax, and three feet of that can be coiled into a
reasonably compact bundle (and I've used that as a delay line).
Twisted pair is going to have a mixed dielectric and experiment would
seem to be called for. Twisted transformer wire - with enamel
insulation - makes a remarkably compact transmission line, but I've no
idea of the propagation delay, beyond that it too has a mixed
dielectric (air plus enamel).
Come to think of it, buried strip-line in a printed circuit board is a
non-dispersive transmission line. You'd need a four layer board in
which to bury the strip-line - and six layers would allow to stack two
layers.
Figuring on 0.004 inch tracks and track spacing, and an 0.004 inch
wide ground finger between adjacent tracks, the structure is only
0.016 inches wide, and you could get five feet of strip-line into a
square inch of board space (half a square inch with a six layer
board). It's difficult to get higher than 50R track impedance in
buried strip-line - the tracks start to get very narrow - but ti might
ber worth looking at.
--
Bill Sloman, Nijmegen
Bill Sloman
Guest
Fri Sep 03, 2010 5:05 am
On Sep 3, 2:07 am, John Larkin
<jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
Quote:
On Thu, 2 Sep 2010 06:53:29 -0700 (PDT),BillSloman
bill.slo...@ieee.org> wrote:
On Sep 2, 8:39 pm, brent <buleg...@columbus.rr.com> wrote:
On Sep 1, 1:06 am,BillSloman<bill.slo...@ieee.org> wrote:
An infinitely long FIR filter, for one.
Isn't that an oxymoron?
Absolutely. A rectangular pulse response is equally unattainable.
What about this?
in----+-----delay line------------+
| |
| sum -----integrate----out
| |
+-----------(-1)------------+
Its impulse response is a rectangular pulse and it's linear.
Not with a real delay line. They always degrade the rise and fall
times to some extent.
--
Bill Sloman, Nijmegen
Bill Sloman
Guest
Fri Sep 03, 2010 6:05 am
On Sep 3, 12:53 pm, John Larkin
<jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
Quote:
On Thu, 2 Sep 2010 19:05:45 -0700 (PDT),BillSloman
bill.slo...@ieee.org> wrote:
On Sep 3, 2:07 am, John Larkin
jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
On Thu, 2 Sep 2010 06:53:29 -0700 (PDT),BillSloman
bill.slo...@ieee.org> wrote:
On Sep 2, 8:39 pm, brent <buleg...@columbus.rr.com> wrote:
On Sep 1, 1:06 am,BillSloman<bill.slo...@ieee.org> wrote:
An infinitely long FIR filter, for one.
Isn't that an oxymoron?
Absolutely. A rectangular pulse response is equally unattainable.
What about this?
in----+-----delay line------------+
| |
| sum -----integrate----out
| |
+-----------(-1)------------+
Its impulse response is a rectangular pulse and it's linear.
Not with a real delay line. They always degrade the rise and fall
times to some extent.
Yes. That would make the fall time of the impulse-provoked rectangular
pulse slower, but it still winds up at zero.
Since I want a flattish pulse to analyze, and recovery to baseline
before the next one hits, that might do.
I'm sure that it would. A pulse with fast - but finite - rise and fall
times is all that you need (provided that the transition times aren't
too long) and that is compatible with real components. And I'm pretty
sure that around a square inch of buried strip-line would serve as a
more than adequate delay line to do the job.
--
Bill Sloman, Nijmegen
miso@sushi.com
Guest
Fri Sep 03, 2010 10:49 am
On Sep 2, 6:59 am, John Larkin
<jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
Quote:
On Wed, 1 Sep 2010 23:20:38 -0700 (PDT), "m...@sushi.com"
m...@sushi.com> wrote:
On Sep 1, 7:44 pm, John Larkin
jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
On Wed, 1 Sep 2010 18:04:58 -0700 (PDT), "m...@sushi.com"
m...@sushi.com> wrote:
On Sep 1, 4:15 am, John Larkin
jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
On Tue, 31 Aug 2010 21:44:07 -0700 (PDT), "m...@sushi.com"
m...@sushi.com> wrote:
On Aug 31, 9:29 pm, John Larkin
jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
I'm expecting to get some photodiode pulses that are just a bit too
fast to handle with cheapish amps and comparators and such. It would
be nice to have an analog filter that would accept a roughly gaussian
pulse, maybe 2 ns wide, and stretch it to, say, 5 or 6 ns wide,
substantially flat on top if possible. Rep-rate might go up to 40 MHz
maybe.
