LTspice tube models and their accuracy

Hi,

this far all my tube work has been manual bias point calculations by hand and this has provided such accurate results I've never needed LTspice. I've used LTspice for opamp and PSU work and become quite fond of the software. Results are trustworthy and very accurately translate to real world it seems.

I've read reports it could also create useful ballpark figures or sometimes even perfectly accurate predictions for real world tube designs. But then I've also read the quality of the tube models vary a whole lot, much more than transistors and opamps directly from the manufacturers. They seem to vary so much in fact I've shied away from starting any tube design in LTspice.

Please share experiences. And please post or link to tube models or whole collections you know to be accurate, trustworthy or at least useful for general design decisions.

I'm a member of the yahoo ltspice group and found plenty of models there, but their sources and accuracy is often questionable.

Mike

Set up the tube with sweep voltage and current sources, re-create the plate curves.

Are they close in the mid-range of normal operation?

Are they close out in the corners, abnormal operation?

Take the grid positive and verify that it goes toward 1K resistance for more than +1V or +2V. (I've seen several models that omitted grid conduction, and "allowed" impossible results.)

That would work, but kind of defeats the time saving idea I had in mind when starting this thread. Which I could have mentioned too!

I can trust the transistor and opamp models from manufacturers and it's so quick to get a working proto running, itching to built. But comparing tube models to old datasheets with themselves of somewhat suspect accuracy is not my idea of fun. Maybe someone already did that? Sounds like it could fetch a decent price.

positive grids I could live with, but plate resistance that is completely off the wack is pretty unforgivable, as are the various omissions and versions of "heaters" I've seen. Sometimes they remember to mention there's a heater you have to "warm up" in simulation before the big moment, sometimes they don't. Makes one want to forget about tubes and reach for the nearest opamp.

I think you have to be very careful using spice models for tubes - not only careful of the models themselves but of what you attempt to simulate with them. There's o lot's of info on the web about how various people have derived their models and how they compare with data sheets or actual tubes. Probably the best known and most respected models are those of Norman Koren but even they disagree with real tube measurements. Some of the most accurate |I have found are those made using curve captor (open source project on sourceforge) which allows you to input the data from the manufacturers curves and generate a spice model form that data.

Bottom line in my experience is that the models are only good for verifying a basic topology and obtaining order of magnitude figures for distortion and open loop bandwidth. After that, there's no alternative but to build and test the real thing. Part of the problem is the wide diversity of types of tube with the same part number e.g.12AX7. Parameters vary between regular and long plate versions, between manufacturers and between batches from one manufacturer.

Cheers

Ian

Been using it and made my own models...gave it to a few guys. No more.

analag said:

made my own models...

Click to expand...

Could have mentioned that more often while posting the numerous wildly untested looking gadgets.

Anyway, thanks for the options all of you.

> the time saving idea

With a socket nailed to a slab of wood, you can generally try a REAL tube faster than you can set-up a simulation.

> I can trust the transistor and opamp models from manufacturers

In extremes, you can't.

The originators of SPICE did a good job of modeling the BJT with a minimum of parameters. In the "normal" zone, the results are better than reality (since most BJT parameters vary 2:1 either way). Works down to quite low current too, because SPICE has to model turn-on behavior; but not precise down to silly-low currents. Excess current action is often very-very approximate.

> plate resistance that is completely off the wack is pretty unforgivable
> mention there's a heater you have to "warm up" in simulation

Don't understand. If plate resistance is not pretty-close, results will be WAY off at any but some nominal condition. (Do realize that the data-list Rp is often a show-off, and Rp at "audio normal" currents is sure to be 2 to 10 times higher.) And if you can't tell the heater is cold, maybe tubes are not your thing?

PRR said:

Don't understand. If plate resistance is not pretty-close, results will be WAY off at any but some nominal condition. (Do realize that the data-list Rp is often a show-off, and Rp at "audio normal" currents is sure to be 2 to 10 times higher.)

Click to expand...

Du.uuh!  PRR, definitely dun unnerstan dis.

Are you saying datasheet Rp is not usually consistent with the curves for triodes ?

There are two common values of plate resistance and neither is constant. If you look at the plate curves for a triode and follow one curve for a fixed grid bias voltage hen at any point the plate voltage divided by the plate current is the static or DC plate resistance.. If you move down to a larger grid bias but keep the plate voltage the same, the current is less so the static plate resistance is higher. This is not surprising because a tube is nothing but a voltage controlled resistor.

If you now pick a point on one curve and move along it until the plate voltage increases by one volt, the plate current will increase by some small value. This small change in voltage divided by the corresponding small change in plate current is the incremental or dynamic plate resistance. It is really the slope of the curve at any point. Since the curves are not straight lines, the slope must vary as you move up and down the curve and hence dynamic plate resistance varies.

Dynamic plate resistance rp and incremental transconductance gm are related by the well known equation my = gm * ra. Of these three, the only one that remains reasonably constant is my. Gm increases and rp reduces as plate current increases, that's why manufacturers quote them both at a rather high current.

Cheers

Ian

PRR said:

Don't understand. If plate resistance is not pretty-close, results will be WAY off at any but some nominal condition. (Do realize that the data-list Rp is often a show-off, and Rp at "audio normal" currents is sure to be 2 to 10 times higher.) And if you can't tell the heater is cold, maybe tubes are not your thing?

Click to expand...

Indeed the datasheet plate resistance is something the manufacturer would like their product to be. Wouldn't want that in the models, but the real world instead, or some average at least. Here's my favourite tube data page with quite a bit of revelations on datasheet vs. real world, positive and negative: http://www.klausmobile.narod.ru/testerfiles/index.htm

The point about heaters was just that I wish their behaviour was modeled the same in all tubes, or at least documented. Do I set 5 or 12 second warm up etc. crucial info.

