Detecting tiny pulses after a very large one

[Jeroen Belleman]

I'm trying to find a way to detect tiny pulses following a very
large one. I have this beam current transformer sitting in a
particle accelerator, delivering 4ns, 600V pulses in response to
the passage of the main bunch of particles. This bunch fills
one of a continuous sequence of 'RF buckets', while the others
should be empty. In practice, a tiny bit of beam, on the order of
1e-5 times the main beam, leaks into adjacent buckets, and this
bothers the LHC.

If I attenuate down far enough to protect the digitizer's input,
there is no hope of seeing any of this tiny spill, so I must
clip the main pulse and spare the small stuff. The RF buckets
are at 80 MHz, so the clipper must recover fast. To preserve
the 3GHz bandwidth of the signal, it must be a low capacitance
device too. Small enough to hide it by necking down a 50 Ohm
stripline, for example.

I've dabbled a bit with various combinations of attenuators
and Schottky or ESD protection diodes in Spice, and it doesn't
look straight-forward. I'd be abusing the diodes badly, far
exceeding their maximum current. Beefier diodes are slow and
have too much capacitance.

Anyone here wants to share some wisdom?

Thanks,
Jeroen Belleman

 

[Phil Hobbs]

How about a Nuvistor? Sounds like a job for a tube, for sure.

One other possibility would be to use Schottky diodes to shunt the big
pulse, and a second identical winding wired the other way round, to
cancel the circulating current in the first one and so turn off the
Schottky in time for the little pulse.

You'd obviously need to delay the signal from the second winding, e.g.
with a bit of coax.

Cheers

Phil Hobbs


--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510 USA
+1 845 480 2058

hobbs at electrooptical dot net
http://electrooptical.net

 

[Jeroen Belleman]

How about a Nuvistor? Sounds like a job for a tube, for sure.

Haha! Here I was, looking for depletion mode MOSFETs to see if
they could be of any use here... Thanks for the hint. I'll
check it out.

The beam transformer is a 'wall current monitor'. It doesn't
really have anything that looks like a winding.

Cheers,
Jeroen Belleman

 

[George Herold]

I'm trying to find a way to detect tiny pulses following a very
large one. I have this beam current transformer sitting in a
particle accelerator, delivering 4ns, 600V pulses in response to
the passage of the main bunch of particles. This bunch fill
one of a continuous sequence of 'RF buckets', while the others
should be empty. In practice, a tiny bit of beam, on the order o
1e-5 times the main beam, leaks into adjacent buckets, and this
bothers the LHC.

If I attenuate down far enough to protect the digitizer's input,
there is no hope of seeing any of this tiny spill, so I must
clip the main pulse and spare the small stuff. The RF buckets
are at 80 MHz, so the clipper must recover fast. To preserve
the 3GHz bandwidth of the signal, it must be a low capacitance
device too. Small enough to hide it by necking down a 50 Ohm
stripline, for example.

I've dabbled a bit with various combinations of attenuators
and Schottky or ESD protection diodes in Spice, and it doesn't
look straight-forward. I'd be abusing the diodes badly, far
exceeding their maximum current. Beefier diodes are slow and
have too much capacitance.

Anyone here wants to share some wisdom?

No wisdom, just 'silly' ideas. If you can't shunt the big pulse from the low level input, can you turn-on a low level channel after the big pulse goes by... (Thinking of a box-car averager but maybe without the average function.)
I guess you'd need a second detection path.

George H.

 

[[email protected]]

https://www.macomtech.com/pin-limiter-diodes

Mark

 

[Jeroen Belleman]

https://www.macomtech.com/pin-limiter-diodes

Mmmh. I tried a Mini-Circuits VLM-33-S+ limiter, which
I presume, uses this sort of diodes. It should be fine to
protect a receiver front-end from gross RF overload, but
for my purpose, it was awful, useless.

That is not to say that some other circuit incorporating
these diodes cannot be made to work. I have to look into
that too. I looked at Schottky diodes first because those
recover so much faster.

