Ultralinear vs Distributed Load in the Late 70s through Early 80s Fender Twin Reverb

Okay, I kind of played it fast and loose on the technical side with my last two updates concerning the "Ultralinear" Twin.

If you'd like a primer written by someone much more well versed in the effects of connecting the screens to taps on the OT, read this: Oestex - Ultra-Linear

Say we're designing an amp, and we want to connect the screens to taps on the primary side of the OT. If we connect the screen taps to the OT center tap (0% of the primary turns), we get pentode operation, albeit with no ripple filtering. if we instead connect the screen taps to the plate taps (100% of the primary turns) we have triode operation. Anywhere in between we get what is known as "distributed load" operation, which has performance somewhere in between pentode mode and triode mode.

Pentode mode has higher output, but also higher distortion. This is how nearly all of guitar amps are typically connected. Certainly all amps you can buy in the store. Many guitar amps have a "half power" switch, sometimes properly labeled as a "triode/pentode" switch which allows the user to pick which mode the output tubes are connected in. These switches aren't very popular with users (though they have some fans) partly because the loudness doesn't seem to drop to half (pesky logarithms!) and partly because the tone is much duller in triode mode.So, I'm sure the average guitarist has some familiarity with amps with this switch. To summarize for guitar purposes:

Pentode:

• Higher output
• Brighter tone 
• Crunchier distortion
• More distortion

Triode:

• Lower output
• Darker tone
• Muddier distortion
• Lower distortion 

Well, friends, there's a whole spectrum in between 0% and 100%. Someone back in the 50s (or possibly the 30s, there are disagreements about who/when) figured out the optimum ratio for balancing high output and low distortion and they decided to call it ultralinear. It turns out (hah, "turns") that the optimum screen tap turns ratio for hi-fi is in the ballpark of 40%, though the optimal tapping point is different for every power tube. At this magical percentage, output actually increases (slightly) over typical pentode connection, the amount of distortion decreases to something similar to triode connection, and the resulting tone is somewhere in between.

Brilliant stuff if you're building a stereo. We're building guitar amps; who cares? Cut to CBS/Fender in the late 70s.

Some clever engineer realized that they could slightly increase the output of their flagship Twin Reverb by using an ultralinear OT. Fender, even in Leo's day, was always about maximum clean power, so this makes a lot of sense. CBS (and Leo, admit it) also really liked pinching pennies, so the prospect of saving the cost of a choke and a filter cap must have played into the decision. To polish it off, they upped the plate voltages (and by extension the screen voltages) to 500V and blam: the Twin is now putting out 135W. At least by guitar math. Remember that the 100W Twin was identical to the previous 85W Twin. It's pretty easy to move the goal posts when measuring output power, especially if you're selling amps to musicians.

Anyway.

So it's been largely assumed that CBS/Fender used the typical 40% taps for the balance of high power and low distortion. But they didn't!

I measured the turns ratios in my Twin's OT and here's what I found:
Ra-a = 2.26k
Screen tap at 12.5% of the winding 

Wild, eh? Pretty far from 40% of the turns. Here's a graph of the power, distortion, and output impedance of various tapping percentages for a 6L6GC:

Table courtesy of jazbo8 and bob p over at M-E-F

 The striped box indicates "ultralinear" conditions, and the green box indicates where Fender put their taps.
So we're basically still in pentode mode! Huzzah, the "Ultralinear Twin" is really pretty close to just being a Twin!

As we can see, the power is slightly less than true-pentode connection (yeah, yeah, these are "beam tetrodes" but they're wired up like pentodes so go stuff it), the distortion is a hair lower, and the output impedance is lower as well. Those last two, well, just think of them as being effects of the NFB that you get from connecting the screens to taps on the OT. Needless to say, we absolutely do not need an additional global NFB loop, the kind you typically see on guitar amps.

So how is fender making more power if the screen tap they chose (or really any screen tap for the 6L6GC) indicates it should be putting out less power than a typical pentode-connected power amp?

That's where the increase in plate voltage comes in.

