Looking for EL84 Tone Comparisons by Tube Brand

[roberto]

Oh, wait. My attitude about the variability of humans is showing, isn't it?

Believe me, I really didn't notice.
Well, it would be unrealistic to negate that I've often noticed that my perception of the sound is triggered by many factors. Even if I try not to.
I remember once I was modding a Soldano SP77 just to boil down a bad day. I was trying a combination of stages (something like half Diezel and half Bogner XTC, but with totally different frequency cuts) and I wasn't happy at all with the results.
Then I saw my phone ringing and it was a friend of mine: "I'm at the door, what a sound!" and then he took the preamp and still has it.
Of course many times happened the opposite.

I'm a Mec. Eng. that has always been fashinated by electronics (I've a degree in Mechatronics plus two years specialisation in Mechanical Engineering applied to Food and Beverage industry), and tubes always fashinated me.
I have to admit that I've been triggered many years by the "tubes sound better" motto, and my love for pentodes (mainly EL84, partially EL34) was a good shield against other options. Then as always happens I had a new point of view (by making both worlds cooperate in some circuits) that moved my trigger. EG: even without going in AB2, power amp driven by a DC-coupled CCS-loaded mosfets goes to another level of definition IMO.

What could be the investment for the kind of equipment you have in mind to perform these tests?
Engineering apart, as that could be some sort of cooperation, hoping to have the possibility to contribute.

I'd interested in those numbers for myself, and I'd let people debate, at least on something new.

[pdf64]

...But to be labeled and sold as a "12AX7" it has to be within the minimum and maximum ranges of the component datasheet...

Not to query your general points, but apart from their limiting values, the tech info provided by tube manufacturers, even in the golden age, seems to be typical / nominal values for the various characteristics, upper and lower limits for those typical values being notable by their absence.
So the info seems to be more of a list of untoleranced nominal values, with nothing there by which a wonky but functional tube could be claimed to be non compliant to spec.

[AmpAwe]

Do any of you think it is true, or not true, as some tube retailers claim that they listen to and test every single tube and resell bad tubes to others at whole sell and sell the cream to "their" customers? Or is that too controversial of a question?

[martin manning]

The demand for testing and matching tubes after they leave the factory exists because manufacturers do not seem to be producing to the tolerances that were typical when tubes were more widely used. There are now a number retailers selling specially selected tubes, be they burned-in, matched sets, matched sections, or low-noise. I imagine nearly all of a retailer's power tube inventory could be sorted and sold as matched sets, but I do have to wonder what happens to the rejects from the section matching and noise tests. If they go back on the shelf to be sold as non-premium product, then the quality of that group will on average be lower. I really doubt that any of them would be sent back up the supply chain, unless they were completely non-functional.

Personally, I don't think that specially selected tubes sold by at least some of the larger retailers are tested very carefully, as I've been disappointed when checking the ones I have purchased using a curve tracer which I believe to be accurate. Smaller outfits that specialize in such testing will probably do better, but charge more, as it takes more time.

[R.G.]

In no particular order...

Tubes were sold both ways, with only typical values and with min/max - or, if the makers could get away with it, only a min or only a max. Consumer market tubes were probably specified as little as the makers could get away with. Tubes sold to the military or big commercial interests were given better specifications, and were probably selected units. The probable path would be to get in a big order for better specified tubes, and then to select from the main production line until you could fill the order. Selecting parts out of the main production run was the only way to get assured limits on parts before modern manufacturing made tighter tolerances common. You see the same in resistors. In the 1940s, 20% was the common, unspecified tolerance. 10% was marked with a silver stripe, and 5% was marked with a gold stripe, and still are. In the 40s and 50s this was done by selecting the middle out of production runs to sell for a higher price. In some times, not specifying a 10% or 5% part meant you could be reasonably sure that there would be NO part within 5% or 10% in what you got. Hey, it meets the spec...

Same thing applied to tubes. Remember that this was an era when manufacturing was crude. Machines were crude, materials were not as pure, and production spreads were wide. This is one subtlety of the old name for economics - the dismal science. Every western-world manufacturer would love to sell you whatever they have and not talk about specifications. They got into competing on specifications when governments/militaries started demanding it and wielding a BIG money stick to get it. This changed a lot when the Japanese leaned to manufacture well enough to send quality products to the western world at low prices. It made a joke of US automakers' insistence on flash over quality, and came very close to ruining several very large western manufacturers. As an engineer, I was extruded through several in-depth training sessions on "how the Japanese do it" in the 1980s. Quality had become a selling point.

