Fast impulse from fast-rise-time square wave (microwave)

[Marco]

Marco, what software are you using for your simulation shown in the schematic and graphing output?

The demo version of Simetrix, it's limited in nodes but the GUI is the best I've found. It's also very resilient against convergence problems relative to other Spice simulators I tried.

[RoGeorge]

SimSmith might be a free alternative. I think is based on SPICE, same as LTspice, Pspice, etc.
http://www.eas.uccs.edu/~mwickert/ece5250/notes/ADS_session2.pdf

In the first 5 minutes there is an intro about what can, and what can not do, by W0QE:
 

[Marco]

Linear and frequency domain simulation are the exact opposite of what he needs.

He needs an open source equivalent to Agilent ADS (Doesn't exist.)

[ezalys]

I have access to ADS and microwave office. What I'd prefer to do is learn to design this with pen and paper and maybe do some final optimization in mwo.

[rfeecs]

This is going off in a lot of crazy tangents.  In reality, you don't worry about the line impedance so much.  You don't have nice impedances on the different ports.  You may be driving diodes, etc.

People figured out samplers using diodes long ago, as in the sampling scopes of the 1960s.  You can look at the manuals for Tektronix samplers like the S4 and S6 to see how they did it.

The S6 used the short circuit transmission lines.   They referred to them as "clip lines".

The S4 used a "traveling wave" sampling gate, patented by Tektronix:
https://patentimages.storage.googleapis.com/52/97/2c/b6fbddbb95dcb8/US4647795.pdf

The gate is generated using just one edge of the strobe line and is set by the length of a transmission line.  Here's a very brief explanation:
http://w140.com/frye_s4_gate.pdf

[ezalys]

Yeah I'm kinda disappointed at how off topic people are being. You can't source cheap tunnel diodes or high frequency SRDs anymore in low volume and NLTLs and all that are very expensive and tricky and what the hell does chirping have to do with anything? I just want a fast pulse from a fast transition and to understand how such a circuit functions.

But okay. Yeah. The S6 and S4 are nice designs that I've looked into before. The S4 seemed a bit complex but the S6 seemed pretty understandable. I like the clip-line idea a lot. I imagine I could prototype such a thing on a PCB using little pieces of coax. 100 ps seems really short... The scheme reminds me of finite differencing. You can choose a shorter dt (shorter clip-line) but then you sum with a less time-shifted version of the same signal, meaning if the rise time is a bit slow relative to l/c, you really kill the pulse amplitude. Is this intuition correct?

[rfeecs]

But okay. Yeah. The S6 and S4 are nice designs that I've looked into before. The S4 seemed a bit complex but the S6 seemed pretty understandable. I like the clip-line idea a lot. I imagine I could prototype such a thing on a PCB using little pieces of coax. 100 ps seems really short... The scheme reminds me of finite differencing. You can choose a shorter dt (shorter clip-line) but then you sum with a less time-shifted version of the same signal, meaning if the rise time is a bit slow relative to l/c, you really kill the pulse amplitude. Is this intuition correct?

Sounds about right.  But ideally, if your rise time is much shorter than the clip line delay, you will get the full amplitude and the pulse length will mostly be determined by (2x) the line delay time.

[Marco]

This is going off in a lot of crazy tangents.  In reality, you don't worry about the line impedance so much.  You don't have nice impedances on the different ports.  You may be driving diodes, etc.

Whose reflections on the transmission line feeding them will ring with whatever you use to differentiate the pulse. You can't keep everything compact enough at these frequencies to avoid it, everything is a transmission line.

People figured out samplers using diodes long ago, as in the sampling scopes of the 1960s.  You can look at the manuals for Tektronix samplers like the S4 and S6 to see how they did it.

With a discrete design with modern components the design space is at the same time not what it was in the 60s, nor what it was for Tektronix once they started wire bonding hybrids and using MMICs. With the hybdrid in the SD-xx they could just scale their old designs, but even with 0201s that's unlikely to be optimal.

[ezalys]

Okay that seems pretty reasonable. The onet8501p has a 35 ps max rise time so a 100-200 ps stub seems pretty reasonable.

Also, I'm trying to scale in cost, not in frequency.

[Marco]

Supposedly the designer of the early Tek sampling heads did a 20 GHz sampler on FR4 with discretes well over 10 years ago. So in my opinion that's what you should shoot for

[awallin]

seems radio-astronomers doing VLBI do what they call "phase-cal" with a pulse train that has harmonics up into the multi-GHz range.

they make a square-wave with sharp edges, use a capacitor to differentiate the square wave, and then an RF-switch to pass only pulses of the desired polarity, see e.g.
https://www.haystack.mit.edu/geo/vlbi_td/BB/023.pdf

[Gerhard_dk4xp]

Looks like you get unpleasant component values if you want to differentiate in a 50 Ohm system.
This is a 50 Ohm microstrip on 0.8mm FR4, the usual Bungard stuff that you use for self-etching in .de.
First picture is just a scalpel cut through the stripline. That gives a nice peak, but watch the vertical scale.
0.5 pF 0603 has the rising edge quite OK, but the decay is not differential-like, it simply passes the
step for the foreseeable future.

Blue trace: reflection back to the TDR source,
yellow = after 20 cm of coax + cut microstrip into normal 20 GHz sampling channel.

I cannot easily attach a coax line in the middle of the board because there are no vias
to the GND side. At least not today. Bedtime now.