Jeroen Belleman

 

[George Herold]

On 10/16/2013 8:51 AM, Jeroen Belleman wrote:


Too bad. That two-winding hack would have been slick.

Hmm well I wonder if you could make a transformer such that it would saturate with the big pulse... but let the little guys through... Maybe if Jeroen can knock the 600V down by a factor of 10 then some diodes can do the rest.
(I have no idea if ferrites work at 3 GHz.)

George H.

 

[Joerg]

I'm trying to find a way to detect tiny pulses following a very
large one. I have this beam current transformer sitting in a
particle accelerator, delivering 4ns, 600V pulses in response to
the passage of the main bunch of particles. This bunch fills
one of a continuous sequence of 'RF buckets', while the others
should be empty. In practice, a tiny bit of beam, on the order of
1e-5 times the main beam, leaks into adjacent buckets, and this
bothers the LHC.

If I attenuate down far enough to protect the digitizer's input,
there is no hope of seeing any of this tiny spill, so I must
clip the main pulse and spare the small stuff. The RF buckets
are at 80 MHz, so the clipper must recover fast. To preserve
the 3GHz bandwidth of the signal, it must be a low capacitance
device too. Small enough to hide it by necking down a 50 Ohm
stripline, for example.

Ok, looks like the amp must be 3GHz in BW but "80MHz buckets" seems to
indicate about 12nsec between them which is large.

I've dabbled a bit with various combinations of attenuators
and Schottky or ESD protection diodes in Spice, and it doesn't
look straight-forward. I'd be abusing the diodes badly, far
exceeding their maximum current. Beefier diodes are slow and
have too much capacitance.

Anyone here wants to share some wisdom?

Phil's nuvistors are a good idea since tubes can take a huge punch
without penalties but they are only available as (mostly Russian) cold
war era NOS. Also, I doubt they can do 3GHz but I may be wrong, maybe
there are some. If you need only a one-off solution that doesn't have to
live for half a century and you can find nuvistors that run in the
gigeehoitzes it may be an option to buy some through an auction site.
They often come in cartons of 10 or 20. The wire/pin inductance will
already present a problem in a GHz application.

In ultrasound we have the same problem at a lower frequency range but it
should scale. We have large pulses of more than 100V and right
afterwards must acquire uV-level signals. Way I usually did that:

1. T/R switch. Essentially two diodes back to back in series with a
small current through them. So small that Trr didn't matter but one
could also use Schottky here. We couldn't because of the high pulse
amplitudes. This makes sure that the amp doesn't fry. Possibly your main
pulse energy is low enough that you don't need it.

2. An amplifier that is immune to any lingering saturation effect. My
favorites are grounded gate FET stages. A professor at our university
said this technique is stupid. He had no clue.

Another option may be opamps with GBW over 15GHz that have a very
defined saturation behavior. Often you can only find out by trying
because the datasheet is silent about it.

 

[Joerg]

On Wed, 16 Oct 2013 11:36:52 +0200, Jeroen Belleman wrote:

[...]


I think people are misreading your "RF bucket" terminology -- I'm taking
what you say to mean that the desire is to have a pulse that's 4ns long,
centered in a 'bucket' that's 12.5ns long, with tails that die off
quickly enough that the bleed into adjacent 'buckets' is less than 1e-5
times the pulse height?

Do you know in advance which bucket you expect the pulse to be due? Do
you need to see the pulse at all in the "desired" bucket? Can you
actively blank? If so, a T/R switch, timed for that bucket, may work.

Active T/R switches can have their issues because they can generate
spurious little pulses on their own. Things that aren't really there.

Or -- could you bias a diode such that it normally has current flow, but
your large pulse reverse-biases it? ...

That's ye olde passive T/R switch. Woiks, BTDT.

... That, possibly followed by more
mundane protection, at least has the potential to be pretty fast. DC
bias may be a problem, but could be recovered if you're got a portion of
the beam that's otherwise known to be at zero current.

