Here’s a picture of the front of the monster amp…
This baby’s good for over 300 watts per channel. As it stands now, both channels are slaved together, since I have no need for a stereo rig. In case of the possibility of ever need two seperate channels, the tube sockets are already mounted with heaters connected.
The case you see in the picture is an all aluminum job I scavenged from a high voltage power supply. The supply itself was a MONSTER, with DC voltage selectable from 450 to 5500 VDC. I still have all of the transformers, one of which is a good 45 pounds! The four black circles you see at the bottom left are the fuses. There is one for the entire amp, one for the heater secondary, and two for each center tap B+ connection to the output transformers. The red LED is the indicator for the heater supply. The large black knob on the right is a three position power switch, OFF, HEAT, and STANDBY. The standby position applies power to the auxillary power transformers which supplies the 12AX7′s, 12AT7′s, and EL84′s with -300/-150/+150/+300 VDC. The small silver SPST switch is for the main B+ tranny, and there are two other LED’s for those two plate supplies (yellow and green–for GO).
Here you can see the two Hammond 1650W output transformers. They are mounted in the front of the unit since they represent such a tremendous weight (nearly 60 pounds!). Were they mounted in the rear, if the unit is rack-mounted they would place a great torque on the front plate, as well as the screws holding the rest of the chassis to the faceplate. Mounted in front, they are at least directly attached to the mounting plate for the entire amplifier.
Also visible behind the forest of tubes are 8 LCR can caps. These babies are good for 500 mics @ 500V. They are only used for the main B+ supply which is around 650VDC. As a result of the voltage requirements they had to be series-connected. You may notice that the units on the left have some black PVC “electrical” tape over the tops. This is because the cans of these caps (the negative terminals) are sitting at over 300VDC! I had the experience of placing the palm of my hand on one once. The tape went on soon after!
Those “small” tubes visible between the KT90′s are EL84′s (aka 6BQ5). These tubes are run as cathode followers, directly coupled to the grids of the KT90′s. This allows a very stable bias injection scheme (to the grids of the EL84′s) as well as a very low drive impedance to the KT90 grids. As a result, the KT90′s can push about 15V into the grid current region, which maximizes plate voltage swing for a given B+.
This is a top view of the amp, bleached out due to the flash going off. The transformer to the bottom right of the picture is a Thorardson filament transformer to supply the massive heater requirements… 12 KT90′s, 4 EL84′s, 2 12AT7′s, and 2 12AX7′s add up to some serious amperage–about 24A @ 6.3VAC! The tranny is rated for 25A. I got it for a steal on E-Bay. You can also see the banana jack speaker outputs on the far left and right, on the back of the amp. They are hard-wired to the 4 ohm taps on the Hammonds.
As it stands right now there are NO volume controls on the amp. In fact, the only pots in the entire design are the four bias adjustment pots on the back pannel. Normally, I leave them on the fully “cold” position, which puts about 70-75 negative volts on the KT90 grids. I have no desire to burn up tubes if there is no sonic benifit, and so far I have not heard any by running them hotter. As it stands right now each KT90 dissapates about 13W at idle.
There’s one thing you may also note, and that’s the tight spacing between the KT90′s. Obviously, cramming a 600W+ tube amp into an 8 space rack mount chassis is not easy business, and something had to give. Instead of having a hum-wracked design, I decided to go with the slightly tigher tube spacing, since I was planning to impliment a forced air cooling scheme anyway. None of the tubes actually touch each other, and so far I have had no problems with overheating whatsoever. I would NOT try to run this amp without the fans going, though!
When I get around to updating the schematics, I’ll be posting those here too. There were a few changes that were made from the first approximation to its present incarnation, and there’s a few more I’ll be doing too. As it stands the plate tranny for the KT90′s is a surplus toroid from a SS amp that I modified for 240VAC on the secondary. This was not enough to simply rectify, so it’s currently a voltage-doubler arrangement. At the moment I have a 1400VA do-it-yourself power tranny on order from Toroid Corp. of Maryland, which I will wind myself to about 450VAC. That will do away with the doubler. I may actually switch to a choke input filter and get rid of some of the LCR’s–they take up a lot of space, and present a very tough initial charging current to the tranny. Lights definitely go dim when I hit that silver switch, and I’ve opened up some 1N4007′s in the process.