Now, Fender has always pushed the voltage ratings of tubes, and it's fair to point out that voltage itself isn't what kills tubes, it's how much current is flowing at that voltage which kills tubes. Literally, W = V*A. The typical solution is to include a screen grid stopper. When the plate voltage dips below the screen voltage, all those electrons that would have loved to keep whizzing by instead are drawn to the screen grid and screen dissipation increases dramatically. Putting a resistor on the screen grid means as the current increases through the screen, the voltage at the screen will decrease. Good ol' Ohms Law n'at. This is a huge concern at overdrive, when we're going to slam the plate voltage as high as it can possibly go and then back to as low as it can possibly go.

In a typical pentode-connected power amp, all the current goes to heating up the screen grid; it's wasted. In this distributed load connection, some of that current goes to the load - it makes more output - so the screen dissipation is actually staying constant. The formula here is P = EI - ei, where uppercase 'E' and 'I' are the DC voltage and current, while lowercase 'e' and 'i' are the AC voltage and current.

source: http://www.pearl-hifi.com/06_Lit_Archive/02_PEARL_Arch/Vol_01/Sec_2/100_UL_Screen_Grid_Dissipation.pdf

Pretty neat, huh? Basically the rule of thumb for DL operation is "if the screen is fine at idle, it's fine up to full output." The question I have to ask though, is what happens at overdrive? This is how guitar design is still something of an uncharted frontier; overdrive conditions are largely dismissed in the golden-age literature (why the hell would anyone want to make distortion...?!). To meet this "if it's fine at idle..." condition, the screen voltage (and thus the plate voltage; with regards to DC they're going to be nearly the same) needs to be under the maximum listed on the datasheet. While there are some current-production tubes which list Vg2max as 500V, the historical standard is 450V, and that seems like the more reliable number.

Back to the guitar world, the screen dissipation at overdrive for 6L6s in Fender's distributed load experiment has been measured at ~20W. That's just a hair bigger than the 5W that they're rated for. So it seems fair to assume that these amps only don't eat tubes like a Mesa-Boogie would like popcorn because they're not routinely driven to overdrive. Reportedly increasing the screen stoppers to 4k7 can save the screens, but the output and tone suffer heavily.

To me, all this adds up to a poor tube choice for the iron. Sure, we could add a variety of bandaids to try and get the plate and screen voltages into the safe region of <450, but they all involve throwing power away as heat, which just seems stupid.

Personally I think the best solution is to run a pair of KT88s in the outer two sockets and double the speaker load (8R into the 4R tap, or what have you). KT88s will eat this plate and screen voltage for lunch, and at 12.5% they'll actually be closer to the ultralinear region:

KT88 distributed load characteristics. Source: oestex link above

Going off the datasheet values for a pair of KT88s with B+ = 510V and Raa = 4k5, we're looking at 100W of output maybe. Probably a hair less, something in the 85-90W region, which is preferable. The bias voltage will have to be ~10-20 ish volts more negative, so the headroom is going to certainly increase. Then again, since I removed the global NFB loop, the headroom is likely to be less than it was in stock condition. Also the tone is likely to change, but we'll see how much.

So, this is my preferred option because it requires the least adaptation of any tube change I can think of. There are other options, of course. So, assuming we want a 30 year old amp to make it last to 60 and beyond:

• Run the amp with the stock quad of 6L6GCs, preferably the most rugged you can find. This may be the Sovtek 5881/WXT whatever designation they're calling it now. Put in the largest screen stoppers you can tolerate and don't let it get too loud. THIS IS STUPID LOUD IS AWESOME TINNITUS FOR ALL
• Run the amp with a quad of 6L6GCs in pentode connection. This will involve adding a new power supply node, and if you want to stay all Fendery about it a choke and a new filter cap. If you don't want to try and stuff a choke on this crowded chassis a resistor will work fine. Looking at the Ra-a of the OT, you probably want the screen voltage to be somewhere in the 300-350V ballpark. If you connect the choke to the B+ node that's feeding the plates, you're going to want at least 1K resistors on the screens.
  
• Run the amp with a quad of KT88s or 6550s in distributed load connection. If a quad of KT88s will even fit, they're way too close to each other for safe heat dissipation. You'll need to add a separate heater supply for the preamp tubes. KT88 screens are more rugged than 6550s, so for distributed load connection these are preferred.
• Run the amp with a pair of KT88s or 6550s in pentode mode. You'd need to lower the screen voltage considerably for this option. Fortunately there's a nice node right between the stacked reservoir caps (where the PT CT connects). And yes, you can draw current from here without disrupting the voltage balance of the stacked reservoir caps! Check the 'evil twin' schematic; this is where Fender connects the plates for the 25W low power node. 
• Run the amp with a pair of KT88s or 6550s in distributed load connection. Double the speaker load. The heater current draw is slightly less than the quad of 6L6GCs it was designed for, so no problem there. KT88 screens are more rugged than 6550s, so for distributed load connection these are preferred. 

 Any change to KT88 or 6550 will require these mods:

• You'll almost certainly have to modify the bias circuit to safely run these tubes. You'll need a raw bias voltage adjustment and a balance adjustment as a minimum; independent bias pots for each tube are preferable.

•  Pin 1 will have to be rewired and connected to ground.  

•  The 'bear trap' tube retainers will have to be replaced with spring-type retainers.

•  The sockets may have to be rotated - check the Genalex datasheet for details. 

•  Additional cooling (via fan) is a good idea, possibly necessary. 

You know, I haven't looked into EL34s yet, because I certainly don't plan on adding an auxiliary heater transformer. Plus I'm a beam tetrode kinda guy. I'll leave the research into these and KT77s for someone else.

I am accepting donations at this time for funds to get a pair of KT88s. :)

Seriously, AES?

A COMPARISON OF CURRENT PRODUCTION 6 L 6 GC TUBES

CE Distribution / Antique Electronic Supply / Amplified Parts just put out this PDF where they plotted the frequency response of a single tube of each of the 6L6 family tubes they sell. 

 Okay, so this looks like a really useful document initially. They put a bunch of tubes on a tube tester, picked the average one, plugged it into an amp, plotted its frequency response, and let a guitarist subjectively describe the tone.

That last bit is a little like asking a mechanic what oil you should put in your car and then asking a guy who likes to drive how the different oils taste on bread, so I'm going to try to avoid commenting on "sizzling leads" versus "balanced and tight." If all you want is lurid prose about how many orgasms each tube brought the guitarist, there's no need to bring a tube tester or frequency plotter into this.

So ignoring the poet, we have the manufacturer's rated specs on the left. These are pretty much crap, as modern manufacturers largely just copy whatever was on the original data sheets. The maximum plate voltage is just a suggestion, let's be honest. Dissipation kills tubes, not voltage. And do we expect guitarists to know what their screen voltage is? Considering how common screen grid failure is, it would've been nice to include maximum screen dissipation buuut it doesn't matter anyway because the manufacturer's numbers are junk.

Back In The Day(tm), for a tube to be sold as a "6L6GC" it had to meet certain specs for plate dissipation, transconductance, envelope size, etc. No one cares about this any more, so it's understandable that there's some variation in rating and size from one manufacturer to another. Not to mention some of these tubes aren't even 6L6GCs, but Russian military tubes which were designed to be clones of 6L6GCs. That's another whole bag of worms, and it's fairly irrelevant, but it is interesting that some New Sensor 6L6GCs can take up to 40W on the plate comfortably.

What would really be interesting would be a comparison of transconductances, but considering a lot of these tubes are "close enough" to a 6L6GC I suspect there wouldn't be too many surprises. Specs on those Russian tubes (6P3S-E) are hard to track down, so it'd be nice to finally get some solid data. Anyway.

There are handy bar graphs of "lows," "mids," and "highs." This was kind of a mistake, or rather, how they define "low" and "high" is problematic. The open low E is 82 Hz, but this chart defines "lows" as 50 Hz. Even a baritone B string is 62 Hz, so the measurement of "lows" on these bar graphs is useless.

"Mids" are defined as 700 Hz, which is pretty reasonable.

"Highs" are defined as 6 kHz, which is not pretty reasonable. Most guitar speakers have already started rolling off around 5 kHz. 3kHz is around the "icepick" region and probably would've been a better choice.

To generate the frequency response graphs (and the guitarist gibberish,) they built a little single-ended guitar amp to put the tubes in. I take issue with a couple things here. First off, they didn't make a neutral Hi-Fi amp, they made a guitar amp. They describe the preamp as being a Blackface Fender design, and that's good for frequency plotting because there's certainly no limit to the low end, but unfortunately there's also a freaking tone stack! Yes, you can get a Blackface tonestack to be almost linear by turning up the mids to 10 and turning the bass and treble to zero, but did they do this for the frequency response plotting? Did they instead take the tone stack out of the circuit? These details aren't provided.

Anyway, these are graphs of just individual tubes so buying tube X and actually getting frequency response X is probably a crapshoot. Those little bumps are the result of manufacturing variances; if they averaged 10 of the same tube the responses would look much more similar brand-to-brand. Plotting dB on a linear scale instead of a logarithmic one is a little disingenuous too; even speaker manufacturers don't try that. Anyway, the biggest difference they show is roughly 2dB which is just under the threshold of what the human ear can discern as a difference in volume.

There is also the sizeable mid-hump which all the tested tubes exhibit. In the guitar community, 6L6s are widely considered to be somewhat "mid scooped" and the swap to EL34s will return these missing mids. At the very least, we can now clearly show everyone that the mid scoop comes from the circuit around the tubes, not the power tubes themselves. I suspect this bump in the frequencies is more related to the circuit, than a characteristic response of a 6L6.

In this case, the circuit these power tubes are plugged in is quite simple. Like I mentioned above, we have to assume they took the tone stack out of the circuit before measuring the frequency response of each tube. The circuit itself is single-ended, with no negative feedback. This is potentially a huge oversight - most guitar amps use a global NFB loop around the power tubes and phase inverter. I honestly can't think of a single 6L6-based amp that doesn't. I'm sure they're out there, of course.

The point being that NFB reduces distortion - the output is fed back to the input out of phase, so if the tube generates a boost at the output, that gets fed back to an earlier stage as a notch, and the result is a flat response. So, it makes sense that you would remove NFB if you wanted to plot a theoretical response of each tube to see if there really are significant differences from brand to brand.

But what happens when you put these tubes in your amp?

Even without NFB, a push-pull amp will cancel a noticeable amount of (even ordered) distortion. As the fluctuations in frequency response are largely sporadic, there would be significant cancellation of some deviations from linearity. Some would reinforce each other, so I suspect the overall response would look less choppy, though approximately the same.

With NFB though, all of that choppiness is going to cancel, so the only difference you're going to see is the overall y-axis offset - also known as the difference in emission & transconductance from tube to tube.

So to wrap this whole thing up:

• All these tubes are basically identical.
• There are some apparent deviations in frequency response at around +/- 2dB. 
• These differences are inaudible.
• Even if they were audible, most amps have NFB, so these deviations would cancel out.
• Only one tube of each brand was tested, so we have no idea if there are any trends among brands, or if CE Distribution just happened to select tubes that were basically identical. 
• This document is just a piece of marketing material, and shouldn't be viewed as anything but an advertisement. 

Bias Supply

This is how crazy easy this is. I bought a Triad VPL36-140, which is a tiny transformer that you can wire for either 18VAC or 36VAC.

Wired for 18VAC, I get ~24VDC. I'm not accurately simulating the load current, but 18*Sqrt(2) minus a couple volts for drop across the diode = yep.

Wired for 36VAC, I get ~49VDC.

Since I need only around -21V to bias the 7591s in Bodie, I'm thinking the 18V route is the way to go. Sure, even though I calculated the source impedances for both configurations (secondaries in series or parallel) I'm not sure I trust PSU Designer to accurately predict the load regulation on a tiny little 5VA transformer, especially considering the load current is going to be practically nil. Even assuming ~20k for a bias pot and a resistor to make sure the pot can't dump all the V- to ground, I'm still getting ~23VDC.