The board was an experiment to find the right eps-r of 0.8mm Bungard-FR4. 56 mil seems
to be the correct microstrip width for 50 Ohms. The circle was intended to be a resonator
that could be seen on the VNA. That was before I had the TDR.

regards, Gerhard

[rhb]

I'd like to note that a coax stub and mercury wetted relay were a common (only?)  method for generating short impulses before semiconductors got fast enough to replace them.

[Marco]

I don't think a volt or so is going to cut it to really kick a Schottky sampling gate into gear. If you are satisfied with single digit GHz you could copy this design using the MSPD2018 Sampling Phase Detector, which is essentially a packaged sampling gate.

[Gerhard_dk4xp]

It certainly works for radars. In the 70s there were radars with 1MW peak and 10kW mean outputs.

AFAIK with radar you never have the pulse-compressed (non-chirped) signal at high power---there's no reason you'd want to.  The chirp is collapsed as part of the receive/imaging path.

But that's the point of the chirp, that you can run the transmitter longer at its abs. max. ratings and then compress the output in time to get a peak power larger than abs. max.
Also, the transmitted signal would not be so visible to a simple diode detector.

cheers,
Gerhard

[Gerhard_dk4xp]

Why does a stub work? How long should the stub be?

Good question, I thought I understood why ...  I only have a very approximate understanding though on reflection The principle is easy, the current pulse splits at the stub, then when the negative reflection comes from the grounded end it cuts the signal to the other branch to zero. So instead of a square wave you get pulses at the edges.

That said, how exactly it works with impedances I have no idea. When I put it in a simulator it tells me the stub impedance should be 25 Ohm to make it work perfectly, with other impedances you get multiple reflections and a screwed up trailing edge. Is the simulator screwing up? I have no idea.

Your simulation was good, including the 25 Ohms.
I used 28mm pieces of unknown 50Ohm cable, probably rg405.
One 50Z-Line produced the messed-up trailing edge, two in parallel gave a nice pulse.

Blue = reflections back into the TDR
Yellow = transmission into load channel.

[Gerhard_dk4xp]

Then I shorted the lines to 9mm. That is no longer pretty. Too much
parasitics in comparison to line length. If you want to go faster, you
will probably need etched microstrip.

cheers, Gerhard

[ezalys]

Gerhard_dk4xp, I'm curious, what scope is this that you're using?

[Gerhard_dk4xp]

HP54750A with 18 GHz differential TDR plug in and a dual 20 GHz ordinary sampling plug in.
(54754A and 54751A)  I also have a 50 GHz plug in, but I avoid using it unless really needed.
It has 2.4mm connectors and is easy to kill.

regards, Gerhard

[Leo Bodnar]

I can see a lot of people are trying to use 50Ω microstrip or coax as a stub.  However, Tek sampling heads use really wide and low impedance stubs - they call them clipping lines.  Here is a snapshot of 40GHz sampling head front end.  I have loosely commented relevant components.  SRD generates the step pulse that gets differentiated and becomes a sampling pulse. Sampling signals are differential (unfortunate collision.)
Differentiating differential pulse signal here you go.
Leo

[Leo Bodnar]

Food for thought - the clipping lines are low impedance probably because SRD is a low impedance pulse source and 50Ω clipping line will not be able to fully suppress the signal. 

Notice that differentiating stubs are located close to the pulse source instead of dragging the pulse halfway across the head first. 

Microstrips delivering the gating pulse to the sampler are quite long - perhaps to buy the sampler some time to finish its sampling business until reflections start coming back.  If the gate has closed by that time then they don't really matter as long as they don't reopen the sampler again.

Leo

[Marco]

However, Tek sampling heads use really wide and low impedance stubs - they call them clipping lines.

Yeah, but their substrates are a lot thinner ... you can go lower than 50 Ohms, but not too much lower with a standard PCB. Flex PCBs are quite reasonably priced though, so that would work.

[Leo Bodnar]

Alumina's E_r is almost 10 so they need to be thin.  This seems to be using quartz which is closer to FR4.  The main reason is probably low loss?

I know you have figured that the stub in the middle of 50R line should be 25R impedance but I will throw in my simulation I did for somebody else anyway.
It uses voltage sources with 50R impedance but it is equivalent to current sources with 50R termination and for this simulation makes no difference.
With 25R clipping line you lose some of the pulse aplitude, of course.
Cheers
Leo

[Gerhard_dk4xp]

I have 2 defective samplers from 83483A / 54751A plugins    ( I made one working
plugin from two semi-defective ones) . Part number is Agilent 5086-7699, available as
a replacement part. Someone is selling on ebay for $120 (working with garanty),
somebody else for $160 (defective) and someone for 530 Pounds ("untested", haha).

This is the type of the samplers in the 20 GHz plugin that I used to make the
measurements above.

Repairing the plugins is easy, you just need a Torx bit and a SMA/PC3.5 wrench.
You can just exchange the SMA bricks (really PC3.5), nothing else.

I decided to open one of them. Pictures are what my Sony xpedia Z5 cell phone can do,
but even under the *50 microscope I cannot see all the details. The substrate seems to be
quartz, it is translucent. The squares on the paper are 5mm.
The cavities were filled with absorber foam to avoid waveguide resonance effects.
I think that is outside of the diy area.

[Gerhard_dk4xp]

similar sampler for a microwave frequency counter:

http://www.hpl.hp.com/hpjournal/pdfs/IssuePDFs/1986-02.pdf

cheers, Gerhard