Are hybrid parameters useful for design?

I've never used them and I don't know anyone who does. They are not on transistor data sheets. S-parameters are used for small-signal RF/microwave stuff, but only RF parts have s-params available.

We use simple mental models of transistors, which is fine for most design, without simulation. Complex/tricky things get Spiced, which includes nonlinear effects that h-params don't include.

Academia tends to be about 40 years behind engineering practice.

Are you in India? India has a great tradition of academic rigor. The US tends more to hands-on, hack-and-slash engineering, whatever works. Simple transistor models are usually good enough for us, at least in discrete-component designs.

I must disagree with you, John. I know you have heard of hfe. And, I still have numerous modern data sheets with such things as hre, etc.

John

I got exposed to h-params in school, and in the old GE Transistor Manual. But I don't use them... never did, actually. HFE is what we call beta in our working transistor mental model. Some of the concepts overlap. I suppose some of the disagreement is terminology. But the h-params of a transistor are just linear approximations at one bias point which, to me, represent a lot of work to get a non-general approximation of transistor behavior.

Do you use h-parameters to design transistor circuits?

I recently Spiced a circuit that really depends on transistor Vbe exponential behavior, and Cbe leakage, over temperature. H-params aren't useful for nonlinear modeling.

All those partial differential equations are semi-useless. You don't need them for simple circuit design, and they are useless for more complex nonlinear stuff.

I think these are some of the issues that the late Dr. Middlebrook attempted to address in his "Design-Oriented Analysis' approach to feedback amplifier design. Even some relatively simple feedback circuits can result in 'high-entropy' equations, completely useless for design, when analyzed using the usual techniques presented in EE textbooks. His approach hopes to provide a methodology for obtaining expressions and data about circuit parameters which are meaningful for design purposes.

It has been a while since I've read his work - there's a lot of material and in most cases AFAIK simulation is still required to fully apply the techniques involved to "real" circuits without a completely overwhelming amount of math.

Part of this is because -- as JL claims -- academia is behind the curve of practice.

But part of it is because if you just dive into doing design-by- simulation, all you'll ever learn is how to do cut and try design, without ever understanding the underlying reality. That's great for doing mediocre designs quickly, but it'll sputter to a halt before your goal if you ever need to really make things work.

Your "useful for analysis, but not for design" criticism is a matter of not appreciating the approach one usually takes for design: there is no good way to design from first principles. You decide what you're going to do, then you either copy something someone (possibly you) have done before or you make a guess at something that'll work. Then you analyze how it'll work, and use the results of the analysis to tweak the design

-- or to reject it altogether as soon as possible, so you can spend your time figuring out another approach.

Your "everyone uses IC's now" comment is mostly true -- but often that's because people don't feel comfortable designing with discrete components, and would rather use three op-amps and some circuit tricks than a single transistor and one op-amp. Knowing how to use transistors isn't the be- all end-all of circuit design any more, but it can still come in mighty handy.

I suspect that JL has forgotten just how much he did learn when he was dragged through all the h-parameter stuff. You may not learn things that you'll use later directly, but you certainly do learn how to do circuit analysis.

All design is making a guess about something that might work.

Then you analyze

Right. Good design is a wideband pseudo-noise concept generator followed by a very narrowband good-idea-bad-idea filter. The filter part requires good theory and math, at the intuitive first-cut level, and then at the detail level. Spice can be the secondary filter, *if* you don't miss the gotchas.

Sure, I've forgotten a lot of stuff. But I was already professionally designing aerospace and automation electronics when I learned h-param theory, and never found a reason to use it, then or now.

We have a large-signal model of how a transistor behaves, and it's good enough that we can predict small signal impedances and gains and currents and stuff, and it works for nonlinear designs, over temperature, and other places where h-params don't work. Transistors are actually pretty simple.

You can use a large-signal transistor model to predict small-signal behavior. It doesn't work in the other direction.

So what, John?

Allow me to cut and paste your response above (maybe because you have reading comprehension problems):

You say that they are not on transistor data sheets and you are wrong. THAT was the essential reason for my reply; not all your other verbiage.

No. So what? Are you just trying to pick an argument? Did you READ what I wrote?

I don't give a s**t and SO WHAT?. LOOK AT WHAT I POSTED!!! I am beginning to believe that others on this group are correct. All you want to do is argue. This is another of your attempts to move the target with a lame attempt to put the ball in your court.

F**CK YOU, DIPSH**T.

John S

I'm pretty much in your camp, Tim, but I'd want to dolly back the camera a bit and talk about the place of analysis in circuit design. I'm a sort of hybrid EE/physicist--all my course work is physics, astronomy, and math, but my EE history is longer--I got a soldering iron for my 10th birthday and have been building circuits and other gizmos ever since. My first EE job was before I went to grad school, when I spent a couple of years designing frequency control electronics for the first civilian DBS system (Spacetel from AEL Microtel, introduced in 1982).

