©2001-2003, 2012 Jeremy Epstein. No reproduction, retransmission or web distibution without prior written permission.

This page has information about my
"free lunch" amp.
diy audio
octal riaa preamp
6C45P riaa preamps
teres turntable
tuned quarter-wave pipes

The "Free Lunch" Direct-coupled 2A3 amplifier, by Jeremy Epstein

"Rewire! Reuse! Recycle!"

1. Evolution

While exploring the delightful, cheap little "Darling" amps (which use the 1626 indirectly heated power triode,) I learned how to work with a direct-coupled design. I came to enjoy their simplicity and the good sound a D-C amplifier circuit achieves.

But a "Darling" amp, even using two 1626's in parallel, simply does not put out a whole heck of a lot of power : less than two watts. I don't have enough room in my living room for a big pair of horn speakers, so there is an upper limit to the speaker efficiencies I can get, and a lower limit to the useful power range of an amplifier. Having heard some nice 2A3-based designs, I had come to agree with the title of Joe Roberts' article in Sound Practices: "I Never Met A 2A3 Amp I Didn't Like." So I decided to build one, a simple direct-coupled amp using the same 8532 driver stage the "Darling" amps use.

I learned a little bit along the way, and as I came to understand Miller capacitance better, I drew the conclusion that a low Rp value was very desirable in a driver tube - the lower source impedance of the driver stage translates into better high-end bandwidth. I read up and the 417A/5842 recommended itself to me: the gain was about right and the Rp was about four or five times lower than the 8532. Installing the new driver gave me the results I had hoped for - better high frequencies and an overall impression of more speed. Very nice. The driver runs at a fairly high current, over 30mA, so you are basically using a small power amplifier (with gain) to drive a bigger one. Some builders separate the two functions of the driver tube into a voltage amplifier for gain and a lower-mu, low-Rp tube like the 12B4A for controlling the 2A3 grid.

High Gm tubes, like the 417A and 437A, as well as the 6C45Pi-E, are very prone to parasitic oscillation at high frequencies, above the audio band. These oscillations can show up in many ways. I apparently had a hum problem, which I could not get rid of by any normal means. It turned out to be a parasitic oscillation in the driver tube, riding on the small residual hum signal, and vastly exacerbating it. It goes without saying, it was a big pain in the ass to figure this out. Grid-stoppers are a must with these tubes. Referring to the schematic (Figure 1), the grid stoppers are the 100-ohm resistors in series with the grids of the driver tubes. For best results, they should be mounted directly on the tube socket lug, with the lead between the resistor and the lug as short as possible. What I usually do is to wire up the resistor and the grid lead, attach this assembly to the lug, and then slip heat shrink tubing over the lead wire, the solder joint, the resistor, and the lug. I have heard of an alternative grid stopper, made by directing the grid lead through a ferrite bead. I can't comment on the effectiveness of this myself but it is said to work well.

This 417A/2A3 amp was a strong performer in one arena I find very important these days : it gave a dramatically wide, deep, and tall representation of the recorded space. I find that an amp has to be "firing on all cylinders" before a convincing sound image is achieved, and if it has significant flaws, they will show up immediately as a poor spatial representation. Naturally, since this was still a single-ended amp using a directly-heated triode output tube, this spatial drama did not come at the expense of the relaxed musicality that I also strive for. Dynamics were superior with the new driver.

One day, I learned that Sovtek was on the verge of releasing their newest tube: a new production monoplate 2A3, for a very reasonable price. I called a friend, an engineer who was working on the development of this tube, and I cajoled him into letting me buy one of the first pairs from the initial run.

I found that the Sovtek monoplate 2A3 is a very good tube, especially when you consider it only costs about 30 dollars. And it has the nice feature of being able to be run at about 50% more plate dissipation than a classic 2A3. I upped the B+ a little and started running my 2A3 at a very hot operating point, about 285 V p-k and 82 mA. I get a little more power out of it that way and I like the sound. This operating point is not recommended for your vintage monoplate 2A3's. I have also tried out the AVVT 2A3M mesh-plate 2A3 in this amp and I adjusted the bias to reduce the quiescent dissipation ­ the meshplates are pretty expensive and I would hate to kill them. (See the circuit description at the end of the article for bias adjustment information.)

After living with this amp and the Sovtek 2A3's for a while, I visited the working loft of a friend, Dave "Clippy" Slagle, a notorious NYC-based DIY extremist. Our experiments during that visit were chronicled in an article in VALVE Vol.7 #1, "Sovtek 2A3 and AVVT AV2A3 Operating Points". He showed me a schematic (inspired by Mr. Komuro) for a clever direct-coupling scheme which used two B+ supplies (derived by tapping a voltage doubler at two spots) and a choke-loaded driver stage. The voltage drop across the choke's DCR set the bias for the final output tube. I drew myself a copy of it and pondered it for a while. I also convinced him to part with some of the mysterious Russian tubes he had been raving about, the 6C45Pi-E, which was another low-Rp, high-Gm tube that he favored over the 417A. I swapped the russkie for the 417A and I liked what I heard ­ there had been a slightly "dark" quality to the midrange that the russkies seemed to remove. (Maybe I hadn't killed all the parasitic oscillations, who knows?)