One reason the tube makers could get away with only typical specs on their consumer grade stuff is that the huge quantities of tubes made meant that once the machines were tuned in and adjusted, tubes - which are primarily mechanical devices - respond well to being made in high quantity runs.

[martin manning]

As an engineer, I was extruded through several in-depth training sessions on "how the Japanese do it" in the 1980s.

Me too. W. E. Deming taught the Japanese statistical quality control beginning in 1950. It's all basically repackaged Deming.

[R.G.]

Do any of you think it is true, or not true, as some tube retailers claim that they listen to and test every single tube and resell bad tubes to others at whole sell and sell the cream to "their" customers? Or is that too controversial of a question?

Oh, sure. As the obvious - and ugly! - question.

Sadly, the sellers/retailers are humans, and humans have a huge capability to bend and rationalize bending the rules for personal gain.

It is possible that some retailers even do listen to every tube and sell only the best. A retailer that was a saint would test out the best and then grind the rejects into powder; I can't imagine that happening. The rejects are sold somewhere. Something very like this happened in the effects world. The searching for germanium transistors for Fuzz Face and other early pedals led to individuals and dealers literally scouring the planet for leftovers. Some of these had revered type numbers, like AC128 and NKT275, and this was valued all out of proportion to how they really performed. Dusty-bin collections were sorted either by listening, or by testing for gain and leakage, and sold for a premium. The others? Let's just say you probably won't like the ones you get "unsorted" on ebay.

Another issue is if they listen to every one - who says that what they think is "sounds good" is what you'll think is good. I can get behind measuring for noise - meters and oscilloscopes can tell you that one - but "tone"???

"Matched" is another issue. "Matched" in what sense? For power tubes, the first item matched had better be the plate current with a specified grid voltage, or the same thing said another way, the grid voltage to get a specified current. Get this wrong and "pairs" won't balance for audio, as well as offering the possibility of offset-current saturation in PP output transformers and maybe even one-sided thermal runaway. I doubt that matched output pairs get matched for anything else. Well, it's possible, but what exactly are the items matched other than bias? No one talks about that. As a side note, installing an independent bias setting for each output tube renders this kind of matched pair/quad/octet irrelevant.

If you are matching preamp tubes, what the devil do you match?

[martin manning]

Another issue is if they listen to every one - who says that what they think is "sounds good" is what you'll think is good. I can get behind measuring for noise - meters and oscilloscopes can tell you that one - but "tone"???

For preamp tubes, I suppose one could measure frequency response and THD with the tube in some standard gain stage circuit, or maybe just measure Cg-a as an indicator.

"Matched" is another issue. "Matched" in what sense? For power tubes, the first item matched had better be the plate current with a specified grid voltage, or the same thing said another way, the grid voltage to get a specified current. Get this wrong and "pairs" won't balance for audio, as well as offering the possibility of offset-current saturation in PP output transformers and maybe even one-sided thermal runaway. I doubt that matched output pairs get matched for anything else. Well, it's possible, but what exactly are the items matched other than bias? No one talks about that. As a side note, installing an independent bias setting for each output tube renders this kind of matched pair/quad/octet irrelevant.

I have to disagree there. Power tubes can be matched for both current (at some particular Va, Vg2, and Vg1), and Gm at that operating point. Separate bias trimers can equalize the static current, but not the dynamic response.

If you are matching preamp tubes, what the devil do you match?

Usually a dual triode tube is selected for matched sections, which might be desirable in a phase inverter.

[R.G.]

For preamp tubes, I suppose one could measure frequency response and THD with the tube in some standard gain stage circuit, or maybe just measure Cg-a as an indicator.

I'd go for that. Measuring frequency response is a perfectly valid and repeatable test. But the tone folks are not going to be happy with the definition that tone equals frequency response. If it is, there's no particular reason to measure the tube's frequency response, as the outside circuit is a vastly bigger contributor to frequency response.

I have to disagree there. Power tubes can be matched for both current (at some particular Va, Vg2, and Vg1), and Gm at that operating point. Separate bias trimers can equalize the static current, but not the dynamic response.