An LC phase-linear lowpass filter with a reasonable number of poles
would make a slower sorta gaussian blip, not very flat, with a
substantial tail, which would limit my rep-rate to some extent.
If I run the pulse through a tapped analog delay line, maybe five 1 ns
taps, and sum the signals that appears at each tap, I can get a pretty
flat pulse. That amounts to a FIR/transversal filter with all
coefficients = 1, tweakable a little maybe. That's OK if I can get and
afford such a delay line and can sum the tap signals without great
hassles.
We were playing around with using a 3 or maybe 5 pole LC lowpass
filter, but summing the signals from intermediate nodes, instead of
just taking the last one. This looks promising but mathematically
messy to do really well, a "lost in space" situation maybe. A filter
that makes a beautiful output pulse can have some ghastly intermediate
waveforms.
Any ideas? What sort of filter has a rectangular-pulse impulse
response?
John
Why wouldn't you just stretch the pulses with logic circuits? Your
delay element would be a string of inverters.
It has to be analog and linear. Downstream will be amplifiers and
comparators, as noted.
Another possibility is two cascaded Bessel filters. The first shapes
the 2 ns pulse into, say, a 6 ns gaussian pulse, and the second pretty
much just delays that. The sum of the filter outputs will be pretty
much flat and will settle out fast.
John
John
I claim pulse stretching is not a linear operation.
What I want to do certainly is. A passive LC filter is linear. The
output is strictly proportional to the input.
John
Then you are not stretching the pulse. Think of it this way. The input
pulse is of amplitude A and width W. The output pulse is of amplitude
A and width Y, where Y>W. The energy of the output is greater than
that of the input, hence my reference to Parseval.
Why would the output amplitude be unchanged? That's your assumption,
not mine. COE is meaningless here anyhow, since the network, even if
passive, even if the output amplitude is unchanged, could impedance
scale. Voltage is not energy.
Oh, eh yeah, QED. ;-)
By your reasoning, any lowpass filter (which my stretcher is) is
nonlinear. And a coaxial cable must be nonlinear. Only a Brickenbox
network is linear.
I guess you have your own working definition of "nonlinear", which
isn't the one they taught us in engineering school.
We thought a network was linear if the output scales exactly with the
amplitude of the input.
John
Uh, energy needs to be conserved. Your stetcher is so vaguely defined,
it is hard to judge. Now if you say the amplitude is lower, then so be
it. It would have helped if you thought of that sooner.
When the hell did I say lowpass filters are not linear? Uh, nowhere,
thank you very much.
miso@sushi.com
Guest
Fri Sep 03, 2010 10:57 am
On Sep 2, 9:07 am, John Larkin
<jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
Quote:
On Thu, 2 Sep 2010 06:53:29 -0700 (PDT), Bill Sloman
bill.slo...@ieee.org> wrote:
On Sep 2, 8:39 pm, brent <buleg...@columbus.rr.com> wrote:
On Sep 1, 1:06 am, Bill Sloman <bill.slo...@ieee.org> wrote:
An infinitely long FIR filter, for one.
Isn't that an oxymoron?
Absolutely. A rectangular pulse response is equally unattainable.
What about this?
in----+-----delay line------------+
| |
| sum -----integrate----out
| |
+-----------(-1)------------+
Its impulse response is a rectangular pulse and it's linear.
John
A rectangular pulse requires infinite bandwidth, hence not obtainable.
One assumes your signal will have finite rise and fall times.
Are we to assume the input pulse is wider than the delay of the delay
line? So the sum output goes from zero to minus in to zero to plus in
to zero. Integration yields a triangle.
