Do tube curves tell most of the story?

[bigwill]

If I were to import all of the data from a tube curve graph onto say, an Excel spreadsheet, and made some provision to approximately interpolate between each grid voltage curve, would it be possible to approximate the behaviour of the tube in software in various configurations?

Obviously it wouldn't take AC behaviour into account but would it be good enough to roughly compute transfer curves (ie input voltage vs output voltage) etc?

[EC8010]

Remember that valve curves are an average, so be wary of reading too much into them. Distortion predictions using a wooden ruler on paper curves are probably legitimate. Computer simulations to 16 significant figures are not. Real valves vary. A lot.

[bigwill]

Originally posted by EC8010
Remember that valve curves are an average, so be wary of reading too much into them. Distortion predictions using a wooden ruler on paper curves are probably legitimate. Computer simulations to 16 significant figures are not. Real valves vary. A lot.

Oh I realise it wouldn't be useful for obtaining distortion figures or anything like that - just determining linear operating points and sensible anode / cathode resistances.

I thought if I simulate tubes with their curves I don't have to learn the complicated maths

The way I might implement it is by using a successive approximation approach using data from a table to guess what the valve will do.

Say you had a B+ voltage of 300V, a cathode resistor, anode resistor and a zero referenced signal on the grid of a triode.

The simulation would pick an arbitrary resistance for the triode, say 100 ohms. For this instant in the simulator this is the resistance of the triode regardless of the grid or the voltage across it.

Now the simulator would have some initial voltages on the tube to work with, a cathode to anode voltage and a grid to cathode voltage. From these, the actual resistance of this state would be looked up.

Hopefully, going back to the first step with this new resistance value repeatedly will yield a result that gets closer and closer to the actual circuit values, negative feedback in software if you will....

[Miles Prower]

Originally posted by bigwill
Oh I realise it wouldn't be useful for obtaining distortion figures or anything like that - just determining linear operating points and sensible anode / cathode resistances.

I thought if I simulate tubes with their curves I don't have to learn the complicated maths

There aren't any "complicated maths" involved. Unlike solid state devices where the emitter, base, collector, source, gate, drain are connected, VTs consist of parts just hanging there in free space. Each element can operate independently, and so the fixed relationships you see in solid state (i.e Vbe= 0.6V, or the relationship between pinch-off voltage and Idss in the JFET) don't exist. You're better off with a plate characteristic graph and loadlines. Nor is uber-accuracy necessary. VT circuits are quite forgiving. After all, back in "the day" they didn't have ultraprecise, laser trimmed, 0.1% metal film resistors. The most common types were +/-20% C-comps, and +/-5.0% resistors were "ultra-precision".) My results in doing designs like this have been quite good. The difference between design values and measured values have been coming within a +/-2.5% margin of error (and that was with RCA plate characteristics and Sovtek tubes). If anything, distortion estimates have been running higher than measured.

As for the rest of what you suggest, why bother? All I use for VT design is the plate characteristic and the GIMP. Draw a loadline, and if you don't like the results, erase and draw another. It's NBD.

[7N7]

Originally posted by bigwill



...

I thought if I simulate tubes with their curves I don't have to learn the complicated maths
...

Miles is correct: there are no complicated maths, just arithmetic and Ohm's Law.

Even I can do it...

7N7

[rdf]

Originally posted by Miles Prower
All I use for VT design is the plate characteristic and the GIMP.

Same here. Another neat trick is to create a blank template in Gimp with a generic load line ticked off and numbered equidistant from a centre point. By scaling it to an imported curve layer it's a trivial matter to rotate, shift, and count off the most most linear region.

At a higher level though it brings up the valid question of why we still use curves at all. Though handy the technique was established when slide rules were standard lab instruments. We take a series of accurate data points, manually extrapolate them into curves, then manually estimate linear regions from the curves. It makes infinitely more sense, if albeit less fun, in century 21k to throw the data to a computer algorithm and let it spit out the answer. I read somewhere that when tubes were on the wane some labs were modelling curves in three dimensional clay. Software for the three dimensional visualization of data is everywhere.

Save for Spice (the data typically reconstructed from curves!) this is one area where things truly appear to have stopped dead.

[Robert McLean]

Bigwill :

If I were to import all of the data from a tube curve graph onto say, an Excel spreadsheet, and made some provision to approximately interpolate between each grid voltage curve, would it be possible to approximate the behaviour of the tube in software in various configurations?

Yes, absolutely. For example the CurveCaptor program that you will find references to on this forum does more or less that for triodes. For Pentodes I use spreadsheets of my own. The software used to model the circuit behavior is SPICE, you will find many references on the web. If you do a search here on SPICE you will find information on how to do all this.

Obviously it wouldn't take AC behaviour into account

This is not true. For "normal" frequencies, tube behaviour can be considered to be quasistatic, ie so far as the tube is concerned for any instant in time there is a voltage and current relationship defined by the tube characteristics. The fact that the voltages and currents may be different at some other point of time doesnt matter. "AC" and "DC" are human concepts, not tube concepts. We think 100KHz is way fast, the electrons in the tube couldnt care less. So AC signals can be analyzed graphically on the plate curves, and they can also be modeled in software, to the same degree of accuracy (or inaccuracy if you prefer ). [ by normal frequencies I mean those with periods much greater than the transit time of the tube., ie hundreds or thousands of megahertz ]


Now, others have mentioned that it is not really necessary to do this. You can design a tube circuit with graph paper and a slide rule. Who needs computer models. And there are limits to the accuracy of the models for the reasons mentioned, and more. This is all true. But I dont think that is the point. For me, I just like doing the math and playing with the spreadsheets and so on. It is another type of DIY, along with drilling holes in chassis ( chassises ??) and soldering wires to tube sockets and so on.

[richwalters]

Originally posted by rdf


Though handy the technique was established when slide rules were standard lab instruments.

I still use this nostalgic instrument. Accuracy is simply not required in building a tube amp.

richj

[bigwill]

Thanks for the interesting replies people.

The reasoning behind me doing this is simply because I enjoy tweaking things. I think it would be kind of neat to see roughly how the circuit would work before building it. The actual thing itself wouldn't be that difficult to program

If I come up with anything useful I'll post it on here!