Again, that's correct. Matching for grid voltage at a single DC point is what I'd call a one-point match. Matching for an average gm at some other point would be a two-point match. Really matching output tubes would probably need overlaying plate curves.
My suspicion is that there might, possibly be one supplier of matched tubes that match output tubes for anything other than DC bias. Maybe - and I'd be surprised - there could be two. The standard of the output tube matching industry has to be a one point DC match, on economics reasons if nothing else. That's because it's easy to just measure the grid bias for a given DC current in a box of 500 power tubes, record every one, then sort in a spreadsheet and find which ones make "good enough" matches. That's selecting from points on a line. Adding a matching for gm means measuring both bias voltage and gm for each tube, then looking for two-way matching on what amounts to a plane. Matching gm at more than one point, gets you to a three dimensional matching problem and tossing in frequency response, noise, hum feedthrough and so on, like the word "matched" implies to the uninformed consumer gets seriously out of hand. As a matter of practical economics, that's only feasible if the raw unsorted tube supply is very consistent indeed. And if it is that good, matching is far less necessary in the first place. Looked at another way, every additional parameter to be matched reduces the yield of matched sets, which is the same as increasing the fallout rate.
So yes, you -can- match for a much larger set of parameters. But you can also match the effective gm of one-point matched tubes with parts outside the tube. One simple, direct way to do this is to let vary the phase inverter's output slightly per tube. I've done this once, as a matter of fact. The Workhorse amps had switchiable bias targets for 6L6 and EL34, and normal and "hot" for each of those. I made the obvious mistake and got a mixed pair in one test, and tried trimming the plate resistors of the PI. Worked just fine. I awarded it my personal 9.75 for "tone".

Usually a dual triode tube is selected for matched sections, which might be desirable in a phase inverter.

True, you could. However, my experience with PI tubes is that the plate resistors have a very much larger effect than any differences from tube to tube. There is an application where matched halves of a duo-triode would be desirable, and was actually done. It was for the tubes in the Philbrick triode based opamps from the era of analog computation with these beasts. I got to work with one of the Philbrick setups back in college - the Navy had donated some old test setups from their work on onboard gun aiming computers. In that case you really, really want matched halves for the input characteristics, although gain matching wasn't particularly of interest.

I'd love to see some real (and verifiable) data about what actually gets matched in matched tube sets.

[martin manning]

Matching for grid voltage at a single DC point is what I'd call a one-point match. Matching for an average gm at some other point would be a two-point match. Really matching output tubes would probably need overlaying plate curves.
My suspicion is that there might, possibly be one supplier of matched tubes that match output tubes for anything other than DC bias. Maybe - and I'd be surprised - there could be two.

I suspect you are right that most sellers of "matched" power tubes are matching wrt Ia only, but here's one outfit that does both Ia and Gm: https://www.apexmatching.com/how-apex-matching-works Antique Electronic/CE Distributors uses that service. I'll bet that there are other boutique tube sellers out there that will do the same.

But you can also match the effective gm of one-point matched tubes with parts outside the tube. One simple, direct way to do this is to let vary the phase inverter's output slightly per tube.

Ampeg's SVT (1970) included both a separate bias adjustments for each side and a phase inverter plate load adjustment to balance the output stage (the procedure is described on the schematic). Dumble included a PI trimmer in most if not all of his designs, and I believe all the Dumble clone and derivative amps have retained that feature. IIRC, Fuchs amps have separate bias adjustments and a phase inverter trimmer. The sections in some dual triodes can be so far off that an excessive difference in load resistance is required to balance the PI output, and I would think that starting with a tube in which sections are at least close is probably better in the end.

[R.G.]

Matching for grid voltage at a single DC point is what I'd call a one-point match. Matching for an average gm at some other point would be a two-point match. Really matching output tubes would probably need overlaying plate curves.
My suspicion is that there might, possibly be one supplier of matched tubes that match output tubes for anything other than DC bias. Maybe - and I'd be surprised - there could be two.

I suspect you are right that most sellers of "matched" power tubes are matching wrt Ia only, but here's one outfit that does both Ia and Gm: https://www.apexmatching.com/how-apex-matching-works Antique Electronic/CE Distributors uses that service. I'll bet that there are other boutique tube sellers out there that will do the same.

Good to know. So the guess was pretty close - one advanced matcher, and maybe one or two more.

But you can also match the effective gm of one-point matched tubes with parts outside the tube. One simple, direct way to do this is to let vary the phase inverter's output slightly per tube.