Well, time has passed and I finally bit the bullet. I managed to stop playing my amp long enough to actually get back inside and do some more mods to it. Here’s what I did:
First I ripped out the old “Williamson” style input stage, which is nothing more than a fancy name for a common cathode stage direct coupled to a concertina style “split-load” phase splitter. I was unhappy with the performance of the single ended input stage. I originally used it because it looked cool, was direct coupled and all, and allowed me a handy point (in the form of the unbypassed Rk of the first stage) to inject some NFB.
That said, I had opened up the loop one day, removing the lead from the OPT secondary—lo and behold, the amp TOTALLY opened up. I mean, the sound threw off a giant wool blanket and got RIGHT in my face. The downside to opening up the loop was that the nasty PS hum was increased, and that pretty much prompted all of these other mods as well. I knew I had to get rid of that damned hum. I also knew that I did not have to provide a handy point for injecting the NFB, since I was convinced NEVER to use it, at least not in THIS amp.
So, if it wasn’t going to be a Williamson input stage, what then? I looked long and hard at the schematics, and took stock of my gear. The last thing before this monster amp is a Peavey 15 band 2/3 octave graphic EQ. It’s not the quietest unit, and I’m sure it’s not the cleanest, but it allows me a great deal of control over the sound, and that’s something that comes in REALLY handy when playing in “difficult” rooms. It also allows me to throw in a 18 dB/octave LF rolloff @ 40Hz, and that keeps the excursions under control. In any event, this unit has both balanced and unbalanced outputs. It was time to use the balanced outs.
I punched a hole in the back of the chassis for an XLR jack, and removed the ¼” jacks.
Now, with a nice balanced signal coming into the amp I wanted to preserve it, so the first stage was to be a differential driver. I used a 12AX7 for gain here. Plate resistors were to be 100K, since the only load was going to be the grids of the next differential driver, and they represented an easy to drive dynamic impedance. The B+ for thes stages sits at about 300 to 320 VDC. If I wanted the plate voltage to sit at about ½ to 2/3 of B+, that meant I had to drop about 100 to 150V across the Ra’s. That meant that each section of the 12AX7 was going to pass about 1 to 1.5 mA [V=IR so 100=I (100,000) and 150=I (100000)].
I have symmetric preamp power supplies, and I wanted to tie the low side of the shared Rk to a negative voltage, so that the Rk could be made bigger in value, and therefore the “tail current” would be stabilized. With a good solid “tail current” the distortion-reducing properties of the differential driver are preserved and accentuated. So, from my negative supply of -300 to -320V, a good value of cathode resistor (to pass 2-3 mA) was 100K. Don’t forget you have to sum the currents of BOTH sections of the driver when you size the Rk!
On the input side, I placed two 68K grid stoppers right on the socket, to roll off the RF response of the amp. These also had grid leak resistors of 1.5M to ground. Finally, two 100 nF film caps went to pins 2 and 3 of the XLR. These caps would prevent any DC offsets from being injected into the input stage. I left pin 1, the ground pin, floating. It would be grounded at the other side of the cable, and this prevents any ground loops.
On the output side, the plates of the input stage differential driver were cap coupled to the next stage using 100 nF film caps, sourced from Mouser. The “low” side of these caps is tied to both a 1M grid leak resistor and the grids of the 12AT7. I chose the 12AT7 because this stage had to drive a dynamic load of two bias injection resistors (each at 150K) and the grids of the EL84 cathode followers. That meant an impedance around 75K.
To get predictable performance from a tube you should keep the Ra lower than the Rl (or load impedance). This assures that the plate voltage swings are developed where they should be—across the Ra. So I had a minimum value of Ra, which was 75K. In general, you should use tubes that have plate impedances approximately equal to your load impedance. Strictly speaking, my static load impedance would simply be Ra, but under signal conditions it would be Ra || Rl, or 75K || 75K, which is about 35-40K. A 12AX7 would be operating into to low a load impedance to be ideal, but a 12AT7 would be just about right. A 12AU7 would be even better in that respect, but I have trouble finding decent 12AU7’s and I don’t usually like the sound of them. So a 12AT7 it would be. [Lttle note here: the plate load of a tube is also in parallel with the plate impedance of the tube itself, which the tube must work against in order to generate plate voltage swings. So the analysis is not complete as I have given it, but it is a good approximation.]
