Fast-assed signal switch

[Tim Williams]

http://webpages.charter.net/dawill/Images/Fast S&H.gif

I think I have everything in there that needs to be...

Basically, step recovery diode (SRD) is forward-biased, and along with it, a
schottky balanced switch thing, which is just an average FWB used different.
When reverse-biased (through two inductors, which are coupled to encourage a
somewhat balanced cutoff edge), the SRD loses charge then whams the
schottkies off. RFCs and such isolate the GHz trash from the relatively low
impedance transistors nearby, improving rise time I would suppose.

Not shown: input buffer, output holding cap, output buffer, etc. I suppose
I should've titled this "Fast Switch.gif", but who cares, I'd use it just
for S&H anyway.

An offset adjust (an intentional offset to one mirror and an adjustable one
on the other) would be helpful, and some snubbers should probably be
provided for the reverse-holding transistors' coupled inductor.

The "SW" input is high = off. It is logic level (with respect to -V); what
voltage depends on the emitter resistor (4.7k specified is for about +/-10V
supplies and drive from, say, an RTL stage running from same). A small
capacitor across the 4.7k might be provided to increase turn-on time. Plain
old 2N4401's should suffice for switching relatively slow SRDs (t_rr ~
100ns, t_snap < 500ps), which should likewise suffice for turning off the
input cleanly for inputs slewing slower, which is like a couple GHz.

Add a triggered delay and all that junk (see: Analog Sampler) and this
oughta extend one's low bandwidth scope into the GHz range..!? Ooh, tasty.

Comments? Stupidities? (I'm sure there are some! ;o) )

Tim

 

[D from BC]

http://webpages.charter.net/dawill/Images/Fast S&H.gif

I think I have everything in there that needs to be...

Basically, step recovery diode (SRD) is forward-biased, and along with it, a
schottky balanced switch thing, which is just an average FWB used different.
When reverse-biased (through two inductors, which are coupled to encourage a
somewhat balanced cutoff edge), the SRD loses charge then whams the
schottkies off. RFCs and such isolate the GHz trash from the relatively low
impedance transistors nearby, improving rise time I would suppose.

Not shown: input buffer, output holding cap, output buffer, etc. I suppose
I should've titled this "Fast Switch.gif", but who cares, I'd use it just
for S&H anyway.

An offset adjust (an intentional offset to one mirror and an adjustable one
on the other) would be helpful, and some snubbers should probably be
provided for the reverse-holding transistors' coupled inductor.

The "SW" input is high = off. It is logic level (with respect to -V); what
voltage depends on the emitter resistor (4.7k specified is for about +/-10V
supplies and drive from, say, an RTL stage running from same). A small
capacitor across the 4.7k might be provided to increase turn-on time. Plain
old 2N4401's should suffice for switching relatively slow SRDs (t_rr ~
100ns, t_snap < 500ps), which should likewise suffice for turning off the
input cleanly for inputs slewing slower, which is like a couple GHz.

Add a triggered delay and all that junk (see: Analog Sampler) and this
oughta extend one's low bandwidth scope into the GHz range..!? Ooh, tasty.

Comments? Stupidities? (I'm sure there are some! ;o) )

Tim

It's a sup'd up 4066??
D from BC

 

[John Larkin]

http://webpages.charter.net/dawill/Images/Fast S&H.gif

I think I have everything in there that needs to be...

Basically, step recovery diode (SRD) is forward-biased, and along with it, a
schottky balanced switch thing, which is just an average FWB used different.
When reverse-biased (through two inductors, which are coupled to encourage a
somewhat balanced cutoff edge), the SRD loses charge then whams the
schottkies off. RFCs and such isolate the GHz trash from the relatively low
impedance transistors nearby, improving rise time I would suppose.

Not shown: input buffer, output holding cap, output buffer, etc. I suppose
I should've titled this "Fast Switch.gif", but who cares, I'd use it just
for S&H anyway.

An offset adjust (an intentional offset to one mirror and an adjustable one
on the other) would be helpful, and some snubbers should probably be
provided for the reverse-holding transistors' coupled inductor.

The "SW" input is high = off. It is logic level (with respect to -V); what
voltage depends on the emitter resistor (4.7k specified is for about +/-10V
supplies and drive from, say, an RTL stage running from same). A small
capacitor across the 4.7k might be provided to increase turn-on time. Plain
old 2N4401's should suffice for switching relatively slow SRDs (t_rr ~
100ns, t_snap < 500ps), which should likewise suffice for turning off the
input cleanly for inputs slewing slower, which is like a couple GHz.

Add a triggered delay and all that junk (see: Analog Sampler) and this
oughta extend one's low bandwidth scope into the GHz range..!? Ooh, tasty.

Comments? Stupidities? (I'm sure there are some! ;o) )

Tim

It's more conventional to keep the diodes off most of the time and
couple SRD-generated impulses into them to turn them on very briefly.
The resulting signal is usually only a few per cent of the input (the
ratio is called "sampling efficiency") so the resulting charge glitch
is gained up, sample/holded (is that a verb?) and fed back to
re-center the diode bias. This gets around the problem that the tau
formed by the diode resistance and the load capacitance can be really
slow.

I'm guessing that any imbalance in your SRD drive would blast through
into the output and swamp the signal.

I have a bunch of HP and Tek sampler manuals; I'll post some
schematics when I have a chance. There are probably manuals on the web
somewhere... HP1810, HP187, Tek 1S1, 1S2, and the S1/S2/S3 sampling
heads. Most of the old manuals had a good technical discussion.

There's also Tek's wonderful paperback "Sampling Oscilloscope
Circuits", 1970, original price $1. They show up on ebay.

John


John

 

[colin]

http://webpages.charter.net/dawill/Images/Fast S&H.gif

I think I have everything in there that needs to be...

Basically, step recovery diode (SRD) is forward-biased, and along with it,
a
schottky balanced switch thing, which is just an average FWB used
different.
When reverse-biased (through two inductors, which are coupled to encourage
a
somewhat balanced cutoff edge), the SRD loses charge then whams the
schottkies off. RFCs and such isolate the GHz trash from the relatively
low
impedance transistors nearby, improving rise time I would suppose.

Not shown: input buffer, output holding cap, output buffer, etc. I
suppose
I should've titled this "Fast Switch.gif", but who cares, I'd use it just
for S&H anyway.

An offset adjust (an intentional offset to one mirror and an adjustable
one
on the other) would be helpful, and some snubbers should probably be
provided for the reverse-holding transistors' coupled inductor.

The "SW" input is high = off. It is logic level (with respect to -V);
what
voltage depends on the emitter resistor (4.7k specified is for about
+/-10V
supplies and drive from, say, an RTL stage running from same). A small
capacitor across the 4.7k might be provided to increase turn-on time.
Plain
old 2N4401's should suffice for switching relatively slow SRDs (t_rr ~
100ns, t_snap < 500ps), which should likewise suffice for turning off the
input cleanly for inputs slewing slower, which is like a couple GHz.

Add a triggered delay and all that junk (see: Analog Sampler) and this
oughta extend one's low bandwidth scope into the GHz range..!? Ooh,
tasty.

Comments? Stupidities? (I'm sure there are some! ;o) )

Ive made this and it works fairly well,
http://www.elecdesign.com/Files/29/4749/Figure_01.gif

I since modified it to be a SRD type design but was with a monolithic device
including the coupling capacitor and schotky diodes.

you can get very fast comparators wich are easier than using discrete
transistors.

as stated in other post, the way SRD are used here is to use the turn off
spike to turn on the schotky diodes for a breif time.
theres no point in having any other mechanism to turn them on in a sample
circuit with using a SRD.

Colin =^.^=

 

[Tim Williams]

Ive made this and it works fairly well,
http://www.elecdesign.com/Files/29/4749/Figure_01.gif

Heh, an analog type D flip-flop. With diodes.

I since modified it to be a SRD type design but was with a monolithic device
including the coupling capacitor and schotky diodes.

you can get very fast comparators wich are easier than using discrete
transistors.

Yeah, but that's cheating. ;-)

as stated in other post, the way SRD are used here is to use the turn off
spike to turn on the schotky diodes for a breif time.