Simulation has a lot in common with prototyping--you can poke something till it works, but you won't know *why*, and even more importantly, you won't know whether the performance you're getting is anywhere near the fundamental limits.

Doing the analytic math to find out how good the circuit _could_ be is the only way to be sure that your design is any good. It can very often be worthwhile to trade off, say, 10 dB of SNR for a factor of 3 decrease in cost, but you're always better off knowing that that's what you're doing. It's also easy to lose 10 dB of SNR for no benefit whatsoever, and that's just plain stupid.

I don't think that I've ever multiplied an H matrix by a set of input parameters to get a set of output parameters, but I've done a whole lot of algebra/calculus/analyis to find out the ideal performance of circuits, which allows me to make intelligent tradeoffs of cost vs. performance.

Being a Spice jockey is not the way to wizardhood.

Cheers

Phil Hobbs

HFE, static beta, is on most transistor data sheets. I acknowledged that. The other large-signal and small-signal h-params usually aren't. Transistor (and diode) data sheets used to have a lot more info than they do nowadays.

The OP asked if people use h-params in design. That's what this thread is about. Why shouldn't we ask you, too?

If you don't design electronics, why are you so opinionated, and so angry, about electronics design? Why are you going personal and ballistic and cursing in a technical discussion?

The thing about matrix math is that, at least for me, it hides all the causalities in one mathematical black box. I'd rather use step-by-step analysis (Thevenin reductions, algebra, like that) so I can get a feel for what's actually causing what.

I agree that Spicing something is no substitute for a qualitative and somewhat quantitative understanding of what's going on. But Spice is sure pragmatic.

bit

cost,

also

Sure. Not many of us would try doing an analytic optimization of some complicated circuit using Gummel-Poon models and so forth. Spice has an important place, and so does prototyping. But for theoretical pefornance calculations, a page or two of algebra is definitely the way to go.

Cheers

Phil Hobbs

I can't imagine any educated engineer not doing back of the envelope design on discrete devices, even if it is just a simple gm*ro to estimate the gain of a stage, gmC for bandwidth, etc. But there are plenty of clueless engineers out there that just plop stuff down in spice.

If you are selling moderate performance boards by the hundreds, then just how optimal do you need to be? If you are selling chips by the hundreds of thousands, then it is a different story. In analog chip design, you need to meet the spec over mil temp at the minimum power. If you don't, somebody else will.

How could you possibly have missed them on the 2N3904 datasheet??? Do you even look at the datasheet? They're on bunches of others, mostly older, too.

The most popular transistor model is hybrid-pi, hands down. Usually as a first approximation prior to simulation, a geometric mean of small signal parameters at the corners of region of operation is sufficient for preliminary performance estimation.

Please cease and desist presuming to speak for how things are done in America. Mnay people have heartburn with your methods.

look at the datasheet? They're on bunches of others, mostly older, too.

Yeah, mostly older. Of the transistors that we use, most data sheets have hFE, and only one out of the dozen that I checked had any other h-params.

approximation prior to simulation, a geometric mean of small signal parameters at the corners of region of operation is sufficient for preliminary performance estimation.

Mnay people have heartburn with your methods.

Design anything interesting lately? Did you use h-parameter analysis?

It's just one way of looking at things. The discipline is probably not wasted.

RL

even look at the datasheet? They're on bunches of others, mostly older, too.

first approximation prior to simulation, a geometric mean of small signal parameters at the corners of region of operation is sufficient for preliminary performance estimation.

America. Mnay people have heartburn with your methods.

Yes, but not "electronical"- too boring. And didn't I espouse hybrid-pi? That is not h-parameter. All that h- y- g- and who knows what else was a force fit into two-port network theory of the time.

India has a tradition of excellent rote-learners. The earlier tradition of rote-learning verses to impress others (to rise in social rank) is replaced by the slavish solving of complex mathematical problems of no originality. There is no love for one's field, only ambition to be better off than the rest. That is why, you get brilliant engineering students from the IITs ending up in management, banking and finance. The country does not have appreciable input in well-cited peer-reviewed scientific papers. Patents are minuscule. Meritocracy is absent, so meritorious Indians run away abroad and try never to come back. I have seen the academia from up close, and the lesser said the better...

My precise questi> I'd rather use step-by-step

h-parameters are a subset of the entire analysis thing. We must restrict ourselves to discussing about h-parameters

Why are you and snipped-for-privacy@gmail.com personally attacking JL. He has stated a point, but I did not feel he wanted to pick up an argument! Please, I sought your, and others', help with information. Please let them help me with the information they can provide, which they believe is correct. It is his personal belief. You state your point and stand back and let someone else say his.