Using the same method as the input stage, I sized the plate current of each tube to be about 2 mA. That meant I needed a 4 mA “tail current” at the cathodes, which brought me to an Rk of about 75K. Be sure to check your voltages and currents across your resistors to ensure you do not dissapate too much power across them! Don’t forget to take into account the AC voltage signal swings that occur as well. In this case, as the shared Rk of a differential pair, there would be little or (ideally) NO signal voltage appearing across the Rk.
Moving right along, I removed the 150K grid stoppers from the EL84’s. They really weren’t doing anything useful, just rolling off the HF a bit with the input C of the tube. This also lowered the grid circuit impedance of the EL84’s which made them a little more stable. Oscillations were not a problem.
For simplicity’s sake, I decided to go with a single input and driver stage, and then split off four signals to the EL84’s from there. There is really no need for me to have a stereo rig at this point, and the extra tubes just load down the power supplies and add heat. In case of future stereo uses the sockets are mounted and heaters wired.
So, from the 12AT7’s plates we go through four 1 mic foil caps. The other side of these caps is not ground but the negative bias injection voltage, derived from four bias pots located on the back pannel. The normal voltage of these four lines (two of each phase) is something like -80VDC. This voltage is connected directly to the g1 of the EL84’s.
The EL84’s are triode connected with plates directly attached to 150VDC. G2 is connected to plate through a 100 ohm resistor. The cathodes are connected to two things: a 10K cathode resistor, and the grids of their three respective KT90’s. The cathode resistor is sized to pull about 7.5 mA through the EL84, which makes it 10K. This current is a bit low for normal “plate loaded” operation but CF’s do not require much plate current at all to be linear as hell, so this serves many purposes: 1) cuts the plate dissapation down to about 2 W. The tubes should last a LONG time, plus less heat is pumped out into the amp, and 2) lowers the strain on the preamp power supply. I could cut the Rk’s in half to 5K, which would raise the current to 15 mA per tube, and the idle dissapation to about 4 W, but I have not determined if it is worthwhile… a future tweak, perhaps?
The “lowside” of the 10K Rk’s are tied to a –150VDC supply. As it stands now there are 3K grid stoppers right on the sockets for the KT90’s. I did this because I wanted to preserve my tubes, as osciallations with multiple large power tubes are easy and quick to do damage. The downside is that the low drive impedance of the CF’s is able to push the tubes well into grid current, and these series resistances limit that excursion. I will probably attemp to “hard wire” across these grid stoppers and determine if the tubes will take it, ie not oscillate. I will CERTAINLY be able to get more power out of the amp with them off, but that is not really a design goal at this moment.
In terms of the low drive impedances, being direct coupled, the grids of the KT90’s experience an impedance which is both the value of the Rk (10K) as well as the cathode impedance of the EL84 itself. A quick approximation of this cathode Z is 1/gm, where gm is the mutual conductance of the tube. It is easy to see here that tubes with a high gm make the best CF’s, since their output impedance will be lower. The output impedance of an EL84’s cathode is about 90 ohms. Yes, 90 ohms.
With a source impedance that low, the Miller capacitances of the paralleled KT90’s is not quite so much of an issue, even when triode connected. Additionally, because they are direct coupled the CF has the ability to drive the KT90’s into grid current, where the grid is actually driven positive WRT the cathode. This pulls the plate and the cathode MUCH closer in voltage, since the current through the tube rises dramatically. The downside is that when the grid is positive WRT cathode, it begins to attract electrons itself. In a high-impedance, typical cap-coupled driver, this is unacceptable, and sounds farty—like a Fender cranked way too high. Grid blocking occurs, and the normal operation of the output stage is thrown out the window. A finite amount of time must pass before the grid circuit returns to normal.
All of this is moot in the direct-coupled CF driver. Electrons which are collected by the KT90 grids simultaneously appear at the EL84 cathodes, become part of the space charge, and are ultimately collected at the EL84 anodes. Recovery from grid current is pretty much instantaneous. This allows excursions into the class AB2 or B2 region.
The benefits of grid current are essentially a better use of available B+. Since the tube conducts more current with positive grid voltage, the plates are able to be “pulled” closer to the cathodes. The normal “saturation voltage,” which is the minimum voltage the tube can create across itself, is usually limited by the plate resistance of the Vg=0 plate curve. The slope of the plate curve for a POSITIVE grid voltage is much steeper, meaning the R(on) of the tube has been decreased. The plates can still soar to approximately 2B+ when the tube shuts off, but they are also able to be pulled to nearly 0V while conducting. Voltage swings are increased, current swings are increased, power delivered to the OPT is increased.