Tim Williams
Guest
Fri Sep 03, 2010 12:00 pm
<miso_at_sushi.com> wrote in message
news:a2222499-cd69-45fd-a5a4-f2141f1bbc32_at_z28g2000yqh.googlegroups.com...
Quote:
Its impulse response is a rectangular pulse and it's linear.
^^^^^^^
A rectangular pulse requires infinite bandwidth, hence not obtainable.
One assumes your signal will have finite rise and fall times.
Are we to assume the input pulse is wider than the delay of the delay
line? So the sum output goes from zero to minus in to zero to plus in
to zero. Integration yields a triangle.
Read again...
Likewise, its frequency response has 1/4-wave nulls, since its frequency
response is sinc. Hmm, remove the integrator and you remove the 1/x
dependency; it becomes a proper comb filter. They never mentioned that in
class.
Tim
--
Deep Friar: a very philosophical monk.
Website:
http://webpages.charter.net/dawill/tmoranwms
John Larkin
Guest
Fri Sep 03, 2010 1:08 pm
On Fri, 3 Sep 2010 00:49:39 -0700 (PDT), "miso_at_sushi.com"
<miso_at_sushi.com> wrote:
Quote:
On Sep 2, 6:59 am, John Larkin
jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
On Wed, 1 Sep 2010 23:20:38 -0700 (PDT), "m...@sushi.com"
m...@sushi.com> wrote:
On Sep 1, 7:44 pm, John Larkin
jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
On Wed, 1 Sep 2010 18:04:58 -0700 (PDT), "m...@sushi.com"
m...@sushi.com> wrote:
On Sep 1, 4:15 am, John Larkin
jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
On Tue, 31 Aug 2010 21:44:07 -0700 (PDT), "m...@sushi.com"
m...@sushi.com> wrote:
On Aug 31, 9:29 pm, John Larkin
jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
I'm expecting to get some photodiode pulses that are just a bit too
fast to handle with cheapish amps and comparators and such. It would
be nice to have an analog filter that would accept a roughly gaussian
pulse, maybe 2 ns wide, and stretch it to, say, 5 or 6 ns wide,
substantially flat on top if possible. Rep-rate might go up to 40 MHz
maybe.
An LC phase-linear lowpass filter with a reasonable number of poles
would make a slower sorta gaussian blip, not very flat, with a
substantial tail, which would limit my rep-rate to some extent.
If I run the pulse through a tapped analog delay line, maybe five 1 ns
taps, and sum the signals that appears at each tap, I can get a pretty
flat pulse. That amounts to a FIR/transversal filter with all
coefficients = 1, tweakable a little maybe. That's OK if I can get and
afford such a delay line and can sum the tap signals without great
hassles.
We were playing around with using a 3 or maybe 5 pole LC lowpass
filter, but summing the signals from intermediate nodes, instead of
just taking the last one. This looks promising but mathematically
messy to do really well, a "lost in space" situation maybe. A filter
that makes a beautiful output pulse can have some ghastly intermediate
waveforms.
Any ideas? What sort of filter has a rectangular-pulse impulse
response?
John
Why wouldn't you just stretch the pulses with logic circuits? Your
delay element would be a string of inverters.
It has to be analog and linear. Downstream will be amplifiers and
comparators, as noted.
Another possibility is two cascaded Bessel filters. The first shapes
the 2 ns pulse into, say, a 6 ns gaussian pulse, and the second pretty
much just delays that. The sum of the filter outputs will be pretty
much flat and will settle out fast.
John
John
I claim pulse stretching is not a linear operation.
What I want to do certainly is. A passive LC filter is linear. The
output is strictly proportional to the input.
John
Then you are not stretching the pulse. Think of it this way. The input
pulse is of amplitude A and width W. The output pulse is of amplitude
A and width Y, where Y>W. The energy of the output is greater than
that of the input, hence my reference to Parseval.
Why would the output amplitude be unchanged? That's your assumption,
not mine. COE is meaningless here anyhow, since the network, even if
passive, even if the output amplitude is unchanged, could impedance
scale. Voltage is not energy.
