2xKT88 in UL Twin?

[potatofarmer]

Okay, so I've been looking at retubing my '81 "ultralinear" twin, and the insane screen dissipation pdf64 measured a couple years ago ( http://music-electronics-forum.com/t18237/ ) has me second-guessing putting a new quad of 6L6GC's in here. What information I've been able to find about ultralinear/distributed load operation seems to agree that for reliable operation you have to keep the screen voltage (and by extension, the plate voltage) under the Max Vg2 listed on the datasheet. Some current manufacture tubes state 500V is acceptable on the screens, but the historical spec is 450V and I trust that number a little more.

Currently I have 2k/5w screen stoppers on the output tubes, which may not be enough, but it's better than nothing. Also it appears that the sockets have been replaced at least once in the past, unless Fender decided to use a stack of washers to hold the bear trap and socket retainer on the outside of the chassis.

I should also mention the "vintage value" of this amp is roughly zero. The cab has been water (or at least humidity) damaged, there are rust spots on the chassis, the grille cloth mounting frame is cracked in several places, and so on and so on.

Anyway, I'm looking at these plate voltages (~510V loaded), the odd screen tapping point (as measured on a similar OT by bob p: http://music-electronics-forum.com/t33707/ though I should verify this on my OT), and I'm thinking... why not KT88s?

Pro:

• better able to handle high plate and *screen* voltage in UL for increased reliability
• screen tap point correlates better to KT88 than 6L6 - closer to actual UL than just DL (see graph of power vs winding % of screen tap here: ultra-linear ) - will still have higher Zout than triode mode (oh darn) but will reap nearly all the other benefits of UL operation
• KT88 have a better reputation for UL performance
• silverface PI already optimized(-ish) for lower grid leak load
• potential increase in power output, if PT can deliver current
• because it's interesting

Con:

• cost of new tubes
• danger(?) to iron
• extensive work for little(?) benefit
• breakup even later
• potential increase in power output, if PT can deliver current
• I'm not actually insane

Considerations:

• max DC resistance @ >35W = 100k
  
  silverface twin = 47k/68k = no problem
• heater draw
  
  2x KT88 = (1.6A ea)(2) = 3.2A
  4x 6L6GC = (0.9A ea)(4) = 3.6A
• OT Primary impedance
  
  4x6L6GC in a Fender = typically 2k, possibly 2.4k in "UL" amp
  use speaker mismatch to get up to 4k-4.8k
• OT primary DC current at idle
  
  2x6L6GC = ~60-80 mA per side
  1xKT88 = (.7)(42W/510V) = ~57mA -> "design max" is 35W so ~48mA per side max expected idle current
• biasing
  
  balance circuit crucial (will it be enough?)
  increase neg. bias voltage
  add bias adjust pot (going to do this anyway)
• sockets
  
  replace bear traps with spring retainers
• fit/clearance
  
  in outside sockets, shouldn't be a problem (based on steelguitarforum thread)
• screen grid stoppers
  
  already at 2k/5w, change to 1k/5w

So there are a few things I know I need to do first - verify the primary impedance of the OT, verify the screen tapping points, design/modify a better bias supply - but I'm wondering what my blind spots are here. Or if I'm doing something wrong or anything like that I'd greatly appreciate the feedback!

For the forseeable future this will just be an exercise in wasting time and paper, but by the time I can free up enough cash to try this I'd like to be sure about what I'm doing.

I should mention in addition to adding larger screen stoppers, I've also removed the pull-boost and disabled the global NFB loop (well, I put it on the pull-boost switch so the "pull to sound worse" functionality is still intact) and done a couple minor preamp tweaks too. I currently have the inner pair wired as triodes, and I admit it sounds pretty good like that, but I'm still worried about long-term reliability. Steve Conner metioned once (I think in a thread I linked to here) that doing so may lower the screen dissipation for all four tubes, but I don't quite see how...?

Thanks!

[mhuss]

The idle bias will be a bit different, but otherwise it should just work. If the output impedance doesn't match up, you'll just have less output power.