Screens have 2K stoppers attached right on the sockets which also decrease power output but greatly extend the longevity of the tubes. There is a front-mounted 4PDT switch that has been recently added to switch the KT90’s from UL to triode mode. Since the amp has no NFB I have no use for pentode mode at this time; it would raise the output impedance of the amp to an unusably high value.
The KT90’s are connected in groups of three, with each group of three’s cathodes passing through a 10 ohm current sense resistor for use in biasing. I have had to use 2W units here as lesser ones will open prematurely.
Lastly, the output transformers are Hammond 1650W’s, 1K9 p-p impedance. The units are consecutively numbered from the assembly line. (Thank you Shawn Romaniuk!) Primary center taps are fused with 1A slo-blo pannel mount fuses, and are fed from a 1000 mic @ 1000V cap bank (eight 500 @ 500 LCR can caps). The LCR’s are bypassed at HF’s with 10 nF film caps. I have not observed any PS resonances due to this paralleling of caps. The main B+ for the amp is a voltage-doubled 245VAC winding of HEAVY guage (I’d say 16 AWG or larger) that’s good for about 625VDC at idle.
Here are some pics. You can right-click, and select “view image” to get a bigger picture.
This is a shot of the back of the amp. Visible, from top to bottom, are the two Hammond 1650W’s, the eight LCR can caps, the forest of tubes, and the heater tranny on bottom right. Also note the four bias test points, and four bias adjustment pots.
Here’s a piccy from above, illustrating the two 12VDC fans (over-voltaged to 16VDC) blowing directly over all 12 KT90′s. I have since replaced the hack-ass ty-wraps with bolts, and covered the fans to prevent inadvertant obstruction.
Finally, the rat’s nest under the hood. Well, *I* know where everything is! Obviously, right in the top center is the toroidal plate tranny. To the right of that are the filter caps for the +300/+150/0/-150/-300 supplies, and to their right, the two trannies for those supplies. The 12VDC (floating) supply for the fans is to bottom left. Note the MOV to prevent surges installed at the 120VAC barrier strip to top left.
Note that first of all, I am working on a schematic to reflect all of the changes that have been implemented so far. It should be up within a couple of days.
I “hang out” a lot with the tube HIFI community on the web, frequenting the REC.AUDIO.TUBES newsgroup as well as belonging to a number of E-mail listservers. Although the ultimate goals of the HIFI community are greatly different than the goals of musical instrument amplification, many of the techniques can be applied. In the pursuit of ever LOWER THD, the main focus of the HIFI pursuit offers insights as to how the distortion is generated in the first place. Of course, this is what I’m interested in.
I am of the design philosophy that the power amp should be designed as efficiently as possible. Usually, that means proper design techniques, such as maximizing undistorted output voltage swing, proper impedance matching and loading, and high bandwidth. Low THD is not a primary objective, but is usually achieved in the execution of a well-designed power amp circuit. Along these lines I started to pursue circuit tricks used in the fringes of the HIFI tube community.
One of the greatest benefits comes from the use of discreet SS devices to “help” the tube circuits perform well. In this spirit, I designed and implimented a Constant Current Source for the EL84/6BQ5 cathode followers which drive the KT90 output tubes.
There are three primary dynamic characteristics of a tube, all related to one another. They are Mu, or the amplification factor, often seen as ” µ “; plate impedance, or ” rp “; and transconductance, also known as mutual conductance, or ” gm.” The fundamental relationship of these three variables is µ = gm * rp. The units for the variables are ohms for rp, and mhos for gm. The amplification factor of a tube is a pure ratio, and has no units . The ohm, a unit of resistance, can be defined by Ohm’s law as Voltage / Current. Likewise, the measurement of gm corresponds to a plate current change for a given control grid voltage change, or a Current / Voltage ratio. As you might expect, the unit of the mho is the reciprocal of the ohm. This all makes sense, since when we refer back to the equation of the variables above, the inverse units of rp and gm will cancel, leaving a unitless result, the mu.
Now, ideally, none of these would change with the different operating points of a tube. Actually, it would be sufficient if only TWO of them did not change and were held constant, since the third is a product of the two. Sadly, this is not the case, and is the source of distortion in all tube circuits.