2. Revolution

One day, my pondering yielded a bit of a "Eureka!" moment. There was already a dual supply of sorts in my 2A3 amp ­ the main B+ supply was the top, and the bypassed cathode of the 2A3 was the bottom. Since the cathode was an AC ground, I could use this as a DC supply for my driver and try Dave's circuit with only minimal rewiring. I'd get the bigger driver swing available from the choke-loaded driver, and lower the source impedance of the driver once more. But what to use as plate chokes? I had read enough about choke-loaded parafeed amps to know that the bandwidth of the choke-loaded stage would be compromised if the choke were of poor quality. So just looking around in the basement for doorstops was out.

Fortunately, I had some suitable parts on hand. The thermionic love-in known as nyNoise had brought a houseguest named Thomas Mayer all the way from Munich to Brooklyn a few months before. Thomas is quite well mannered, and he has earned the nickname "Iron Man" for his devotion to transformer coupling. So, his idea of the perfect house gift was a pair of Lundahl 1660 interstage transformers! (It's true my wife wasn't as impressed as I was, but she did envy his comprehensive collection of black clothing.)

These IT's are designed for coupling single-ended stages and are air-gapped for 18mA of DC current through each half of the primary. I strapped both halves of the primary in parallel and made 32H, 36mA plate chokes from them. The DC resistance of these chokes was too low to get the needed voltage drop across them for biasing the 2A3's correctly, but a relatively small value (1K) resistor, in series with the choke load, did the trick. Lundahl also makes a choke (LL1668) wound especially for plate load service that would probably be much better than the jury-rigged interstage I used, and Magnequest has a range of plate choke options too.

Well, this connection sounded really, really good, more dynamic, more open, faster, and ­ yup, a bigger, deeper soundstage. The trick of recycling some of the 2A3's current to power the driver cuts the demand on the power supply somewhat ­ you are basically getting a free power supply for the driver. Also the power that was dissipated as heat in the 2A3's cathode resistor is no longer wasted, but becomes the driver's quiescent power dissipation, allowing the amp to run a bit cooler. The choke-loaded driver stage allows better gain and a larger voltage swing than a resistively coupled driver, and the choke also provides a nice low-DCR path for the 2A3 grid to (AC) ground. The placement of an active element in the bias-set circuit (the driver tube) adds some positive feedback to the topology, and perhaps this contributes to the open, dynamic sound. Despite the positive feedback, the active bias arrangement actually stabilizes the 2A3 bias point. (These last two conclusions were drawn from Bob Danielak's PSpice analysis of the amplifier circuit, for which I thank him.)

The way these technical benefits dovetailed with the clear sonic improvements led me to dub the circuit "Free Lunch."

I enjoyed this configuration a great deal, and, since Dave hadn't steered me wrong yet, I decided to try another of his favorite techniques ­ battery bias for the driver stage. I used a pair of AA Ni-Cd's in series for about 2.4V - the current through them keeps them charged automatically. They have been in there for months and the 70mA of driver current have not caused an explosion. Yet. This was a real ear-opener ­ what I heard was a big improvement in bass note definition, as well as a clean-up of the mids and highs as compared to a bypassed cathode resistor or an LED. (The LED was actually tried first ­ it actually did the same good things to the bass but the upper ranges were somehow not quite right to my ears.)

Excited by what I heard, I showed this circuit to the JoeNet pundits and rapidly found that some other people had been working towards a similar topology. One builder, Brian Clark, had been calling his version the "Monkey On A Stick," which sounded to me like some sort of obscene carnival snack (kind of like a super-disgusting corn-dog.) I was directed to John Broskie's "TubeCAD Journal" on the Web - he had also drawn up the Monkey circuit as a thought-experiment in a discussion of the Loftin-White amplifier. Doc Bottlehead's "Afterglow" circuit, published in the above-referenced Sound Practices article, was similar, but used a solid-state constant-current source instead of a choke. Jack Elliano of ElectraPrint has a design that is essentially the same except for the Ultrapath connection of the final power supply bypass cap. Christian Rintelen, always a pacesetter, has been listening to a similar amp for a while, designed in 1996 by his friend Reto Luigi Andreoli. And of course, there was the Komuro voltage-doubler version Clippy showed me that was the inspiration for mine. It appears that there have been a few different designers, then, contemplating this sort of arrangement, although most have been using stacked power supplies, while my amp uses only a single supply. The Monkey is more unconditionally stable than my design: the placement of the cathode resistor tap ensures that the 2A3 will have bias present even if the driver tube draws no current or is pulled out entirely. Mine does not have this feature­ but I enjoy the sound of the amp best with all the tubes in it anyhow, so this does not concern me. Refer to the diagram (Figure 2) below for a comparison between the two topologies.