An ultralinear topology is a lot less rough on the screens than a typical static screen voltage one. Due to the autoformer action of the OT, the screen voltage mostly stays less than the plate voltage. When the screen voltage exceeds the plate voltage is when the screens carry excessive current and really heat up. Note that in all UL implementations (mostly hi-fi) the static screen voltage is almost identical to the plate voltage, and countless Dynaco and other old hi-fi amps worked for decades at higher than normal screen voltages.

[potatofarmer]

An ultralinear topology is a lot less rough on the screens than a typical static screen voltage one. Due to the autoformer action of the OT, the screen voltage mostly stays less than the plate voltage. When the screen voltage exceeds the plate voltage is when the screens carry excessive current and really heat up. Note that in all UL implementations (mostly hi-fi) the static screen voltage is almost identical to the plate voltage, and countless Dynaco and other old hi-fi amps worked for decades at higher than normal screen voltages.

Sure, I've seen reference to that. Those vintage hi-fi amps weren't driven into overdrive though, and while I can't imagine that'll be a frequent occurrence with this Twin, I still want to design with it in mind.

(Looking at the KT88 datasheet- can't find a 6L6GC sheet with an UL graph) Comparing the graphs of plate current, at low plate voltages the screen current starts increasing earlier yet more gradually in ultralinear mode than the brick wall you hit in pentode mode. It looks like for any plate voltage below 250V, the screen current is going to be higher in ultralinear than in pentode mode, but that's completely not a problem because as far as AC is concerned the screen voltage is going to be around half of the plate voltage. This is of course for the 43% taps, which is where anyone with any sense operates distributed load beam tetrodes... but I think then at a more "pentode like" tap, the AC screen voltage would be higher than a 43% tap, and correspondingly the screen current will be lower, probably yielding the same dissipation. Interesting how distributed load operation makes(lets) the screen suck down plate current like it's a true pentode.

In other words, I need to start figuring out how to determine the screen dissipation, at least on paper. "If it's fine at idle, it's fine" seems like a handy rule of thumb, but I don't know the rationale behind that, and in this amp's case it could be argued that the idle conditions are not fine...

[mhuss]

'Tis true, the olden days (and more modern) spec sheets really only consider linear operation. Traditional theory and equations will not be reliable for this reason. The best thing would be to pop a pair in there and measure the actual values under normal and clipping conditions.

[potatofarmer]

'Tis true, the olden days (and more modern) spec sheets really only consider linear operation. Traditional theory and equations will not be reliable for this reason. The best thing would be to pop a pair in there and measure the actual values under normal and clipping conditions.

I think a pair of KT88s would be operating in a sufficiently safe region that I wouldn't have to dig further, but I'm still questioning whether this circuit could work with a quad of 6L6GCs. If I'm worrying for nothing, all the better.

As far as determining screen dissipation goes, I just found this: http://www.pearl-hifi.com/06_Lit_Arc...issipation.pdf

...though I wish it included the "complex integration" it dismisses in favor of sticking a pair of ammeters into the screen supply. A voltmeter across the screen stopper would make a good approximation of the thermocouple milliammeter, I'd wager. What the hell, I've got some harbor freight specials I wouldn't mind popping in the name of science.

Though, you're right, this still wouldn't help to determine what's happening during clipping.

[mhuss]

A true RMS meter would be able to accurately measure the AC even clipped/distorted. :-)

[jazbo8]

as far as AC is concerned the screen voltage is going to be around half of the plate voltage.

In other words, I need to start figuring out how to determine the screen dissipation, at least on paper. "If it's fine at idle, it's fine" seems like a handy rule of thumb, but I don't know the rationale behind that, and in this amp's case it could be argued that the idle conditions are not fine...

That's not how UL works, please refer to the graph below, note the voltage scale on the bottom for Vg2 - it is not 1/2 of the plate voltage... As for the screen dissipation, you just have to make sure that the total dissipation, i.e., Pa + Pg2 < 40W (shown by the red line below). Usually for Class AB, 70-80%, so 28W-36W.

[Mick Bailey]

A KT88 running upside-down puts a lot of heat into the base - worthwhile researching the mounting position first. Usually you'd mount them vertically upwards, or horizontally.