The mu is probably the most constant of the three variables, since it depends primarily on the internal geometries of the tube and not on operating points. Thus it is as constant as we can expect. The plate resistance will go DOWN as the current through the tube INCREASES. You can see this on the plate family of curves for a tube in the way the lines continually become more and more vertical with higher plate currents. Now, if mu is to remain constant, and the rp goes DOWN with increasing plate current, then what must happen to gm ? Of course, it goes UP, in order to preserve the original equation. Thus the gm or mutual conductance increases with increasing plate current. This is true for all tubes and is a byproduct of the “three-halves law,” which is beyond this article’s scope.
The fundamental point I am making here is that the distortion of a tube can be GREATLY decreased by holding the PLATE CURRENT constant. This may sound strange and counter-intuitive at first, since the tube OPERATES by varying the number of electrons which flow from cathode to plate, but it is nonetheless true. The lowest distortions always come from a load that is INFINITE in impedance. An infinite impedance is another way of saying a CONSTANT CURRENT. You may be starting to see where all of this is going now.
Now let’s take a cathode loaded stage, AKA a cathode follower. In a nutshell, the voltage on the cathode will “follow” the voltage on the grid within a couple of volts. The output of the CF is in phase with the input. Most CF’s are loaded by a cathode resistor to ground, with the grid’s voltage elevated. Assume quiescent conditions of 300Vp, 95Vg, and 100Vk, and an Rk of 100K, to keep it simple. At this operating point, the plate current is Vk/Rk, or 0.001A. Now, wiggle the voltage on the grid a bit. As the grid’s voltage rises, the cathode’s voltage will follow. At the peak of our wiggling, the grid voltage is 185V, and the cathode’s voltage is 190V. Now our plate current is Vk/Rk or 0.0019A. At the other extreme of wiggling, the grid’s voltage is 5V, the cathode’s voltage is 10V, and the plate current is Vk/Rk or 0.0001A. This is a very large difference in currents for our little CF to contend with, and it responds as we predicted by offering fairly high distortion figures.
One way to minimize these is a “halfway point” between what we’ve just described and a CCS, and that is to tie the Rk to a NEGATIVE voltage rail, and make the value of Rk larger, keeping the quiescent conditions the same. This will cause the swings in current to decrease, and the distortion to decrease as well. When you look at our equation for plate current above, Vk/Rk, increasing Rk will obviously make the swings smaller in magnitude. A side benefit is that the loss of signal voltage through the CF will be minimized as well, since the Rk is larger and not so heavily dividing the output voltage. Keep in mind though that an analogy of a CCS is an infinite resistance supplied from an infinite voltage. Obviously this is really what we want here, even if a perfect CCS is unobtainable (which it is). It is still worth the effort, since it will bring us more gain (or really, less loss) through our CF, and it will also linearize the gm of the tube, decreasing distortion.
Some of you may be wondering how this effects the OUTPUT impedance of the CF. This is not really a problem, and I’ll tell you why–it’s already so damned low to begin with! While it IS possible to lower the output Z of a CF by lowering the Rk, you really have to get into the region where the Rk is close to the INTERNAL cathode impedance of the tube, which is approximately 1/gm. When you use Rk’s that low, you’re asking for a lot of trouble, since the gain through the tube will be lowered even further, and the current swings will be exacerbated as well. In other words, if you want a lower output impedance from a cathode follower, use a tube with a higher gm. Don’t try to half-ass your way around the problem.
So now that we can see how it would help things, how does on MAKE a CCS?
In my particular case, I decided to use some high voltage BJT’s as the active elements. One could use tubes or MOSFETS as well–it doesn’t matter. Tubes have the drawback of requiring heater current, and they also have a limit as to the maximum heater to cathode voltage differential. In my case, the cathodes would be sitting at about -150VDC, which could turn into a problem. Plus, it is very hard to beat the robustness of SS devices–they do not need replacing nearly as often as tubes do, and they usually cost less when you have to. I chose to use the BUX-85 BJT since it had a 450Vceo, which is the measurement of how much voltage the BJT can handle from collector (or plate) to emitter (or cathode). Plus it was about 80 cents!!!
Essentially, we can control the collector current (or Ic) by adjusting the voltage from base to emitter (or Vbe). This is just like controlling the plate current of a tube by adjusting the voltage from grid to cathode. So if we make the Vbe fixed, we can have a fixed Ic, which is what we want–a constant output current. In my case, it achieved this with a zener diode. If you are a tube purist you could use a regulator tube to the same effect. By connecting the zener diode from the negative voltage rail to the base of the BJT, I set the Vbe to a specific value. The value of zener voltage depends on a couple of things: the desired Ic and the value of Re, which is described below.