Why? Isn't the instant when the schottkies are turned off the instant when
the signal is disconnected? Therefore, it shouldn't matter how long the
thing is switched on, so long as enough of the input voltage gets through
the diodes (perhaps into a rather small initial holding cap, as the 1pF in
the above circuit). Turn-off should be with as high a dV/dt as possible, to
maximize the input dV/dt that can be handled without slew errors (in an SE
switch, the slew is skewed; in a balanced switch like this, both up and down
slew rate).

The "hump" of an SRD turn-off, coupled appropriately, will turn on, and off,
the schottkies fast enough, but not necessarily in enough time to get
anywhere up the RC time constant of the input, diodes (whatever effective
resistance they have) and holding capacitor. "Sampling efficiency" doesn't
make sense to me: after coupling in a bit of charge, another bump of the
same level will go more in that direction, despite the input being constant.
Therefore, it is approximately an integrator circuit, and therefore needs a
differentiator to have any use as an oscilloscope attachment.

Tim

 

[John Larkin]

Heh, an analog type D flip-flop. With diodes.


Yeah, but that's cheating. ;-)


Why? Isn't the instant when the schottkies are turned off the instant when
the signal is disconnected? Therefore, it shouldn't matter how long the
thing is switched on, so long as enough of the input voltage gets through
the diodes (perhaps into a rather small initial holding cap, as the 1pF in
the above circuit). Turn-off should be with as high a dV/dt as possible, to
maximize the input dV/dt that can be handled without slew errors (in an SE
switch, the slew is skewed; in a balanced switch like this, both up and down
slew rate).

The "hump" of an SRD turn-off, coupled appropriately, will turn on, and off,
the schottkies fast enough, but not necessarily in enough time to get
anywhere up the RC time constant of the input, diodes (whatever effective
resistance they have) and holding capacitor. "Sampling efficiency" doesn't
make sense to me: after coupling in a bit of charge, another bump of the
same level will go more in that direction, despite the input being constant.
Therefore, it is approximately an integrator circuit, and therefore needs a
differentiator to have any use as an oscilloscope attachment.

Tim

Get the Tek book. It explains all of this stuff.

John

 

[Tim Williams]

Get the Tek book. It explains all of this stuff.

Way to dodge the question ...NOT ;-)

Unless you're offering, of course.

Tim

 

[John Larkin]

Way to dodge the question ...NOT ;-)

Unless you're offering, of course.

Tim

Oh all right.

Imagine a 25 ohm source (50 in, terminated), a series switch, and some
small hold capacitance. Say that the switch has another 10 ohms of
series resistance, and the hold cap is 3 pF. If the switch is closed,
the hold cap follows the input with about a 100 ps time constant. So
your off-switch will be speed limited by that tau.

If we blip the switch closed only for, say, 25 ps, the cap samples the
signal for 25 ps but it only charges up to about 25% of the difference
between the pre-sample cap voltage and the new voltage. That's a 25%
sampling efficiency, but we're 4x as fast. All it takes to fix the
efficiency problem is a cheap opamp, to give low-frequency gain. Most
sampling scopes only run a few percent efficient.

And fast samplers usually use only 2 diodes, to reduce the switch
resistance.

This was all worked out in about 1961, when HP discovered the SRD
(known then as the Boff Diode) and invented the dual-diode feedback
sampler, first used in the HP185 scope.

The only two serious advances since then have been the shockline
sampling pulse generator and the 6-diode traveling-wave sampler. The
fastest electrical samplers are now upwards of 250 GHz, using shock
lines and dual diodes. PSPL sells a commercial 100 GHz sampling head,
but not the whole oscilloscope.

John

 

[colin]

Heh, an analog type D flip-flop. With diodes.

thers no D type there that I see,
although I added an ecl one in mine becuase i dont see how that triggers
cleanly.

Yeah, but that's cheating. ;-)

I cheated even more and replaced those transistors in that diagram with the
emiter folowers from the ecl trigger flip flop

Why? Isn't the instant when the schottkies are turned off the instant
when
the signal is disconnected? Therefore, it shouldn't matter how long the
thing is switched on, so long as enough of the input voltage gets through
the diodes (perhaps into a rather small initial holding cap, as the 1pF in
the above circuit). Turn-off should be with as high a dV/dt as possible,
to
maximize the input dV/dt that can be handled without slew errors (in an SE
switch, the slew is skewed; in a balanced switch like this, both up and
down
slew rate).

well theres just no point in having an srd and turning it on as well.
you can get a very fast fall time of the switching waveform without srd.
the voltage over wich the diodes turn from on to off isnt much and you dont
need a fantastically high dv/dt to acheive 250ps switching time.