[Pedro Vecino]

I can not find the sense. A quartet of Sovtek 5881 (or the 6L6 Sovtek model) fully support these voltages and have the advantage of sounding good with a conservative bias adjustment.
One problem with this model (the SuperTwin too) is that the hum is very dependent on the current balance in each tube plate to compose a perfectly symmetrical set. For me the solution (drastic) is to install a choke and a new node with his filtering. If one can not live with 500 volts on the screen grids, you can install a 100V zener diode in each socket to subtract it.
Mechanically a quartet of Sovtek 5881 in that environment is much stronger than any KT88 pair installed upside-down.

[potatofarmer]

That's not how UL works, please refer to the graph below, note the voltage scale on the bottom for Vg2 - it is not 1/2 of the plate voltage...

Okay, yeah, I need to do more reading about this. The bit that stuck in my head (albeit incorrectly) was this:

The "Ultra-linear" configuration avoids the conventional conflict between plate and screen voltages by creating a voltage divider network through the output transformer primary to AC earth (transformer centre-tap), ensuring the screen voltage tracks and thus always remains, both below and proportional to the plate signal voltage. By this configuration, the screen is intended to be prevented from exceeding its power dissipation rating.

This statement is subject to the Screen Grid always being operated within its Rated DC Screen Grid voltage and at a DC voltage less than that of the Plate.

Applied plate (B+) voltage for ultra-linear connection should never exceed Grid 2 rated voltage, so standard ultra-linear configuration is only suited to tubes designed for audio applications having a Grid 2 rating approximating either the plate voltage rating (or actual applied plate voltage if less than rated maximum).

ultra-linear

As for the screen dissipation, you just have to make sure that the total dissipation, i.e., Pa + Pg2 < 40W (shown by the red line below).

So we don't have to worry about the screen dissipation? Is this true for all distributed load conditions, that is, independent of where the screen taps are?

[potatofarmer]

I can not find the sense. A quartet of Sovtek 5881 (or the 6L6 Sovtek model) fully support these voltages and have the advantage of sounding good with a conservative bias adjustment.
One problem with this model (the SuperTwin too) is that the hum is very dependent on the current balance in each tube plate to compose a perfectly symmetrical set. For me the solution (drastic) is to install a choke and a new node with his filtering. If one can not live with 500 volts on the screen grids, you can install a 100V zener diode in each socket to subtract it.
Mechanically a quartet of Sovtek 5881 in that environment is much stronger than any KT88 pair installed upside-down.

I actually have an (old, old) quad of those. I tried them briefly but with the bear trap retainers I didn't want to leave them in too long. So I could buy the variant with the full base...

But do they support these screen voltages? There still aren't any real numbers about these anywhere, apart from the reports of the plate dissipation being substantial, Ken Gilbert being unable to get them to redplate by idling them at 5,000W and hitting them with a blowtorch, etc. So I'm wary about relying on these, particularly if they can't be overdriven without facing the wrath of ~20W of screen dissipation.

The "output matching" bias balance control is pretty effective at eliminating hum, even with unmatched sets.

Zener diodes are clever, but then we're throwing away a decent amount of power as heat, which means big zeners and heatsinking / drilling the chassis.

The question of mechanical stability is a good point. I'm not familiar with how well those spring retainers work, and this is a combo so the tubes are going to take a pounding.

A fan might be a good addition regardless of tube type; the glass of the 6L6GCs puts out a surprising amount of heat. I'm currently missing the back panel and I'm thinking that's a good thing. Also the increase in tube fluorescence from idle to moderate or full output is also striking.

I know this amp's been around for 30-some-odd years and hasn't exploded or disintegrated into a pile of dust yet. I may be worried about the screen dissipation for nothing - for the KT88, Vg2max gains ~50V in UL connection. What's the rationale behind this, and would it apply to other beam tetrodes as well, or to other distributed load connections besides the ideal 40-ish percent?

[jazbo8]

Okay, yeah, I need to do more reading about this. The bit that stuck in my head (albeit incorrectly) was this:

"ensuring the screen voltage tracks and thus always remains, both below and proportional to the plate signal voltage."

The key words here are "plate signal voltage" and "proportional" - so it is not the DC voltage, nor 1/2.

So we don't have to worry about the screen dissipation? Is this true for all distributed load conditions, that is, independent of where the screen taps are?