Oh, eh yeah, QED. ;-)
By your reasoning, any lowpass filter (which my stretcher is) is
nonlinear. And a coaxial cable must be nonlinear. Only a Brickenbox
network is linear.
I guess you have your own working definition of "nonlinear", which
isn't the one they taught us in engineering school.
We thought a network was linear if the output scales exactly with the
amplitude of the input.
John
Uh, energy needs to be conserved. Your stetcher is so vaguely defined,
it is hard to judge. Now if you say the amplitude is lower, then so be
it. It would have helped if you thought of that sooner.
When the hell did I say lowpass filters are not linear? Uh, nowhere,
thank you very much.
A lowpass filter stretches pulses. You said...
Quote:
I claim pulse stretching is not a linear operation.
You're quite welcome.
Some pulse stretchers certainly are nonlinear, but I need one that is
linear. The optical pulses I want to measure (frequency, amplitide,
noise) are picoseconds wide, so the signal I get is already just the
impulse response of a photodiode. It's already stretched, but
stretched ugly: the waveform is a 2 ns pointy spike. I'm looking for a
network to re-stretch it into a slower flat pulse. A linear network
whose impulse response is a rectangular pulse is ideal for my
purposes.
The more general problem is:
Given a desired waveform Z(t) and an available waveform V(T), what is
the transfer function of a network N that makes V into Z?
More officially,
V(t) * N(t) = Z(t) where * is convolution
and we have V and Z but want to solve for N.
This is "the deconvolution problem", one of the family of
mathematically "ill-posed problems." Lots of papers have been spawned.
The next nasty step, once you've somehow found transfer function N, is
to design an affordable circuit that does it.
Of course, we're spending too much time on this part of the design,
because it's interesting. The other interesting problem is "how do you
accurately measure the frequency of an oscillator that only oscillates
in short bursts now and then?
John
John Larkin
Guest
Fri Sep 03, 2010 1:20 pm
On Fri, 3 Sep 2010 00:57:20 -0700 (PDT), "miso_at_sushi.com"
<miso_at_sushi.com> wrote:
Quote:
On Sep 2, 9:07 am, John Larkin
jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
On Thu, 2 Sep 2010 06:53:29 -0700 (PDT), Bill Sloman
bill.slo...@ieee.org> wrote:
On Sep 2, 8:39 pm, brent <buleg...@columbus.rr.com> wrote:
On Sep 1, 1:06 am, Bill Sloman <bill.slo...@ieee.org> wrote:
An infinitely long FIR filter, for one.
Isn't that an oxymoron?
Absolutely. A rectangular pulse response is equally unattainable.
What about this?
in----+-----delay line------------+
| |
| sum -----integrate----out
| |
+-----------(-1)------------+
Its impulse response is a rectangular pulse and it's linear.
John
A rectangular pulse requires infinite bandwidth, hence not obtainable.
All circuit theory deals with unobtainables: ideal components, zero
propagation delays, noiseless signals. Otherwise you couldn't get
anything done.
Quote:
One assumes your signal will have finite rise and fall times.
An impulse, from which the term "network impulse response" derives,
has zero width. And an ideal delay line and an ideal integrator each
have infinite bandwidth. So the thing above has a rectangular pulse
response to a unit impulse input. A negative rectangular pulse, since
I got the sign backwards.
Quote:
Are we to assume the input pulse is wider than the delay of the delay
line? So the sum output goes from zero to minus in to zero to plus in
to zero. Integration yields a triangle.
No, what I'm trying to do is stretch a skinny spike into a longer flat
pulse, but preserve the amplitude information. The length of the delay
line determines the output pulse width. In my case, the input is about
2 ns wide and I'd like, say, a 6 ns wide pulse, because that would be
easier to process accurately.
My real-world 2 ns spike, passed through the network above, will make
a rectangular pulse with finite rise and fall times but flat on top,
still fine for my purposes.
John
John Larkin
Guest
Fri Sep 03, 2010 1:24 pm
On Fri, 3 Sep 2010 06:00:42 -0500, "Tim Williams"
<tmoranwms_at_charter.net> wrote:
Quote:
miso_at_sushi.com> wrote in message
news:a2222499-cd69-45fd-a5a4-f2141f1bbc32_at_z28g2000yqh.googlegroups.com...