I took this amp to the Århus Triode Audition in Denmark in August, 2000, where I played it for some diehard DIY mavens including the Iron Man himself and Christian. It was well received, and now I am confident that I'm not the only one who likes the way this amp sounds.

3. Solution

An explanation of exactly how it works is in order, since I have shown this circuit to a few people who have scratched their heads over it. The best way to understand it is to keep track of the current draws and I x R voltage drops, in other words, to do a DC analysis of the quiescent operating points. Refer to Figure 1 below.

Start at the plate of the 2A3. (We will use plus-to-minus current analysis, not conventional minus-to-plus, because in this case it is easier. The result is the same if you do it the other way.) About 82mA of current is drawn through each 2A3. Since the B+ is 495V and the desired Vp-k is 285V, that means that the cathode is at 210V. In order to bias the 2A3 we need about -45Vbias so the grid of the 2A3 must sit at 165V DC This point is also connected to the plate of the 6C45Pi, so this is the driver's plate voltage too.

The 6C45Pi drivers are being run above their maximum Vp-k ratings (150V) by the way, but these tubes are no wimpier than are the vodka-swilling Russian factory workers who produce them. Their maximum voltage ratings are adjusted to ensure survival of 75G shocks in military service! I have it on good authority that these tubes cannot be killed by normal means ­ you will need a wooden stake and a crossroads to stand a chance. They operate at 165V p-k, 2.4 Vbias, 35mA quiescent current.

Note, however, that this 35mA quiescent current is flowing from the cathode of the 2A3 into the choke load circuit, so, only a portion of the 2A3's 82mA quiescent current is drawn through the 2A3's cathode resistor. Subtract 35mA from 82 mA, and you have 47mA remaining. 210V/. 047A = 4468, so we require a 4.5Kohm cathode resistor hereto hit our 210V target. The 4.5K resistor dissipates .047*. 047*4500=9.9W so use a 20W to 25W resistor here.

The plate choke has a DC resistance of about 250 ohms, and so the 35mA of driver current across it drops about 9V out of our desired 45 Vbias. A 1K resistor drops 35V and the resulting Vbias of -44V is close enough for hifi.

Note that bias adjustment of the 2A3 can be done very conveniently in the "Free Lunch" by adjusting the value of the 1K resistor. A larger value will make the grid of the 2A3 more negative, "cutting off" the tube or reducing the current, conversely, a smaller value will run the 2A3 hotter. However, for a cooler operating point, the current through the 2A3 cathode resistor is reduced, so this value must be adjusted too.

An example:

If we wish to run the 2A3 at a cooler operating point, closer to the spec value of 15W plate dissipation, two resistors must be changed. The 4.5K resistor in the 2A3 cathode circuit must be changed to 10K/20W, because there will be much less current flowing through this resistor at the new, cooler operating point, and we need to maintain approximately the same 210V drop across it. The 1K resistor in series with the 6C45Pi plate choke is revised to 1.5K. With these new values, the 6C45Pi draws about 30mA current, and the bias for the 2A3 is increased to about 52V, which "cuts off" the 2A3 slightly for cooler operation. The new operating point yields 16-17W quiescent dissipation, within striking distance of the desired 15W anyway ­ the AVVT plates do not glow red at this operating point. Note that there will always be some experimentation required because the 6C45Pi varies quite a bit in reality from its specs. Try and get a matched pair if at all possible, otherwise you will have a lot of trouble balancing the two channels to the same operating points. Any given pair will require some tinkering with the series resistor, especially with the fixed (battery) bias.

As always, I would like to thank all the teachers I have had who have inspired me and helped me as I fumble around in the confusing milieu of electronics, particularly the members of the Joenet mailing list, without whom I wouldn't even be able to spell "DC." I'd particularly like to thank rocket boy Bob Danielak for his PSpice analysis of the circuit, and technical review of this article ­ any mistakes are mine, any really smart stuff is his. Big brain on Bob. And finally, a special thanks to Zelda's 52-Hour Photofinishing for inspiration, and the constant encouragement to experiment.

Figure 1. The "Free Lunch" amplifier.

Figure 2. A comparison of the "Free Lunch" and "Monkey" arrangements.

Recommended reading and viewing:

"I Never Met A 2A3 Amp I Didn't Like," Sound Practices #15, no longer on the web, but available on the Sound Practices Archive CD at:


"Design Idea : A Safe Loftin-White Amplifier," from TubeCAD Journal, May 2000, on the web at:

http://www.tubecad.com/may2000/page17.html (refer to the upper right figure.)

"Direct Reactance Drive Amplifier" by Jack Elliano of Electra-Print, on the web at: