more LTspice weirdness

How many resistors, values and what network pattern?

I'm struggling to think of a way that could happen without any non-ideal components (or are they real resistance models and high value ones at that with series inductance and stray capacitance included?).

Very high impedance precision resistors IRL have distinctly non-ideal resistance behaviour at 10^9 upwards. Depending on the batch and maker.

What's your reltol?

Ideal current sources are hard to reason about sometimes, I try not to use them in simulation when I can at all avoid it

Show us! Post the asc file.

Jeroen Belleman

It's also about time steps in LT Spice. If I force the time step small, my pulse gets flat.

It looks like the first step of the current source overshoots, and then the voltage creeps back toward right.

That's a well known problem with numerical integration. There are solutions, but they involve adding extra steps.

Mike Englehart admitted here that LTSpice wasn't made quite as fool proof as it might be, to let it run appreciably faster.

It's a lot faster and cheaper to put a circuit into LTSpice than it is to put it onto a printed circuit board (and lay out the board).

The results aren't as reliable, but you can make the point that a particular circuit is sub-optimal very quickly and cheaply.

LT is wonderful. I can evolve circuits way faster than I could do the math myself, or beadboard.

I just used a 1000 amp pulse to evaluate part of a circuit. Can't get that from Amazon.

Of course you can't always trust the part models, or trust Spice itself. The human instinct part is key to resolving that issue.

Don't trash your soldering iron.

I hope that circuit design will be the last skill to fall to AI.

I'm desiging attenuators in my spare time. I'm using LT Spice to tune topologies and values, and testing parts (some to destruction) alongside.

I couldn't do the algebra to get the part values, so I sim it all. Being constrained to parts that I have or can get wrecks a nice pure mathematical approach.

By Monday morning we'll have about a billion 8 KW pulses pushed into a cute little Caddock DPAK resistor. I'm Spicing attenuator circuits using those parts, especially fast step response. They have a lot of parasitic L and C.

I have a cookie can full of dead 20 watt and 40 watt commercial attanuators. Our pulser kills them at apparently tiny joule and watt levels.

The Caddock MP725 dpak resistors seem to be very tough (we'll know more on Monday) and are pretty fast.

I'll torture some Susumu thinfilms too, for pulsed overload behavior.

One handy thing in LT Spice is the BV element, the behavioral voltage source. It will do math. I'm using one to compute the input impedance of my attenuator. And I'm applying a 1000 amp pulse to the output to see the output impedance.

I believe it was Peter Baxandall who said his circuit design was often optimised by analogue computing - he built it and changed component values to get the best performance. The best analogy was the circuit itself.

We rarely build a prototype for fast stuff. Usually we go for the final multilayer PCB and hope we can sell it.

A tiny dermeled proto can be useful for characterizing parts whose data sheets are suspect.

Was that before Spice?

My attenuator problem is a numerical nightmare. It would take at least forever to iterate with solder. And the tolerances of your junk box parts will skew the final design.

Spice sure helps.

Somewhat!.

And only somebody as dumb as John Larkin would bother.

Why?

Pull the other leg.

And time. Very high frequency circuits call for exotic substrates, which Rogers sell. Finding a printed circuit board shop that has the right substrates in stock isn't easy, and they charge heavily to cover the capital cost of maintaining that stock.

All of which are measurable and calculable

Not strictly true. The Cambridge Instruments electron beam tester could produce 0.5nsec long pulses of electrons. We did it by generating pair

It was distinctly helpful. There are models available for 5GHz bandwidth broad band bipolar transistors.

I even put together a three stage Percival distributed amplifier (but that didn't get below 0.8nsec - sad because it used more but cheaper bipolar transistors than the faster production version).

Been there. Done that. Even published - rather against my inclination, but my co-authors wanted the publication. There the bipolar transistors were a pair of 5GHZ BFT-95

Ghiggino, K.P., Phillips, D., and Sloman, A.W. "Nanosecond pulse stretcher",Journal of Physics E: Scientific Instruments, 12, 686-687 (1979).

LTSpice was first released in 1999. Spice itself has been around a lot longer than that - I first used in England around 1974 - but I used it at work on my employers computer. LTSpice was the first version of the program I could afford to put on my home computer, and it became available at the time when home computers got powerful enough to run it as a useful speed.

Enough people used it that the semiconductor industry released models of their transistors for free that LTSpice could run.

Peter was an impressive engineer, and I've been admirer of his work since about 1970, but he did get some stuff wrong.

Audio is not very quantitative. You can design it literally by ear.

LT has ideal and lossy transmission lines.

I decided that I really didn't understand transmission line transformers, so I Spiced some. It was a revelation in several ways.

As Mike says, the real value of Spice is to train your instincts. I would add, and to do the hard arithmetic too.

So work on FR4. Spending big on exotic boards is seldom sensible.

Most shops have the Isola stuff available, when it matters. The Rogers lam was like copperclad shoe leather.

Peter Baxandall got more into audio after he retired, but even then he was a champion of the analytic approach to audio design - and a sworn enemy of the golden-eared boys. The battle between the engineers and the more subjective golden-eared boys played out of the pages of UK magazine HiFi News and Record Review during the 1980s - it was worth reading back then.

I even had a letter or two to the editor published there back then. Surprised the hell out of member of my hockey team.

Until the FR4 screws up your edges.

Some of them are remarkably soft.

That doesn't make it all that easy. Happily I've been able to evade that.

DC-to-GHz is harder.

You do have to pay attention. There is enough literature on transmission line transformers that you don't actually need to "train your instincts" with LTSpice.

Ghiggino, K.P., Phillips, D., and Sloman, A.W. "Nanosecond pulse stretcher",Journal of Physics E: Scientific Instruments, 12, 686-687 (1979).

cites Matick R.E. "Transmission-line pulse transformers - theory and applications" Proc.IEEE 56 47-62

Just as well - LTSpice wasn't around back then.