Fast sampler

Neat. No step-recovery diodes.

The LVDS line receivers are radical parts, for 30 cents.

We also used ATLC2 to get a good match of a cheap fat-pin edge-launch SMA connector to a multilayer PCB. That was fun.

Well, 40 years does get you something sometimes. ;)

Yeah, you tipped us off to the possibility, so Simon cranked on it. Deleting the L2 and L3 grounds, and via-stitching a L4 ground through the whole stack, makes the cheap connectors pretty good.

Cheers

Phil Hobbs

Very nice. Mine was more like Uncle Scrooge's version with every penny being turned around, only one diode :-)

However, that darn diode is 0102 size and I had a hard time handling it with my not so young eyes.

And those cheap yet blazingly fast RF transistors, thanks to cell phones and all. They make nice pulsers. But they are like the princess on the pea, very low Vce and if you go a smidgen above ... poof.

They're not that bad, really--their betas are so high that BV_CEO is lowish, but BV_CBO is 12 volts or more. Their saturation behavior is still pretty BJTish, though. ;)

Cheers

Phil Hobbs

Cute. When I got stuck with driving a diode bridge, I used a transmission line transformer to get a perfectly balanced drive.

As Kibble and Rayner point out, carefully wound (and that means a non-progressive winding) 1:1 transformers can get to one part per billion balance. Other ratios tend not to let themselves get pushed beyond one part in ten million.

I toyed with the idea of using a PHEMT as a series-switch fast sample-and-hold.

Hey, here's another goofy idea:

We used to make fast linear ramps, driving a comparator against a DAC, as a programmable delay. But we got smarter and just used an RC charging thing, and mucked the DAC codes with a polynomial to get our delay.

But what if the comparator sees a fast RC on one input and a slow RC on the other? The exponential curves cancel, and you get a nice slow linear sampling timebase. If you don't quibble too much.

They work well for that. A couple of years back, we did a POC for the Navy that used several SAV551pluses—100 ps is doable. The main problem is that their voltage gain is lowish, so you don’t get as much speedup as with a BJT.

And of course they’re 10x the price.

Not sure about that. For the proto, I used a ramp from an arb to make the threshold—the sampling loop converged at each point, so I wound up with a

The fast bit was all over before the slow bit moved perceptibly.

Cheers

Phil Hobbs

The other issue is that the prop delay depends on the overdrive. Since we’re comparing a ramp to a fixed threshold, which that basically means how far the ramp rises during the time required for the positive feedback to get going.

So we still need an online calibration. Fortunately that isn’t hard—an open-circuited bit of coax is enough. It doesn’t have to be done often.

Cheers

Phil Hobbs

I did caution about quibbling too much. One issue is that the LVDS line receivers have a bunch of offset as the common-mode voltage approaches the positive supply rail. And of course the esd diodes are nonlinear capacitors. And things always ring a little. Geez, nobody's perfect.

One of my guys did a bunch experiments using an LVDS receiver as the comparator in a picosecond-resolution delay circuit. We use a 16-bit DAC and a 4th order polynomial and calibrate the polynomial for every channel. Our P500 has, I recall, nine of those.

The line receivers don't seem to have any significant amount of kickout, either--we can sweep the Rx pulse across the Tx pulse with no apparent funnies due to interaction. Have you folks seen any kickout issues?

Of course the kickout might be delayed, I suppose.

Cheers

Phil Hobbs

We didn't test for that. When a comparator fires, all sorts of stuff happens downstream, that could jostle adjacent channels.

Our comparators typically drive a 1 ns Tiny Logic flipflop as the next step in the signal chain.

Standard ECL wasn't as fast as 15 cent Tiny parts are now.

But it's current steering logic and the supply rails stay a lot cleaner than you see with CMOS switches.

And what's Standard ECL now? It was Motorola 10k back when I was young, and Motorola/Philips/Fairchild 100k a few years later. Motorola ECLinPs took over a about when I stopped using it, about when I started posting here, some twenty years ago.

Not a big problem, just place pillows around it, RC filters on the supply. Drive was never an issue in my case but on rare occasions I have used cheap signal transformers to isolate that.

The larger concern with supply rails is low frequency noise coming in, causing phase noise. That is where capacitance multipliers can shine.

AFAIK it's 100E but I have not used any in ages because I always found them overpriced. I don't like it when two ICs cost more than a crate of beer :-)

I measured one FPGA at around 1 mV per picosecond prop delay, on the 1 volt core supply. Prop delay is about inverse on voltage in cmos.

MC10EPxx, SiGe Eclips Lite.

Really fast and really expensive is Gigacomm, which is actually CML. The NB7V52M flop is only about $13 in quantity.

In my last TDR psec-jitter would have made the client unhappy. The sampling window was 100psec but it wasn't supposed to move unless told to.

So far I've only needed "semi-analog", meaning just one bit and then I did it using RF transistors. It is amazing, you can buy >100GHz fT for less than 20 cents in qties.

When I was a kid I had to shell out around $3 for an AF116 Ge-transistor that had an fT of 75MHz. In 1970's Dollars, which really hurt. Digital wasn't any better. I needed a 1kbit RAM for a project and that set me back about 10 bucks. It still works.

My first transistor was a Raytheon CK722 germanium. I think Ft was measured in KHz. It cost $7, about a month's allowance, or dinner for two at a decent restaurant.

I got tubes for free.

Why is a LASER ruler that can measure distances in air with 2mm accuracy $16,

and an OTDR for measuring fiber $600?

Air is a rather less complex transmission medium than optical fibre, but the most likely explanation is that you can sell a lot more laser rulers than you can sell tools for measuring optical fibres, so you can afford the masks for an ASIC for the former, but not the latter.

Cunning designers have been known to use ASICs in applications for which they weren't designed, but if the application is too specific, it won't work.