This is something that I've been wondering for sometime. What was the nominal manufacturing tolerances, for tubes electrical specs ?

I've grown up with semiconductors, and certain specs for a transistor can be all over the place, and still be good. For example, take a 2N3904 transistor. Its hfe, at Ic=10mA, can be 100, for one device, but 300, for another, and both are good. This is in addition to the hfe varying with Ic, and temperature.

Take a tube, say a 12AU7, and transconductance spec, at Va=250V, is 2200 umhos. How close is the variance from tube to tube, with all those tubes still considered in spec ? Also, under normal operation, how long would this tube typically stay within good operation range ?

Almost as bad as semiconductors. Usually mins are given.

What's the top end on frequency of a transistor? In some cases frequency response is improved so replacements no longer work. Had power supplies go into oscillation by using later manufactured transistors. Still met spec.

Tubes do change more over their life than transistors. A new tube may have 1 1/2 or twice the transconductance spec. I would consider a tube bad below min published in manuals although still usable in circuits. Don't think gain drops off as this is determined by design. Emission goes down so a tube can no longer handle as much current.

If you look at a typical data sheet http://www.hebertech.com/view/174-rca-12au7-a.pdf

you will find Max values for parameters related to voltage, current, and power. Most other parameters are shown as "design center values" or "average values."

Tube design was done in the days of slide rules, not computers and SPICE models. Engineers actually built and de-bugged their designs. Tubes are essentially mechanical constructions, so it the element sizes and spacings are the same, the parameters should be "close." That might not apply completely in the case of cathode emission, however.


I worked at Westinghouse at the end of the tube era and I can recall tubes being "burned-in" for hi rel applications and (rarely) tubes matched for audio work.

Most semiconductor specs are given as "max" "min" or "typical." So a 2N3904 a min hFe, but no max hFE (except for one current value) on the Fairchild data sheet. Not all the values shown on a data sheet are actually 100% tested. They are "guaranteed by design." Fairchild has enough history to know that when a lot makes the hFE spec at 10 mA, it will also make the hFE spec at 50mA, so it will test hFE at one current, but not several.

Rich

Considering the fact that tube batches were randomly sampled, even some new tubes could leave the factory testing more or less than "100%" of design center. In the real world, it probably makes no difference and just because a new tube tests higher or lower than design center doesn't make it any better or worse. The exception, of course, is with higher power output tubes that need to be reasonably matched when used in pairs. Likewise with emission tests. A tube that tests 110% emissions may not last any longer or be any "stronger" than a tube that tests 70%. Chapter 30 of the Radiotron Designer's Handbook explains it.

The GM values in tube manuals are the bogey value of an ideal tube. Hickok and others adapted the +/- percentage in some testers which is generally around 10% during the high quality days but RCA was making or rebranding a lot of junk in the mid-late 60's and on. They were so all over the place that we switched to Sylvania and Telefunken after about 40 years of sticking to RCA in production.

Carl

Thank you for your responses. As it seems history also comes into play.

I have a bunch of 6AF6 eye tubes. from the packaging, some may have been made as far back as the 40s, while a whole sleeve are from the 70's. All of the older ones look well built. The 70's bunch all have cosmetic defects (under the glass), that I presume would have made them rejects in earlier years.

One note: Look again at the data sheet. When you see "2200umhos @ 250v" it is imperative you notice what the plate current was.

As the plate current is reduced, internal plate resistance increases and Gm drops. Mu might drop slightly, but doesn't change to the degree the others do.

So for most realistic operating points, data sheet Gm won't be the Gm exhibited in the circuit. I've heard it said before that data sheet "typical conditions" are really show-off numbers that make Gm look big and rp look small.

A very complete sheet might have a chart showing how Gm and rp change with different applied plate and grid voltages (which really mean changes in plate current). Use those or the plate curves to determine the "bogey value" of Gm (or rp, or mu) for your tube in your application.

It's also worth noting that the Hickok manuals state that when using the English scale, the bottom edge of the "Good" range corresponds to -20% below average for small-signal types, and 35% below average for power tubes. That seems reasonable, and about in-line with tolerance variations for typical electronic parts (i.e., +/-20%).

Final thought:
It is up to the circuit designer to take into account component tolerances and line voltage variations when working up a design. A good design functions as intended with normal variation of the components; a poor design is finicky and needs exact values for everything. Tight acceptance testing for a lot of parts eats up time and profit for a manufacturer.

Some designers (of, say, test equipment) use negative feedback to avoid performance changes as tubes age.

So the real test, as folks often repeat, is in the equipment you want to use the tube in. Except for shorts and maybe gas (which might damage the equipment), a tube which fails on a tube tester might still function well enough in your circuit. Maybe you don't get the far away stations, or as much volume at the speaker, with as low distortion. Or maybe portions of the circuit are wrapped in feedback enough that you hardly know a tube is wearing out until it is nearly dead.

If a tube is doing its job without disrupting anything else, it is good. In the era of marginal designs in radio and TV, sever al types and brands drew frowns from service people, but not all that many. Conservaive designs let many sets live on for decades with widely variable components.

If you have a dozen 6SN7 tubes in your stash,for example. take a look. Do they all work? Does it make sense? Which are better?

Get hold of Tomer, Getting the Most Out of Vacuum Tubes (Gernsback, 1960, available at Tech Books Online). He explains the tube specs game very clearly, with the "bogey" tube that is supposedly representative of the type, and the range of natural variation (bell curve) to either side. He also explains "selected tubes" and a wide range of tube failures. This is a MUST book for anybody who seriously works with tubes above the level of appliance operator.

Larry

When in doubt do a curve trace test, either homebrew or a TEK 570, plus a few others, which has reached obscene prices since the audioholes discovered it. I bought mine when they mostly couldnt even be given away.

Carl