One trick with T/R switches is to use two diodes in series but opposite,
ideally on one die or at least in one package and from the same wafer.
Can be cathode to cather or anode to anode, whatever is on sale. Make
sure the same DC current flows through both. Then the DC almost goes away.

 

[whit3rd]

I'm trying to find a way to detect tiny pulses following a very
large one. I have this beam current transformer sitting in a
particle accelerator, delivering 4ns, 600V pulses...

Have you considered a delay-line amplifier? The general idea is
to feed V(t) and A* V(t-T_0) into a difference amplifier,
and hope the tail of V(t) can be cancelled and you get
a good measure of the excess (at least, for the
short time between T_0 and 2* T_0). This is one
of the classic ways to prevent pulse pile-up.

 

[Robert Baer]

I'm trying to find a way to detect tiny pulses following a very
large one. I have this beam current transformer sitting in a
particle accelerator, delivering 4ns, 600V pulses in response to
the passage of the main bunch of particles. This bunch fills
one of a continuous sequence of 'RF buckets', while the others
should be empty. In practice, a tiny bit of beam, on the order of
1e-5 times the main beam, leaks into adjacent buckets, and this
bothers the LHC.

If I attenuate down far enough to protect the digitizer's input,
there is no hope of seeing any of this tiny spill, so I must
clip the main pulse and spare the small stuff. The RF buckets
are at 80 MHz, so the clipper must recover fast. To preserve
the 3GHz bandwidth of the signal, it must be a low capacitance
device too. Small enough to hide it by necking down a 50 Ohm
stripline, for example.

I've dabbled a bit with various combinations of attenuators
and Schottky or ESD protection diodes in Spice, and it doesn't
look straight-forward. I'd be abusing the diodes badly, far
exceeding their maximum current. Beefier diodes are slow and
have too much capacitance.

Anyone here wants to share some wisdom?

Thanks,
Jeroen Belleman

Perhaps use the master drive pulse, inverted, to reasonable cancel
the detected version as step one.
Step two - try driven pass and shunt switching (TR/ATR style like in
radar).
Step three should then have lower level undesirable signals for
switch diodes, etc.

 

[Robert Baer]

Haha! Here I was, looking for depletion mode MOSFETs to see if
they could be of any use here... Thanks for the hint. I'll
check it out.

The beam transformer is a 'wall current monitor'. It doesn't
really have anything that looks like a winding.

Cheers,
Jeroen Belleman

But it does have an output winding (input "winding" being the beam)
at present. Add one or two more for pulse cancelling.

 

[Jeroen Belleman]

I'm speaking with more and more ignorance as I get further down in my
reply -- size your grains of salt accordingly.

How many bits on the digitizer? 8? 6? Knowing that helps us know how
much dynamic range you need on the protection circuit.

If it was 16 bits you could almost do it just by digitizing what you
see. Surely the catalogs are just bursting with 16-bit, 3GHz ADCs these
days!

I think people are misreading your "RF bucket" terminology -- I'm taking
what you say to mean that the desire is to have a pulse that's 4ns long,
centered in a 'bucket' that's 12.5ns long, with tails that die off
quickly enough that the bleed into adjacent 'buckets' is less than 1e-5
times the pulse height?

Do you know in advance which bucket you expect the pulse to be due? Do
you need to see the pulse at all in the "desired" bucket? Can you
actively blank? If so, a T/R switch, timed for that bucket, may work.

Or -- could you bias a diode such that it normally has current flow, but
your large pulse reverse-biases it? That, possibly followed by more
mundane protection, at least has the potential to be pretty fast. DC
bias may be a problem, but could be recovered if you're got a portion of
the beam that's otherwise known to be at zero current.

The digitizer would be one of these super-fast sampling Guzik boxes,
so expect 6 effective bits or so.

While it's certainly possible to do timed switching, I'd rather not
go that way, because it complicates things considerably.