Well, you still do, without going through all the details, in general, if you operate the amp well below the 40W limit, and do not deviate too far away from the recommended plate load, you should be ok. If you do a search on screen grid dissipation, I think you will find that there were a few threads on the subject.

[potatofarmer]

"ensuring the screen voltage tracks and thus always remains, both below and proportional to the plate signal voltage."

The key words here are "plate signal voltage" and "proportional" - so it is not the DC voltage, nor 1/2.

Yeah, I knew and indicated that we weren't talking about DC conditions. Not sure where the "1/2" came from though.

Well, you still do, without going through all the details, in general, if you operate the amp well below the 40W limit, and do not deviate too far away from the recommended plate load, you should be ok. If you do a search on screen grid dissipation, I think you will find that there were a few threads on the subject.

Sure, and I understand screen dissipation in pentode mode, but distributed load seems to be a special case. I'm sure this has been written about before, I just need to do more reading.

[bob p]

Interesting thread.

When I took the OT measurements on my 6x6L6 Super Twin Reverb, the numbers that I obtained put an end to my habit of referring to the Super Twin as an UltraLinear amp, as it's definitely not designed to operate in what we normally think of in HiFi terms as UL mode. It's definitely a distributed-load (DL) implementation that is quite close to operating in full pentode mode. The net advantage of DL operation isn't immediately obvious, though after doing a fair amount of reading it does look like Fender did this just to maximize power output.

It's important not to extrapolate too much information about 4x6L6 Twin Reverbs from my measurements, because after all, I did take measurements off of a 6x6L6 OT from a Super Twin Reverb. There is no guarantee that my conclusions for the 6x6L6 Super Twin Reverb can be extrapolated over to a 4x6L6 Twin Reverb. I don't know what the turns ratio is on the Twin Reverb OT. That would be a very interesting datum to have.

I've always thought that it would be interesting to repeat the measurements on an OT from one of the "UltraLinear" Twin Reverbs. Chances are that we'll find that all of those "ultralinear" Fender amps aren't anywhere near close to operating in UL mode, and that they're probably also DL mode amps.

PaulP, given how easy it is to take these types of measurements, I'd like to ask you to take them. I'd really like to see the results of some OT measurements for your "ultralinear" Twin Reverb. This would be a good opportunity to determine whether those "ultralinear" Twins actually operate in UL mode or not. I think that more than likely the tests are going to show that your "UL" Twin Reverb is really a DL-mode amp. If that is the case then everyone who has been calling these amps "ultralinear" for so many years has been making a wrong assumption about the OT just because of the superficial appearance of the wiring topology.

My measurements suggest that it's inappropriate to refer to the Super Twin Reverb as an "ultralinear" amp. It would be interesting to see some similar measurements off of an "UL" Twin Reverb. It may be that we need to put an end to a bad habit in referring to these amps as "ultralinear."

[potatofarmer]

Measured the OT. My "line out" jack came pre-disconnected, so that was fun to notice finally.

4 ohm tap:
in: 14.44 V
Brn: 171.7 V
Brn/Yel: 21.47 V
Blu: 171.7 V
Blu/Yel: 21.47 V

(21.47 V)/(171.1 V) = tap at 12.5% of the winding

(171.7 V)(2)/(14.44 V) = 23.78:1
(23.78)^2 = 565.5:1
(565.5)*4 = 2262 -> Ra-a = 2.26k

8 ohm tap:
in: 14.58 V
Brn: 122.4 V
Brn/Yel: 15.32 V
Blu: 122.4 V
Blu/Yel: 15.32 V

(15.32 V)/(122.4 V) = tap at 12.5% of the winding

(122.4 V)(2)/(14.58 V) = 16.79:1
(16.79)^2 = 281.9:1
(281.9)*8 = 2255 -> Ra-a = 2.26k


So there we have it. Did I mention the KT88 datasheet has a sample Ra-a of 4.5k for ultralinear connection?
I double checked on the 8R tap because I don't have a true RMS meter, though a quick check at the peak-to-peak as measured on a scope agreed with the reading on the meter. Also the voltage supply is coming from an old Radioshack prototyping unit, AC power, DC power, signal gen, etc. Looks like it worked fairly well, though I wouldn't put any sort of guarantee on the precision I implied (3 sig figs, hah).