CORRECTION: It's Breitbarth. His story is on LinkedIn.
I bet it could be done, but the physical implementation would be a challenge because it does clever things with transmission lines. But it would be interesting to do it with modern multilayer circuit boards.
See the US patents of "Agoston Agoston" initially at Tektronix.
Joe
I meant make a small box that actual SD heads would plug into.
Tek actually had one, at a trade show, but they never sold it.
I corresponded with him once, after he started Hyperlabs. I sent him my deconvolution program but I don't think he used it.
One of his patents was for a sampler that used a slotline machined into a metal block. But the SD heads were hybrids.
John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics
They work pretty much the same at the end of a 2-meter multistrand cable, so the mainframe side can’t be that tweaky.
Cheers
Phil Hobbs
It would need the weird connectors, which might be the worst part.
And a trigger/timebase, which wouldn't be too awful.
A Pi Pico could be the CPU.
John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics
Dogs expect you to feed them. Women tend to think that they can manipulate you, which comes to much the same thing. You do come across as a gullible sucker, so the result isn't unexpected.
Yikes.
John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics
Maybe too expensive to make to be sellable.
I do recall those pictures.
So make the whole item; no plug-in?
Joe
Don't we talk of finding weird but just-right connectors a lot?
Joe
There are a pretty much infinite number of ways for human beings to interact. Narcissist aren't great at interaction, and have fewer choices.
Yup, connectors are a major nuisance.
I need some low-profile pcb-mount right-angle high-voltage connectors that wouldn't be too weird for customers to mate with.
John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics
That's still a pretty tall order. The SD-24 will do 30-ps 10%-90% with super clean pulse tops, positive and negative polarity.
My current best is a dedicated TDR with 60-ps edges and quite decent pulse tops. It's unipolar, needs ~10 pulses per delay, but a stuffed board with USB serial control costs about 0.3% of what an SD-24 listed for. ;)
The SD-24 uses a single SRD and some LRC/transmission line hackery to drive a 6-diode (bridge + blowby protection) sampler, which lets it measure bipolar signals.
Really a nice piece of work.
Cheers
Phil Hobbs
An axial lead carbon resistor used to cost as much as a Coke.
My first transistor, a CK722, cost $7, which would have bought a nice dinner for two.
Google estimates that a 1N914 cost several dollars ca 1960.
But please find us a diode data sheet that does not have a capacitance spec or curve.
John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics
It must have taken a monstrous amount of work to make the SD-24 as clean as it is.
Nowadays, we could make a sampler/TDR fast and ugly and math it pretty.
John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics
Interesting. Given that there are people still lusting after the SD-24, I'd hazard that there are application even today.
Joe
For a dedicated TDR, maybe so if the noise gain isn't too large and the nonlinearity is small enough. It's a bit amusing hearing that from you, though, since you generally trash the idea of deconvolution when other folk bring it up. ;)
Blowby rejection is quite a different problem, and that's super important for a general-purpose sampling front end. I don't have a good circuit for that. It would be fun to try.
Cheers
Phil Hobbs
Why bother? Your "counter-example" was from the data sheet for a device that started off as a relatively expensive diode. I'm not under any obligation to take you seriously.
Hardly. I've been a fan of time-domain deconvolution for years.
Since it's an ""ill-posed problem", there's no good academic solution, so there's room for creative fiddling. The classic approach, dividing complex Fourier transforms, has issues.
Time domain is better:
There are lots of nice fast opamps around.
Sampling and TDR are fascinating. But could we make money selling a small USB unit? The most expensive part would be the box.
John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics
The two are equivalent apart from roundoff. You transform the signal and then multiply by the reciprocal of the system transfer function, suitably windowed to control the noise gain. Dividing by noisy small values is a good way to make a mess, it's true.
Again, if the nonlinearity isn't too bad. Applying a high-frequency boost of 20 dB to a 5% nonlinearity will not lead to good things.
How would you use it to get rid of blowby?
It's that "IP in a box" thing again. ;)
On power-up, ours defaults to a 10**7:1 time stretch mode, which is pretty generally useful--1 ps --> 10 us.
Cheers
Phil Hobbs
<
Does anybody know what's in the samplers of the 50 GHz channels? I don't dare to open them.
The HP TDRs did use some sort of convolution to improve the traces.
Gerhard
The problem, in either domain, is to intelligently (or pragmatically) deal with what is in effect the issue of dividing by very small numbers, namely applying a lot of gain where there's not much signal.
If the problem is fixing reflections and such, that's not bad. Risetimes can be improved maybe 2:1 before that blows up.
Snoop the input, invert, poke that into the sampler. That's what Tek did.
Easy to copy.
John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics
Have something to add? Share your thoughts — no account required.
Ask the community — no account required