Its impulse response is a rectangular pulse and it's linear.
^^^^^^^
A rectangular pulse requires infinite bandwidth, hence not obtainable.
One assumes your signal will have finite rise and fall times.
Are we to assume the input pulse is wider than the delay of the delay
line? So the sum output goes from zero to minus in to zero to plus in
to zero. Integration yields a triangle.
Read again...
Likewise, its frequency response has 1/4-wave nulls, since its frequency
response is sinc. Hmm, remove the integrator and you remove the 1/x
dependency; it becomes a proper comb filter. They never mentioned that in
class.
Didn't the original PAL color TVs use a delay line for just that
reason, to slice out the interleaved color subcarrier?
John
Tim Williams
Guest
Fri Sep 03, 2010 2:09 pm
"John Larkin" <jjlarkin_at_highNOTlandTHIStechnologyPART.com> wrote in
message news:j2o186t22je083oo3k2l3vr6l4b66ksoje_at_4ax.com...
Quote:
The other interesting problem is "how do you
accurately measure the frequency of an oscillator that only oscillates
in short bursts now and then?
The answer is "duh, you can't", courtesy of Heisenberg.
If you want to do otherwise, you need Star Trek's Heisenberg Compensator
;-)
If the phase is consistent, you can scope it. But you need a sample rate
/ bandwidth high enough, which is technically difficult (GSa / bits /
price). Compare-trigger-and-integrate methods are plausible but accuracy
suffers similarly (jitter, threshold, response), and doesn't integrate
data like a good signal processor, at least without fancy work.
Downconversion won't work very well if the difference is less than a
cycle, i.e., the pulse looks like DC instead of cycles. Indeed, the range
of frequencies for which it looks like DC corresponds to the uncertainty
principle all the same. Yes, you can always run a curve fitting on the
resulting "secant" of a sine wave, but that similarly becomes ill-formed.
But that's not the end of it. Assuming the oscillator has a rectangular
window (or any other type), you can look at the frequency spectrum of the
signal and determine the window type and center frequency. Data is data,
so if you've got everything, you know everything about it, regardless of
the domain you're viewing it in.
Tim
--
Deep Friar: a very philosophical monk.
Website:
http://webpages.charter.net/dawill/tmoranwms
Tim Williams
Guest
Fri Sep 03, 2010 2:11 pm
As I recall, the delay was an entire line (~60us, whatever PAL is
exactly), so they mix the color information from the previous line with
the current line, which alternates or something, to cure the notorious
color drift of NTSC (never twice the same color).
Tim
--
Deep Friar: a very philosophical monk.
Website:
http://webpages.charter.net/dawill/tmoranwms
"John Larkin" <jjlarkin_at_highNOTlandTHIStechnologyPART.com> wrote in
message news:a4q186t8nr238ilo6f5d1hdr15au8b5n12_at_4ax.com...
Quote:
On Fri, 3 Sep 2010 06:00:42 -0500, "Tim Williams"
tmoranwms_at_charter.net> wrote:
miso_at_sushi.com> wrote in message
news:a2222499-cd69-45fd-a5a4-f2141f1bbc32_at_z28g2000yqh.googlegroups.com...
Its impulse response is a rectangular pulse and it's linear.
^^^^^^^
A rectangular pulse requires infinite bandwidth, hence not obtainable.
One assumes your signal will have finite rise and fall times.
Are we to assume the input pulse is wider than the delay of the delay
line? So the sum output goes from zero to minus in to zero to plus in
to zero. Integration yields a triangle.
Read again...
Likewise, its frequency response has 1/4-wave nulls, since its frequency
response is sinc. Hmm, remove the integrator and you remove the 1/x
dependency; it becomes a proper comb filter. They never mentioned that
in
class.
Didn't the original PAL color TVs use a delay line for just that
reason, to slice out the interleaved color subcarrier?
John
Goto page Previous 1, 2, 3, 4, 5, 6, 7 Next