The "hump" of an SRD turn-off, coupled appropriately, will turn on, and
off,
the schottkies fast enough, but not necessarily in enough time to get
anywhere up the RC time constant of the input, diodes (whatever effective
resistance they have) and holding capacitor. "Sampling efficiency"
doesn't
make sense to me: after coupling in a bit of charge, another bump of the
same level will go more in that direction, despite the input being
constant.

extending the on time has no such effect whatsover becuase you are only
interested in the last 250ps or so anyway.
so if the hump is ~250ps thats about right, whatever state the waveform is
in before then is of no interest to you.
in fact it can cuase problems.

Therefore, it is approximately an integrator circuit, and therefore needs
a
differentiator to have any use as an oscilloscope attachment.

its not realy an integrator, but the two types of circuit behave
differently.
with an srd the voltage on the capacitor remains from one sample to the
next,
so adjacent samples that are similar voltage increase the voltage on the
capacitor.
therefore at slow sweep rates the capacitor charges up closer to what the
input is.

therefore with an srd and capacitor the sweep rate affects the high
frequency roll off.

with the on for longer aproach its harder to recover the sampling loss as
the previous sample voltage is destroyed,
however the frequency response is flatter but ofc lower voltage.

Colin =^.^=

 

[John Larkin]

Here's a dual-channel sampler I did for fun. Only one channel is
implemented. It worked pretty well, 70 ps risetime, 5 GHz bandwidth.
It drives a dual-diode half-bridge with complementary impulses,
starting with a step generated by the srd in the center, then
differentiated by the short, fat, shorted transmission lines.

http://s2.supload.com/free/MVC-286X.JPG/view/

http://s2.supload.com/free/MVC-288X.JPG/view/



John

 

[John Larkin]

What is the piece of white coax doing?

It picks off a little of the input signal for the blowby compensation.
"Blowby" is the small amount of non-sampled signal that gets through
the diode capacitance. I figured the same signal could be used for an
internal trigger, which would be handy for high rep-rate signals. The
blowby amp is to the lower left. I'd use an opamp nowadays.

Where is the sampled output? In my HP8411A, the output comes from the DC
bias resistors to the Schottky diodes, which would be equivalent to the to
fatter wires that go down through the hole closest to the sampling diodes.
Is that what you do on your sampler?

Yup. That pair goes off the the charge amp, slow sample-hold, and bias
feedback stuff, not shown.

BTW I can't help thinking that it might go better if you put in a few more
ground "vias" at the ends of the fat microstrip lines above and below the
SRD. Maybe it wouldn't actually work better but I guess the current pulse
into the top transmission line eventually has to be returned to the bottom
transmission line through the ground plane under the SRD.

The topside flood seems to work. And there's lots of capacitance
between the topside and bottom copper pours. There are 8 vias sort of
generally surrounding the sampler circuit.

The only thing that matters here is the peak of the spike that forward
biases the sampling diodes. Otherwise, it can ring and bounce and
twang all it wants, and it doesn't matter.

John

 

[Chris Jones]

Here's a dual-channel sampler I did for fun. Only one channel is
implemented. It worked pretty well, 70 ps risetime, 5 GHz bandwidth.
It drives a dual-diode half-bridge with complementary impulses,
starting with a step generated by the srd in the center, then
differentiated by the short, fat, shorted transmission lines.

http://s2.supload.com/free/MVC-286X.JPG/view/

http://s2.supload.com/free/MVC-288X.JPG/view/



John

What is the piece of white coax doing?

Where is the sampled output? In my HP8411A, the output comes from the DC
bias resistors to the Schottky diodes, which would be equivalent to the to
fatter wires that go down through the hole closest to the sampling diodes.
Is that what you do on your sampler?

BTW I can't help thinking that it might go better if you put in a few more
ground "vias" at the ends of the fat microstrip lines above and below the
SRD. Maybe it wouldn't actually work better but I guess the current pulse
into the top transmission line eventually has to be returned to the bottom
transmission line through the ground plane under the SRD.