Merlin has R1 as 100k in his power supply book, which is probably more realistic for ripple reduction. I'll have to balance that against how long it'll take the bias supply to reach a stable voltage. Considering this is a full-wave bridge rectified supply, the ripple will be at 120 Hz which is much easier to filter than the standard half-wave rectifier in most guitar amps which pumps out 60 Hz ripple.

And hey, finally a little circuit where I can buy almost all the components at Rat Shack!

Vacuum Tube Valley Magazine

http://www.jumpjet.info/Pioneering-Wireless/eMagazines/VTV/vtv.htm

Thanks to jaaxx at ax84.com for finding this.

The Zener Trick

After a little more digging I found out that the reversed zener diode trick was actually patented back in 1963. (Found this out somewhere in the old Ampage archive, no luck finding it again since.) It's still pretty new in the guitar amp world - pretty sure no mass-production amps use it and apart from a handful of small hobby builders (Chuck H at M-E-F says he put it in a prototype amp he built for Dean Markley) it's unheard of.

Just goes to show how behind-the-times guitar tube amp builders are! Then again, I doubt the guys who patented it back in '63 had any idea it would make 'better' non-linearity, which is the other issue with tube guitar amps - it's uncharted territory, at least from an academic standpoint.

Richard Kuehnel (of ampbooks.com) in his book on power amps proposes that while we think of (nearly all) tube guitar amps as operating in class AB1, they should really be labelled differently considering the power tubes are routinely driven past the point where grid conduction would begin if only there were something connected to the grid to supply the current. There isn't a current source though, just a capacitor to charge up quickly and drain slowly leading to delicious farting sounds.

What I'm trying to get at is this whole tube guitar thing is nuts. We have mountains of information on how to build hi-fi tube amplifiers and we can do all the math, plot the load lines and figure out dissipation, harmonic content, linearity... but an equal part (even the greater part for many modders) is just playing it by ear, literally. But then, of course, the educated guesses are usually far more productive and innovative than the blind guesses, which is how an idea novel enough to be patented fifty years ago is still novel today.

Recommended Reading

It can be hard to find decent information which is specific to guitar amps. There are plenty of sites where you'll see advice like "You have a Bender QuintupleRec JCM ABC 3.14 Green? Change R14 to 1M, C2 to 1uF... (etc) - Instant Page/Hendrix/Young*!"

...but if you ask "why that component? why that value?" then your options are a lot more limited. Fortunately there are a number of online communities now, many of which feature People Who Actually Know What They're Doing - both people with electrical engineering degrees and other people who've been amp techs/builders for decades.

To that end, I've been updating the "links" list as I remember more sources I've used. If I had to pick one, hands down, it would be Merlin Blencowe's Preamp Book. The first edition is phenomenal and he just released a second edition that I want really badly. It's a perfect balance of technical and practical - possibly not the best for absolute beginners, but he's not afraid to simplify when appropriate.

Tubes have been obsolete technology for a while now (w.r.t. consumer electronics) so the classic books are old and focused on minimizing distortion - reproducing music, where you want the amp to stay out of the way, instead of producing music where the amp is part of the sound. They're also generally more technical than needed for guitar amps, but still a valuable resource for fundamentals. I'm still working through Amplifiers (The Why and How of Good Amplification), G. A. Briggs, 1952 which has been interesting and is well written with occasional dry jokes - it's not that nerds don't have a sense of humor, it's just that they generally have a bad sense of humor.

Somewhere in between are Richard Kuehnel's books - very technical, but very powerful tools. I have his power amp book, and if you're interested in understanding blocking distortion, this is the best treatment of the topic I've found. Some topics he goes into considerable depth, but others he doesn't write quite enough. Really math heavy, but that's a good thing. Great calculators on his site too.

Across the various boards the people are mixed, but generally if someone is way off someone else will step in. If you see Randall Aiken, Merlin, or RG Keen start talking, pay attention. My favorite forums are the AX84 BBS and Music Electronics Forum. The former is hard to search and the latter is frequently hacked so I usually search both through Google.

*Your choice: Angus, Neil, etc. I've never seen "instant La Monte Young"... oh man, that would be awesome.