new GaN fet

At 4 GHz, a short becomes an open if you hang a two-inch cable on it. RF folks are very fond of that sort of pure frequency domain trick.

Cheers

Phil Hobbs

That's because non-linearities are avoided like the plague. They cause harmonics that screw you /and other/ users up. There are lots of /enforceable/ regulations to prevent that, with an associated squad of men in black suits.

So wouldn't you want a Spice model to evaluate that? Not all parts are used small-signal.

The part is specified at its *3 dB* compression point. Pretty hard to avoid lotsa distortion there!

Cheers

Phil Hobbs

It must be hard to simulate frequency-hopping complex-constellation multiple-antenna RF systems. That's probably the big use for million-dollar oscilloscopes.

Yes. Well before computers. And the key was finding the right mathematical and physical approximations, yielding an analytically tractable theory that was also useful.

Well, the device manufacturers do have such models, but the manufacturers do not release anything that revealing, to avoid educating their competitors.

Yes. NTIA.

Yeah, but DC to RF efficiency is higher, which is OK is one can afford a bigger output filtering system.

Often, intermediate driver amps (which drive the final amp) are 3db, but the final amps (which drives the antenna) are 1 dB.

Joe Gwinn

A few competitors will spend big bucks to measure the enemies parts, and they have the facilities to do it. So why should a thousand end-users have to do the measurements themselves, or spin product revs until it works?

I asked MiniCircuits for Spice models of their phemts. They were adamant, agressive even, that they will NEVER have Spice models.

Their data sheets should have a giant header that shouts

FOR RF USE ONLY

Yes, but they *do* have them.

Actually, much of what they make is better simulated using a full EM field solver, which spice cannot touch.

My guess is that if they published simplified spice models, they would get endless questions about this or that inaccuracy, so they avoid the expense by not issuing such models.

They do, "MiniCircuts", right at the top of every page.

Joe Gwinn

torsdag den 3. marts 2022 kl. 22.13.17 UTC+1 skrev John Larkin:

would that stop you?

and isn't the RF use only implied with specs only covering RF?

The point being that TG was talking out of his, um, lower back.

Cheers

Phil Hobbs

You know this how, exactly?

SPICE is a pretty capable package for solving sparse systems of nonlinear ODEs, with a few features bolted on for other things such as transmission lines. (A transmission line has invisible internal state, and so can't be simulated by ODEs.)

It doesn't model carrier diffusion, which is probably its worst deficiency in high speed circuitry. Second worst is that its noise simulation capability is very limited--it's just a linearized propagation-of-errors calculation based on a single operating point. In reality, many noise sources don't behave that way--for instance, shot noise depends on the instantaneous current, so noise correlations with signal can be far from negligible in real life.

But most of those things aren't done any better by your average full-wave EM code, which has zero information about the circuit properties and probably couldn't do carrier dynamics to save its virtual life, even if you gave it a physical description of the circuit with the required spatial resolution everywhere. (This would not be your elegant lightweight input file, you understand.)

I wrote a clusterized optimizing FDTD code for my antenna-coupled MIM tunnel junction work about 15 years ago, so I have relevant experience.

Op amp makers have the inaccuracy problem in spades, which they handle by lying through their teeth AFAICT--they say that better models would reveal too much about their designs, or run too slowly, or put warts on everybody's nose, or something.

Cheers

Phil Hobbs

Then they lied to me more than once.

Field solver for a phemt?

Best avoid s-params too.

Is that what "minicircuits" means? RF only?

I made a nice wideband step attenuator with a SAV-541. It's not obvious from the data sheet that this would work. I started with Phil's spice model for the SAV-551.

Of course not. We use RF parts in pulse applications all the time.

Yes. They can afford to blow off us time-domain buyers.

Actually, the SAV-5xx data sheets have DC drain curves! That's way more than most RF parts reveal.

We measured a lot more.

Am 03.03.22 um 16:21 schrieb snipped-for-privacy@highlandsniptechnology.com:

We talked about pin diode switches last week. Do you know the carrier lifetime of the diodes? Does Spice have any idea what that might be? And this is THE parameter that makes PIN diodes happen.

And the voltage exactly when/where on a polysilicon gate. End, feed point, somewhere in between?

What with noise calculations in a nonlinear environment? Try that with Spice in a chopper amplifier.

Which operating point would you choose in the chopper amplifier? Switch at A or B? You must make up your mind; noise analysis has not the concept of time.

It takes harmonic balance methods to do it.

Spice is NOT a large-signal simulator. What you get is small signal behavior around approximated operating points that shift around as crazy as a function of time.

Transient analysis is wrapped around the small signal linear analysis as just another layer.

Ask THEM for example: <

X-parameters are like s-parameters, only large signal.

You also get ADS design kits fom NXP, or Infineon etc. <

cheers, Gerhard

On a sunny day (Thu, 03 Mar 2022 13:13:01 -0800) it happened John Larkin <jlarkin@highland_atwork_technology.com> wrote in snipped-for-privacy@4ax.com:

John, that is their market. If you want to use your bike for mountain climbing you cannot demand all bike manufacturers to give specs for that. It is NOT that hard to make a test circuit and measure things, I do that with most new to me devices including Minicircuits (you actually gave me the idea to try some of their stuff (mixers)..) If it works it works, I have done spice simulation that told me things did not work that I used for many years without problems... Spice may be nice for complex filters etc, but there are better programs for that the rest really needs just hands on experience and UNDERSTANDING about how them electrons like to move. And the time factor, test maybe be an afternoon and even bending a PCB trace makes lot of difference, spice you can do for weeks, have then nothing you can be sure about that works :-) Imagine WW3 and you only have a hand full of BC547 or something that survived the EMP and now you need to build that radio No spice bummer... No problem for me.

They have defined it as such, so deliberately discouraged a large class of potential customers. That's weird.

We do measure things. And have to generate our own Spice models. In the case of this new fet, it's arguably a depletion fet and there is no Idss specified. They basicaly say, bias it until it works.

That's crazy.

Just more work for everyone.

They probably support people who will buy millions.

We should distinguish between a "Spice model" and a Spice program.

This is a jfet model:

.model 2N3819 NJF(Beta=1.304m Betatce=-.5 Rd=1 Rs=1 Lambda=2.25m Vto=-3 Vtotc=-2.5m Is=33.57f Isr=322.4f N=1 Nr=2 Xti=3 Alpha=311.7u Vk=243.6 Cgd=1.6p M=.3622 Pb=1 Fc=.5 Cgs=2.414p Kf=9.882E-18 Af=1 mfg=Vishay)

It assigns values in SI units to elements of device physics. That, plus some package parasitics and thermals, defines the device. That could be the input to LT Spice or to an expensive RF suite or to 12 professors with slide rules.

But the values are there. Narrowband load pull measurements are different.

Opamp models can be incomplete, but they are better than s-parameters.

On a sunny day (Fri, 04 Mar 2022 06:36:25 -0800) it happened snipped-for-privacy@highlandsniptechnology.com wrote in snipped-for-privacy@4ax.com:

Well its nature:-) There may be / will be some production spread in the cut-off gate voltage, that is with all FETs. You could make a feedback loop..

But you need that test circuit and measurements anyway? I do.

And it is also the fun of learning something new :-) Am working at version 0.5.1 of my FLIR software, things evolve.

The first step is to select a part. I don't want to buy and test a hundred parts. Basic DC specs are fundamental.