Chris

 

[Chris Jones]

It picks off a little of the input signal for the blowby compensation.
"Blowby" is the small amount of non-sampled signal that gets through
the diode capacitance. I figured the same signal could be used for an
internal trigger, which would be handy for high rep-rate signals. The
blowby amp is to the lower left. I'd use an opamp nowadays.

Thanks - makes sense.

Yup. That pair goes off the the charge amp, slow sample-hold, and bias
feedback stuff, not shown. Ok.


The topside flood seems to work. And there's lots of capacitance
between the topside and bottom copper pours. There are 8 vias sort of
generally surrounding the sampler circuit.

The only thing that matters here is the peak of the spike that forward
biases the sampling diodes. Otherwise, it can ring and bounce and
twang all it wants, and it doesn't matter.

I was just thinking that for maximum bandwidth, the reflection from the
short had better be nice and crisp to avoid smearing out the shape of the
sampling pulse. If it worked well, then that's all that matters.

Chris

 

[Boris Mohar]

Has optical sampling been done? Two photo diodes switched by a short laser
pulse.

 

[Tim Williams]

So, the SRD plinks the short bit of trace (between its caps and the rest of
the ground plane, acting as an inductor), which causes a wave to bounce down
both ends, and I suppose reflecting off the right side, so the total pulse
length is approximately the width of the dual dipole structure? Or is the
pulse the height of the dipole base (from the SRD to where it joins the
ground plane), so the pulse is that short, and the right end is merely
terminated? (I can't get a good look at the right side.) But if it were
terminated, why any right wing at all?...

The schottky diodes are in the SOT-23 I take it, so that when the pulse hits
it, it gets forward biased, grabbing some of the input and dumping it into
1. nearby capacitances and 2. the twisted pair line exiting through the
hole?

So the other end of the output twisted pair contains coupling and bias
which, combined with the 2.7k's, keeps the schottkies biased off?

Where does the DC (from the rectified pulse) go? Seems to me it might build
up, though obviously DC /per se/ bleeds off just fine, being 0Hz. I guess I
should say, charge built up from one cycle.

Tim

 

[John Larkin]

Has optical sampling been done? Two photo diodes switched by a short laser
pulse.

One sampler is a thin film of radiation-damaged GaAs, which forms a
photoconductor with a picosecond-range lifetime. It's been used with a
fs laser to make an optically-gated electrical sampler. One problem is
the cost, and another that the fs lasers sort of fire when they feel
like, and can't be triggered with appropriately low jitter.

Agilent makes an optical signal sampler, the 86119A, that uses some
nonlinear optical effect. It's very expensive but hits 500 GHz.

HP and Tek could certainly make 250 GHz electrical samplers, but seem
to have called a truce at about 70G, probably because signals don't
pass through cables and connectors very well up there.

John

 

[John Larkin]

So, the SRD plinks the short bit of trace (between its caps and the rest of
the ground plane, acting as an inductor), which causes a wave to bounce down
both ends, and I suppose reflecting off the right side, so the total pulse
length is approximately the width of the dual dipole structure? Or is the
pulse the height of the dipole base (from the SRD to where it joins the
ground plane), so the pulse is that short, and the right end is merely
terminated?

Yes, the latter. The very short stubby traces are the clipper lines.
The longer traces to the ends transport the 50-ohm glitches to the
diode pairs.

Imagine that the SRD is a fast-rise current source. The stubby lines
are 25 ohms. The current enters the 3-way junction and the leading
edge goes in all directions. When it hits the end of the 25 ohm
stubby, it reflects back as zero volts, that hits the junction, and
zero volts chases the initial step down the 50 ohm lines. It's all
terminated at the diodes.


(I can't get a good look at the right side.) But if it were

terminated, why any right wing at all?...

That's for the 2nd channel, which I never built.

The schottky diodes are in the SOT-23 I take it, so that when the pulse hits
it, it gets forward biased, grabbing some of the input and dumping it into
1. nearby capacitances and 2. the twisted pair line exiting through the
hole?
Yes.


So the other end of the output twisted pair contains coupling and bias
which, combined with the 2.7k's, keeps the schottkies biased off?