Ampeg's SVT (1970) included both a separate bias adjustments for each side and a phase inverter plate load adjustment to balance the output stage (the procedure is described on the schematic).

Yep. Independent bias was an old (and obvious!) technique before I ever knew what a tube is. And PI balancing is also. But the point is that the PI tube outside the output tubes can be used to make up for any non-grossly mismatched AC gain characteristics of the output tubes. That then begs the obvious question... well, what about the matching needed in the (presumably duotriode) sections in the PI?

The sections in some dual triodes can be so far off that an excessive difference in load resistance is required to balance the PI output,

This is a place I'd really like to see some hard, statistically valid data on. I have not run onto decent surveys of, say, 12AU7 or 12AX7 sides matching. I know of some very old studies of individual lots from military uses, but that's completely unrelated to what we as a consuming public could get today. Another of my uncomfortable suspicions is that we could find a number of grizzled old tube techs and amp builders who could tell us stories about that lot of tubes they got back in the blizzard of ought-75 where all the halves were matched - to halves in other tubes.

Do you happen to know how far off so far off is, in a statistical sense? I have run into the worst possible mismatch, but having one half dead probably doesn't count, nor is it representative.

and I would think that starting with a tube in which sections are at least close is probably better in the end.

So would I. But we are exploring how bad bad has to be before it's intolerable, and whether selection and matching is the most practical and economical method for handling it.

er... I think...

[AmpAwe]

WOW! - lot of replies - I have been offline, will catch up on this thread, hopefully tomorrow - appears very interesting,
THANKS!

[martin manning]

I have not run onto decent surveys of, say, 12AU7 or 12AX7 sides matching. ... Do you happen to know how far off so far off is, in a statistical sense?

I have traced quite a few 12AX7's, but I don't have a database of the results. Anecdotally, I can say perfectly matched sections is a rare thing. Typically the plate current will differ by 10-20% at the data sheet spec point (250V Va-k, -2.0V Vg1).

[R.G.]

I have traced quite a few 12AX7's, but I don't have a database of the results. Anecdotally, I can say perfectly matched sections is a rare thing. Typically the plate current will differ by 10-20% at the data sheet spec point (250V Va-k, -2.0V Vg1).

That hits me as about right for today's market. 10%-20% is actually pretty tight as non-monolithic, non-laser-trimmed parts go in the electronics biz.

In the tube Golden Age, EEs were perfectly comfortable with designing with only typical specs and expecting the results to be good. After all, 10% was a close tolerance resistor up through the early 1960s. There was a whole lost generation of EEs when transistors made it necessary to design for active devices that might vary by three or four to one if not pre-selected. They just could not cope with such wildly variable devices. Even in the early 1970s in my circuits courses, coping with variability was a hidden topic in most circuits classes.

It's pretty trivial to design a one-point triode-half matcher. A socket with matched plate, cathode, and grid leaks for each half, a regulated plate and heater supply, and some A to D capability would do it. Shoot, adding a signal source and a way to digitize plate signal would let you detect AC gain per half automatically too. Add a label maker (to preserve the ID of each tube) and a little programming in BASIC and you're ready for burned fingers swapping tubes.

Here's a link to the Philbrick K2-W datasheet, just for grins.
http://www.philbrickarchive.org/k2-w_refurbished.pdf

[AmpAwe]

If you are matching preamp tubes, what the devil do you match?

In the case of multichannel tube amps that would be the multi channels.

I think that I might be learning enough to soon measure the operational values of my https://www.uaudio.com/hardware/4-710d.html at the tube pins, to manage to change the tubes in it. It's likely fixed bias such that you can just swap the 4 12AX7 tubes without adjustments. Its manual says it must be done by UA techs by sending the unit in. I asked them to provide me the procedure and they said - "NOT gonna happen".

I do wish the tube retailers would define what they do for matching, one I buy from does give some information about how they do their matching but it is very general - they do invite the buyer to ask them for more info - I think I will give that a try.

For the most part I have had great experience buying. With notable exceptions. I bought one $50 12AX7 and it was so amazing it went in V1 of my favorite GTR amp, and so I laid down another $50 and that one was dramatically different with a strong but very early huge distortion, where the first purchased one was super clear tonally balanced and super strong. I should have evaluated before a year went by - but didn't

the post by R.G. » Sun Dec 06, 2020 4:02 pm - make A LOT of sense.
All this info make the voodoo of tube swapping make much more sense