Re: OT: sound speed depends on frequency on mars

I hope you pointed out that buried strip-line isn't dispersive. I have pointed this out here from time to time.

It would be a bit silly. You can make lumped constant transmission lines by linking a series of capacitors with discrete inductors, if you want a high impedance transmission line - people sold them thick film hybrid assemblies, and I even used a few, a very long time ago.

Unless you wore a pressure suit. Not exactly the usual music festival costume, but you'd need to wear one to survive an outdoor concert on Mars.

Of course it's dispersive, maybe a bit less than microstrip.

It's hard to keep up decent impedances on stripline in a multilayer board, especially 8 or 10 layers.

Most ideas seem silly to people who are by nature hostile to ideas.

Dismissing is easier than thinking.

Why do you think that?

Stripline is buried between two ground planes. The only tricky part of impedance control is the thickness of the dielectric in the two layers above and below the strip-line. In a ten layer board this is thinner than it would be in a board with fewer layers.

Pay enough for close-tolerance substrates in the two relevant layers and you should be okay.

Not a problem I've got.

Thinking about what a loading coil might be doing to the impedance of a PCB trace isn't something that you seem to have managed to do.

Because dielectrics are imperfect, especially FR4, and because there are lots of papers online that analyze dispersion in stripline.

And eventually the trace has to be skinnier than PCB houses are willing to etch. Standard pricing seems to be around 5 or maybe 4 mils width these days. We do a lot of 5, to sneak between BGA balls, but sometimes even 5 is too big.

Thin dielectrics have tolerance issues too. I'm talking about real PCBs here, not ideal theoretical stuff.

"Pay enough" can get crazy fast. I don't want to pay hundreds of dollars for a smallish PCB.

I certainly had the idea. I might Spice a bunch of ltlines with inductors between, just for fun. It's unlikely that I'd use such an arrangement in real life, but it's just possible, especially if analog quality of a fast edge matters, like in a laser modulator maybe. It's preferable to just keep all the traces very short, but that's not always possible.

But you can't cite any of them. You wouldn't use FR4 around a stripline if you wanted a low-dispersion transmission line. There are better substrates his frequency work.

Why?

Printed circuit board are always real.

An eight or ten layer PCB isn't going to be small. You only need lots of layers when you have to connect lots of stuff.

John snipped the rest of that senstence, without marking the snip.

In a remarkably half-baked way.

You seem to be intent on re-inventing the lumped constant delay line, without being aware that they were commercially available some thirty years ago, back when I used them. They may still be available.

Back around 1985 we planned on making one of our own when we wanted a microsecond or so of pure delay on the main signal to synchronise it with the small second order corrections we were adding in from a couple of analog multipliers (which had their own propagation delay).

Can't you google?

You wouldn't use FR4 around a stripline

Run the Saturn program. More layers make the dielectrics thinner, so to maintain a useful impedance the traces have to get narrower.

Exactly.

Ideas start out fuzzy, or at least they should. I tell my kids, stay confused for a while.

The Tek 545 30 MHz scope had a gigantic, lumped, tunable, differential delay line up to the CRT, so you could see the edge that you triggered on.

Meander-line sections connected by loading coils could be interesting. One product that I'm considering now is a programmable delay line, and that idea might help.

In the 1980's we had NIM-format boxes with binary-weighted-length cables and cheap slide switches, and CAMAC modules with basically the same cables, but with fancy miniature DPDT relays in metal TO-8 style packages. Physicists would invariably mess up the relay contacts.

Jeroen Belleman

I was thinking of using the cute little $1 Fujitsu telecom relays, which are good up to about 3 GHz.

I never did any NIM, but we did a bunch of CAMAC. It was a strange bus, 24 bits of open-drain read data and a separate 24 bits of write data. Design by physicists!

But the geographical addressing was great. Too bad VME didn't do that.

Can't you? You know where you found the papers (if you did) and are much better placed to define the search terms that would throw them up.

I can do that with a handheld calculator. What you seem, to have missed is that low dielectric constant substrates give you wider traces for a given impedance. I got a 150R line on the surface of teflon-alumina substrate.

If you don't think about what you are doing.

You seem to have chosen to stay confused for decades.

The discrete loading coils are lumped elements.

All true, which doesn't make them any less useful.

Like the MC100EP195?

I've tested that part. It's expensive, drifty, and has an insane amount of jitter. It's funny that its resolution is "about 10 ps"

Maxim and I think someone else made CMOS programmable delay line chips, which were equally bad, not to mention discontinued.

We mostly use fast ramps and comparators and DACs to make programmable delays. Jitter is low and polynomial calibration makes them very accurate. Cheap too.