One way to increase the performance of our small CCS here is to insert a resistance (Re) from emitter to “ground” (which is in my case the negative rail). This serves the same function as the Rk of a tube stage, offering some degenerative feedback to linearize the output current. This is exactly what we want, remember. I chose a value of 1K, which made calculations nice and easy, and it was a decently high value to work with in terms of performance.
One of the cool things about BJT’s is that the emitter’s voltage (Ve) will follow the base’s voltage (Vb) + 0.7V. This is due to the structure of the BJT and inherent to the manner in which it works. If the collector current is primarily the value of emitter current (it’s not quite, but damned close enough in this case), then we know the current flowing through the Re will be equal to the current flowing through the collector. We know that the emitter current is equal to the Ve/Re. We know then that the collector current will be defined by (Vb + 0.7) / Re, since Ve = Vb + 0.7V. Damn, this algebra comes in handy.
So now, having arrived at a value for Re of 1K (somewhat arbitrarily), we can set the current through our CCS to (Vb + 0.7) / 1K. Notice that for higher values of Vb, the offset of 0.7V becomes less and less of an issue. A good approximate rule of thumb for THIS case would be that the value of base potential, in volts, would equal the value of collector current, in milliamps. That’s damned convenient, isn’t it? Really it’s due to the value of Re being 1K. If the Re were 10X lower, at 100R, then the Ic would be 10X higher for a given base voltage, and so on.
I chose a value of current according to the max dissipation of the EL84′s for a given plate voltage and cathode voltage. Because the cathodes sat at some negative voltage WRT ground, that must be added to the value of Va. In my case, the cathode sat around -80VDC, and the plates were at 150VDC. That gave me an effective plate voltage of 230VDC. The max plate dissipation of the EL84 is 12W. I wanted to keep it as low as possible, for longevity / reliability concerns. Luckily, you don’t need to pass much current through a CF in order to keep it linear. I decided on something like 1/3 of the total available Pa, or around 4W. That made my plate current approximately 4W = Ip * 230V, or .017A. According to our rule of thumb above, I could get that with about 17VDC on the bases. I chose the closest zener diode voltage according to what I had on hand in the old parts bin, which was 15VDC.
As for those 1K Re’s, I had some 1W 1% tolerance metal films that fit the bill nicely. With four EL84 cathode followers, I would need four BUX-85′s, and four Re’s, but all four of the CCS’s could share the same reference voltage, which meant only one zener. By referencing the zener’s votlage to the negative rail, it would track sags due to the imperfect regulation, and would keep the currents largely constant. The only other issue I had to deal with is how much current to pull through the ZENER, since the base currents of the BJT’s would have to be accounted for. Luckily, we were only passing milliamps here, a total of .015 * 4 or.06 Amps total. The beta of the BUX-85 is decently high (30), which meant my total base current would be 0.06A / 30, or 0.002A. 2 mA is easy to deal with, and I sized the resistor to ground from the zener to pull about 10 mA as a just-in-case.
Of course, all of this will become MUCH easier to understand once the schematic is completed. As of today, 03/09/00, I figure a couple of more days. Perhaps this weekend I will post it on the web.
To hold you over a bit, here’s a couple of more pictures:
You can see the two 12VDC fans spinning madly away (at 16VDC), with added fan shrouds, and more substantial mounting using bolts. The power supply to the fans (and top cover) is easily detatched with a nylon jacked 1/4″ plug.
This picture here illustrates the current inrush limiter which was necessary to prevent blowing 15A (!) mains circuit breakers upon turn on. It is connected to the center tap of the stacked cap bank, and leads directly to one side of the PT secondary.
Here you can see the two auxiliary power transformers, supplying the +/-300VDC for the gain stage and the driver stage, and the bias for the output stage. They are cheap 120:240ct 100va units from Allied Electronics, full wave bridge circuit into stacked surplus electrolytic caps. The center tap of the secondary is used as a low Z +/-150VDC supply node.
Here is a pdf file containing a schematic of the cathode follower driver circuit in the BAGA.
Here is a schematic showing the first fully differential incarnation of the BAGA. It’s outdated, but it worked pretty damned well nonetheless!
a recent email exchange between VTL (instigator of the kt90 tube type) and myself can be seen here.