Yup. The diodes are biased off about 2 volts.

Where does the DC (from the rectified pulse) go? Seems to me it might build
up, though obviously DC /per se/ bleeds off just fine, being 0Hz. I guess I
should say, charge built up from one cycle.

A small shot of charge gets to the twisted pair, the amount depending
on the difference between the instantaneous signal level and the
previous diode bias. A charge amplifier senses the step, and it's
amplified and goes into a slow (1 us roughly) s/h and then a gated
integrator. The resulting voltage is fed back to approximately
re-center the diode bias for the next shot. It's a feedback sampler,
invented by HP in about 1961 or so. The diodes are sort of a sampled
error detector.

John

 

[Tim Williams]

Alright, excellent. Then, to apply this to things I have on hand, I
might... hmm, I probably have some signal schottkies. I have some RF
circuits, one board here has four glass body diodes arranged in a ring
modulator (inbetween two ferrite transformers- could it be any more
obvious?). I can make out "HP2", and maybe "305" on the next line. HP2305,
any ideas?

Anyway, for pulse generator, I don't recall having any diodes with
reasonable snap, so I might try an avalanche generator. 2SA1206 or
something, I discovered, works at bench supply voltages, while I have some
PH2369's that plink from +100V, as Jim uses in AN47. So I could set up one
of these on, say, 6" of shorted coax, no, twisted pair -- so that when this
thing fires, it runs a good voltage for a moment, then reflects back and
shorts out. In the mean time, some schottkies or whatever are biased by
said pulse and pick up some input from across the input cable, passing it to
something, which I suppose has to be a charge amp, to restore the frequency
response.

Ironically, looking at the finer details of an avalanche generator is one of
the goals of my wanting to look at hell ass fast signals. Fortunately I
have more than two 2369's...

Tim

 

[colin]

One sampler is a thin film of radiation-damaged GaAs, which forms a
photoconductor with a picosecond-range lifetime. It's been used with a
fs laser to make an optically-gated electrical sampler. One problem is
the cost, and another that the fs lasers sort of fire when they feel
like, and can't be triggered with appropriately low jitter.

Agilent makes an optical signal sampler, the 86119A, that uses some
nonlinear optical effect. It's very expensive but hits 500 GHz.

HP and Tek could certainly make 250 GHz electrical samplers, but seem
to have called a truce at about 70G, probably because signals don't
pass through cables and connectors very well up there.

you could let the laser fire at random,
then you measure the time it fires from you trigger point
then plot the sample at the right place.

Colin =^.^=

 

[Phil Hobbs]

It's true that lasers in general, and especially solid state lasers like
Ti:Sapphire, aren't triggerable any time you happen to want. Their
jitter, though, is lower than any other known system whatever--below 1
part in 10**20, iirc. You can make them wider than an cctave, so that
you can harmonically lock one end of the frequency comb to the second
harmonic of the other end, and make something like an old-time ham
marker frequency generator, producing peaks at 100 MHz intervals across
the whole visible and near-infrared spectrum, whose frequency is known
to an accuracy of 10**-18 or better. The key to the accuracy is that
you aren't multiplying up the 100 MHz by 4,000,000 times--you're doing
the mixing at 1x, which gets you a cool 126 dB improvement, and the
frequency accuracy of each comb peak (in Hz) is the same as the rep
rate's.

This got Jan Hall and Ted Haensch the Nobel Prize in 2005--more for the
optics than the electronics, because they figured out how to do the
spectral broadening without messing up the time coherence. The nice
thing is that once they figured it out, it's really easy to do.

There are also streak cameras, which can take data continuously, not
stroboscopically like a sampling scope, and can display sensibly at 1
ps/division. Getting the electrical signal onto the optical one is
usually the tough part, but I have an electrooptical modulator in my lab
that is flat to 1 dB out to 30 GHz, and the technology could go further.
Full scale is about +20 dBm.

you could let the laser fire at random,
then you measure the time it fires from you trigger point
then plot the sample at the right place.

If you get good enough coverage this way, it could work. Alternatively,
with a fs sampler, you could use a variable optical delay, but that
would probably have to be mechanically scanned, which would make the
update rate slowish.


Cheers,

Phil Hobbs