Of course I keep inventing things. That's my job.

It's resolution is about 10psec, because that's the - temperature dependent - delay through individual delay elements. If you want it to be more precise, you have to control the part's temperature, or re-calibrate every few minutes. That's what I was planning to do when I contemplated using it, and figured that I could get it done within a millisecond - which did call for a fast A/D. Which one I can't remember because it was back in 1998.

The RMS random clock jitter is specified on page 10 of the data sheet, and it's around 1psec which pretty standard for ECL parts - not remotely insane.

The nice thing about ECL is that it doesn't mess up it's power rails in the way that CMOS and TTL do, which does get rid of one jitter source.

I once got rid of some nasty sub-nanosecond jitter on a TTL clock by generating it in ECL (run between 0V and -4.5V) and getting it out of an ECL-to-TTL converter.

I had expected the ECL-to-TTL converter to be equally susceptible to noise on the +5V rail, but I was happy to find out that I was wrong.

Monotonicity is TBD! It should say "Fat Chance."

Our ramp delay generators are absolutely monotonic.

Temperature control, and periodic recalibration, are not practical in a sensible instrument. What do you do if the customer makes a trigger when you're in the middle of calibrating? Blow up their laser?

We calibrate delay generators in production test, and they work fine after that.

I measured a lot more. And the horrible delay tempco is essentially jitter, as far as a customer is concerned.

The Moto ECL-TTL converters, like the 10H125 or the ELT21, were slow and expensive and had ghastly jitter. The Arizona Microtek part is better but still pretty bad.

An LVDS line receiver is cheap and hugely better.

I never claimed to be sane. Sane is boring. I do claim to design and sell a lot of electronics.

The modules I mentioned had coiled-up coax inside. You can't squeeze much delay into PCB traces. I've come to think of

I used to spend most of my time with CAMAC in the 1980s, but I haven't touched it since more than 30 years. It was a weird and wasteful bus system, but lots of physics experiments used it at the time.

NIM was simple and stupid. Just a crate with sturdy power supply connectors in the back, with +/-6, +/-12 and +/-24V supplies. There was more, but no one ever used that. Crates with lots of analog would always exceed the current available on the +/-12V, and for I/O, you were on your own. Other than that, it was handy because it was everywhere. It's still used at CERN, whereas CAMAC has vanished. Simple is better.

Jeroen Belleman

A delay box could be a hybrid, some coiled-up coaxes and some PCB wiggle traces.

We invented this for aerospace test

PCI/PXI boards don't have the panel space, power, or cooling to do serious stuff.

As far as I can remember I used a regular Philips 100K ECL-to-TTL converter, and it obviously didn't have ghastly jitter. I was careful about power rail decoupling, and a ham-fisted half-wit could probably have managed to introduce significant jitter. Ran van Dongen, who had designed the original almost-all-TTL system, was neither ham-fisted nor a half-wit, if a bit less ECL-aware than he should have been. He rather liked what I came up with. I mostly used Motorola ECinPS parts which hadn't been around when he had designed the original system

ECL is a low volume product, so it isn't cheap, but when you need it it is worth the money.

But it doesn't produce a TTL output.

When in fact you evolve and sell a certain amount of electronics for niche markets. Your forays into higher volume markets don't seem to have done well. You are insane enough to think this gives you some kind of authority.

DS90LV012ATMF/NOPB is essentially a fast rrio comparator. Its outputs swing from ground to Vcc. It costs us 30 cents. If you buy a similar part that's called a comparator, it costs 5x as much.

Like most cmos rrio gadgets, its offset increases as the common-mode approaches Vcc, but that's no big deal in a polynomial-calibrated ramp.

I mostly like to design electronics because it's fun and keeps me amused. Better than having a real job. People buying it is a nice side effect and pays for test equipment and snacks.

Don't you think designing electronics is amusing? I guess not.

It's certainly fascinating. Having to toss out lots of solutions and start over isn't remotely amusing, but that's what it takes to get to a good design.

It has a lot in common with playing colossal cave

But its is very amusing.

Trying lots of arguably crazy ideas is educational, and has a chance of stumbling onto something really valuable. But one has to do it fast, because there's literally a universe of possibilities to explore.

Exploring Colossal Cave is a good analogy to exploring the electronic circuit solution space, except the circuit space is much bigger hence impossible to explore serially.

Serial implies one-dimensional ordering, while Collossal Cave was two dimensional.

Most circuit problems have a single input and single output, but the space in between can be as complicated as you like.

If the problem you are trying to solve has conventional solutions, these can serve as known routes through the territory you need to explore, and I've had people reject my short-cuts because they didn't understand the problem clearly enough.