Your last suggestion, and Joerg's T/R switch are a good suggestion,
I think. What about the following:

First attenuate the signal down to a more manageable 50Vp or so.
Then use a diode *bridge* biased to some mA using a pair of
50 Ohm resistors to appropriate voltages at the top and bottom.
Apply signal to one side, recover clipped signal at the other
side. Then amplify again to bring out the spilled pulses. As a
bonus, this should be pretty much non-reflective, which
simplifies the interpretation of the output considerably.

Let's see if I can make that work...

Jeroen Belleman

 

[Chris Jones]

Returning to my LOG amplifier idea... in my Garmin GPS chip designs I
used basically a string of PECL stages with a DC loop around the whole
deal to keep all the stages in the linear region, maintaining data and
getting the RSSI function almost for free.

I would suppose your pulses are unipolar, which presents difficulty
with just dropping into my kind of scheme... unless your pulses are
bursts with lots of off-time between bursts, then you could use a very
long time constant for the DC loop.

...Jim Thompson

You can buy log strips from Analog Devices with about the right sort of
bandwidth.
http://www.analog.com/static/imported-files/data_sheets/ADL5513.pdf

I think some other people make them too. (LT?)

Chris

 

[Jeroen Belleman]

You can buy log strips from Analog Devices with about the right sort of
bandwidth.
http://www.analog.com/static/imported-files/data_sheets/ADL5513.pdf

I think some other people make them too. (LT?)

Chris

The target is to deliver a signal from which the amount of spilled
beam in each supposedly empty bucket can be measured. So, while signal
compression would be useful, I don't think a detecting log amp is
the right sort of device here.

Jeroen Belleman

 

[Joerg]

The digitizer would be one of these super-fast sampling Guzik boxes,
so expect 6 effective bits or so.

While it's certainly possible to do timed switching, I'd rather not
go that way, because it complicates things considerably.

Your last suggestion, and Joerg's T/R switch are a good suggestion,
I think. What about the following:

First attenuate the signal down to a more manageable 50Vp or so.
Then use a diode *bridge* biased to some mA using a pair of
50 Ohm resistors to appropriate voltages at the top and bottom.
Apply signal to one side, recover clipped signal at the other
side. Then amplify again to bring out the spilled pulses. As a
bonus, this should be pretty much non-reflective, which
simplifies the interpretation of the output considerably.

Let's see if I can make that work...

Should work. A bridge is even better for T/R switching than a diode
pair. I guess the extra 50 cents or so won't hurt on your project but on
mine with 128 channels and non-government budgets it does.

However, make sure that your amplifier shows next to nothing in recovery
time because it will still rail hard. Just like they do with ultrasound
machines.

 

[Wond]

Hmm well I wonder if you could make a transformer such that it would
saturate with the big pulse... but let the little guys through... Maybe
if Jeroen can knock the 600V down by a factor of 10 then some diodes can
do the rest. (I have no idea if ferrites work at 3 GHz.)

George H.

Ferrite circulators are used in RADAR front-ends, similar problem.
The difficulty here is both signals have the same source, I think?

 

[Joerg]

The target is to deliver a signal from which the amount of spilled
beam in each supposedly empty bucket can be measured. So, while signal
compression would be useful, I don't think a detecting log amp is
the right sort of device here.

Actually, this does work. We used do it in ultrasound on the detection
board in analog systems (now almost extinct). There, we have to be able
to detect 20-30dB of blood flow information that is, in the time domain,
right next to huge unwanted wall echoes that are 40dB above the maximum
flow signal (in heavy smokers, for example).

 

[[email protected]]

The target is to deliver a signal from which the amount of spilled

beam in each supposedly empty bucket ...

If you're going to use a bunch of brain dead metaphors like "spill" and "bucket" in reference to electronic design then you get what you deserve: dumb ideas.

 

[Joerg]

* ---^--- Decent idea; is there any experience as to the recovery time?

It's fast, but AFAICT there ain't no ferrites that would even work at 3GHz.

[...]