> Are you saying datasheet Rp ... ...

12AT7 (used in larger/reverbier Fenders) is a seductive example.

http://www.mif.pg.gda.pl/homepages/frank/sheets/093/1/12AT7.pdf

The short-form data shows Rp as low as 10,900 ohms. WOW! Great in a high-Mu tube. Not much higher than the low-Mu 12AU7.

So we set up with a 250V supply, 47K load resistor, bias somewhat above half-supply, say 150V. Some test and interpretation shows a much higher Rp.

Why the difference?

The 10,900 _IS_ correct.... for _10mA_.

We are working 100V across 47K is 2mA. This data-sheet kindly plots Rp for a range of currents and several voltages, page 4. Our 2mA and 150V (squint) looks like 22,500 ohms.

Even worse if not driving "heavy" (for audio) loads. 12AT7 with 100K plate resistor (~~1mA) has Rp a whole lot closer to 40K than to 10K.

The "10,900 ohms at 10mA" data IS useful to a designer... a radio designer. A VHF amplifier needs high Gm, high gain, and moderate plate resistance. 10mA is not an uncommon idle or weak-signal bias for the radio front-end.

What's also notable: the oft-quoted Mu=60 is a big bug-splat. Mu actually varies from 80 to 30. The lower Mu is at the lower currents where we audio-heads often run tubes. It's really a semi-remote cut-off tube. (12AT7 is really an FM/TV front-end, though other tubes became more popular for RF and 12AT7 found more use in mundane applications.)

> Do I set 5 or 12 second warm up etc

I've rarely seen heater emulation, can only think of a few applications where I would care, and I sure would not trust it in any tricky situation where it mattered.

When _I_ run a tube the warm-up is a very small part of the total hours; to my mind, the tube has been hot forever.

There ARE rapid-start applications like mobile transmitters, where the final is cold until the mike is keyed. Warm-up matters. Do I trust the sim to really track a real tube? There's exponentials in the thermals, heavy exponential in emission, minor variations may make a big difference in when the tube can really push its 20 Watts output.

Then there are tricks like a quad of 6V6, their cathode current run through the heaters of a couple 12AX7 preamps. Good clean DC heat, but there is a double-warmup delay (first the 6V6es, then the 12AX7s). And the AX7s warm a bit slow because the 6V6 won't pass the 3X start-up current that a cold 12AX7 can draw from a constant-voltage source. But do I need to simulate this? The 25 second time to clear sound is a small part of one LP side. It may take 5 minutes to select and find that LP, the amp will be fully-warm by then!

I can't recall a model with heater warm-up emulation, and I would call it "broken" (since I'm not expecting to waste a million signal-cycles before anything happens).

Thanks to PRR & Ian for their explanations.

From Morgan Jones, Valve Wizard etc, I sorta think Ian's version is the most useful ... ie mu is sorta constant for a triode while gm & rp vary.

But surely, the idea behind SPICE is that it simulates something closer to the datasheet curves ...  rather than simpler Circuit Analysis sims which assume eg mu, gm & rp are all constant and linear.

Do the various tube models for SPICE come close to the datasheet curves?

Most Spice tube models do attempt to reproduce the data sheet curves. Simplest and least accurate are base on the basic 3/2 law whilst the better ones, like the Norman Kormen ones attempt to note accurately reproduce the data sheets. I have a small number of models derived directly from the data sheets using a program called curve captor. Within the limits of the curve fitting maths, these exactly reproduce the data sheet curves.

Edit: having said all that, you have to remember that the data sheet represents only typical performance and real tubes will very up to 10% from this. In particular, you will notice differences between simulated and measured DC conditions and the same with dynamic parameters like distortion even if using models derived directly from the data sheets.

Cheers

Ian

> SPICE is that it simulates something closer to the datasheet curves ...  rather than simpler Circuit Analysis sims which assume eg mu, gm & rp are all constant and linear.

SPICE itself will assume whatever it wants. The original project and follow-ons did a lot of work on BJT and MOSFET models. They can be good.... and then you find commercial models where Rb is 10 in *all* transistors.

Later authors proposed several models for tubes.

These are mostly VERY readily available...... and not un-readable........

Here's Koren's improved 12AX7:

.SUBCKT 12AX7 1 2 3  ; P G C;  NEW MODEL
+ PARAMS: MU=100 EX=1.4 KG1=1060 KP=600 KVB=300 RGI=2000
E1 7 0 VALUE=
+{V(1,3)/KP*LOG(1+EXP(KP*(1/MU+V(2,3)/SQRT(KVB+V(1,3)*V(1,3)))))}
G1 1 3 VALUE={(PWR(V(7),EX)+PWRS(V(7),EX))/KG1}
{more stuff not affecting the basic plate curves}

MU=100 is of course Mu.

EX is Child's Law's 3/2, which in real tubes varies from 1.2 to 1.8 depending on construction and current level. (It becomes just 1 for very-very low currents.)

EX predicts the general trend of Rp and Gm for various currents.

KG1 KP KVB are three fudges. Mu is only semi-constant, also EX is not 1.4 over all conditions. I forget which fudge does what; it is probably in Koren's essay. Or you can deduce it from where they appear in the equations for E1 and G1.

MU and EX plot the basic action. Three additional fudge factors limit how far it can model deviations in real devices. But it is enough to give excellent accuracy in, say, 12AX7 from 50uA to 2mA, 20V to 600V.

One particular thing not modeled (in this version) is Contact Potential. This offsets the bottom of the curves. KP KG1 warp the curves the right way but not the right amount. When I try to model super-low voltage operation, I see it does not agree with real tubes.