Help understanding varying power output measurement from tube-swaps in a 70s Model T

[jamesmafyew]

70s Model T (not the super, the earlier version, although their power stages are basically the same) -- a client asked me to modify one to be able to use EL34s. I basically just added a fixed screen node to the power supply by adding another choke (5H/100R) and a couple of caps. He wanted the option of continuing to use 6550s, so I also added a switch to select either fixed-screen or UL operation. I made the bias supply variable so that it could accommodate 6550s/KT88s/EL34s. (Before you say "this is a bad idea," I told him that and he's taking his chances.)

The output transformer on the amp was an aftermarket Mercury replacement. I've seen this one before and it has a 2K primary. It's really big. I didn't measure the primary's DCR when I had it.

B+: 520V at idle, 480V at full power with all tube types. Screen resistors are 1K (not schematic 1.5K... I don't know why)

What I'm interested in is the power output difference:
Fixed screen:
KT88s - 144W
6550s - 148W
EL34s - 157W!

UL:
KT88s - 134W
6550s - 132W
EL34s - 112W

So my question is why would EL34s do more power in this circuit than 6550s or KT88s? Every time I've put 6550s or KT88s in a Marshall or Hiwatt style EL34 circuit (single rail with fixed screens) the power output goes up a bit, depending on the power supply. I put KT88s in a Hiwatt DR504 and it went from 58W clean to 72W clean with no changes to the circuit except for a bias supply mod. I can't remember a time where I saw EL34s do more power than 6550s/KT88s in the same circuit with no changes. I fully expected the EL34s not to do well in UL, but the fixed-screen results have me scratching my head.

This probably reflects my ignorance about load lines... but is the answer that simple -- just a different saturation voltage for the different tube types? Yet this amp was designed for 6550s. Is it an artifact of the fact that I'm testing into a non-reactive dummy load? Would results be different with speakers? Something about the UL OT design? Does the DCR of the OT primary affect it?

Please don't say "uh they're different tube types." Yeah I get that, but these results are not what I expected.

BTW, the bias for 6550s/KT88s was set to about 38mA per tube cathode current at idle. For EL34s, it was set to give about 30mA per tube idle. Obviously the power stage sensitivity was much higher with the EL34s, hit max power way earlier on the master.

(Another interesting and unrelated observation is that the 82k/120k PI plate resistors produce better balance in UL in this amp, but 82k/100k produces better balance in fixed-screen. So that explains that choice)

[Steve Conner]

EL34s have a higher "inner mu" than the other tube types. That means the screen grid is more sensitive, so for a given screen voltage the tube will pass more current. However, they also draw more screen current, so normally they drag their screen voltage down lower than the other tube types and it all washes out.

But, if you drive the screen from a high voltage, low impedance source, you can get a surprising amount of power, at the risk of burning out the screen from overdissipation. Just ask any old Marshall. And on a more serious note, check your Model T for screen glow in the EL34s when cranked.

[jamesmafyew]

Screens are not glowing in the EL34s at full power. I'm guessing it's because they sag down to a very reasonable voltage AT full power -- if the node is sitting at 475V due to the plate supply sag, then the actual screens are even lower.

So in the cases where a lower-voltage EL34 design produces more power with KT88s, it's because the power is limited by the tubes' maximum plate current into a given plate load? And then in this case, the limit for power is probably the supply?

[jamesmafyew]

I was playing around with this spreadsheet:
http://music-electronics-forum.com/t30159-2/#post270483

I found the DCR measurements for the OT in some old papers. 21 ohms CT to plate. Putting in the values that everything sags to at full power, the sheet accurately predicted the circuit's output power. For saturation voltage, I "guesstimated" from datasheets that the EL34 could probably pull down to just under 70V in this circuit, whereas a 6550 or KT88 would be more like 80V. That one difference perfectly predicted the power difference that I observed -- 155W for the EL34 vs about 145W for the 6550 or KT88.

The sheet also indicated that with the screens sliding to below 480V, the EL34s will probably be ok long-term in this amp.

[Tage]

This is a very interesting, I assume it may be the result of the EL34's true pentode construction?

Also I cant seem to find the spreadsheet you refer to, do you have a link James?

[jamesmafyew]

http://music-electronics-forum.com/a...eration-v2.zip

The above is the direct link to the attachment, which was in the forum post direct-linked in my earlier post.

[Steve Conner]

That sounds about right, I guess. But before sending it off I'd check for screen glow with an actual speaker load. Speakers have a big impedance peak at their bass resonance frequency, so if the amp is asked to put out a lot of power around this frequency, the plates will be unloaded and the screen dissipation will go up. For reliable operation with EL34s you have to build in some headroom to allow for this. So plug it into a 4x12, crank everything up full, bash out some power chords and see what happens.

You probably won't notice any difference between dummy load and speaker load until you start to overdrive the power stage. In the linear region, the NFB loop regulates out the difference.

[jamesmafyew]

That's a good point, thanks Steve.

Another thing possibility I hadn't really considered (but obviously should have) is that the amp will end up being run in fixed-screen mode with KT88s or 6550s. This seems like it could be even more problematic than the EL34 situation, because 6550s have a lower max screen voltage (something like 400-425V) and a lower max screen dissipation rating (6W) than EL34s. As far as I know, most new KT88s have a pretty high Vg2 rating but the screen dissipation max is still 8W like an EL34. I guess KT88s would be ok if EL34s are ok, but the 6550s are probably asking for trouble if not left in UL mode.

[J M Fahey]

Just a side note: although the numbers are impressive (and the EL34 results *look* counter intuitive), I'd lean towards the "safe" setting (say, increased screen resistors and such).
Wouldn't mind *at all* about "losing", say, 10 or 20W which anyway won't be noticed live, but making the amp 10X safer.

Personal choice if course.

[jpfamps]

My guess is that the loading is more appropriate for EL34s than either KT88s or 6550 in pentode/tetrode mode, and hence greater power output with EL34s.

Conventional wisdom is that 1.7k is "best" for EL34s at around these HT voltages because "that's what Marshall used" in their classic 100W amps. However the EL34 datasheets would suggest a higher loading will allow greater power delivery, although of course screen grid currents will be higher with the higher loading. Marshall may have considered this when choosing the lower loading, or simply misread the datasheet.

Regardless, early 70s Marshall used an HT around 520 VDC and usually measure around 130W clean.

The Sunn was designed to run KT88s and 6550s in UL, and as expected they both deliver more power than EL34s in this configuration.

[jamesmafyew]

And in fact, Marshall has used a 2k loading for all their 100W 'modern' amps since the JCM900 2100.

I was under the impression that often primary Z is lower than datasheets would indicate, because speaker impedance is much higher than the nominal rating through most of the audio spectrum. I'm sure my results above would change if I were testing into a reactive load.

[jpfamps]

And in fact, Marshall has used a 2k loading for all their 100W 'modern' amps since the JCM900 2100.

I was under the impression that often primary Z is lower than datasheets would indicate, because speaker impedance is much higher than the nominal rating through most of the audio spectrum. I'm sure my results above would change if I were testing into a reactive load.

Your point about reactive loading is good one.

The first JCM900s used 5881s, so a 2k loading would have been the "conventional" choice a la Fender.

I suspect we will never know whether the subsequent use of 2k loading in the JCM900 when Marshall reverted to using EL34s was a conscientious design choice.