In May 1927 one of the very first indirectly heated tubes was introduced, initially called YU227 and later simply 27. It is the cathode version of the 26, which was directly heated. It was conceived to serve as a detector and audio preamplifier and was produced to reduce filament-induced hum compared with the 26. It can readily be used to drive a 45 or other small output tubes of the era. The 27 is the progenitor of a long line of small signal triodes that ultimately led to the 6SN7. Availability of this tube is still very good, including the “ST/balloon” shape.

Years ago, when I began the first experiments that later led to Nibiru, I built what I called a “26 audio processor,” a sort of unity-gain preamplifier that wasn’t a buffer, based on the 26.

The goal was to add a bit more three-dimensionality and “tube sound” to a digital audio source that already had enough output to drive a power amp directly. Using what’s called a buffer (more correctly a cathode follower) — i.e., sending a signal through a tube with 100% intrinsic feedback, operating at unity gain — seemed pointless to me. I had already sensed that the key lay in reactive elements, so I wound a pair of line transformers with about 150 H primary inductance and a 7:1 turns ratio. Since the mu of the 26 is 7, a zero-feedback circuit would let the 26 amplify and the transformer would bring the signal back down to roughly the original amplitude. The circuit worked, but in addition to misjudging capacitor usage, I underestimated the issue posed by the 26 filaments, which must be powered by a stabilized 1.5 V DC at 1 A — otherwise, hum. I tried to size the supply and regulator properly; it worked, but the heatsink needed to cool the filament regulators had to be much larger than the one I used… after an hour you could fry eggs on it! On top of that, the sonic effect was almost negligible — it was hard to tell whether the circuit was in the chain or not — so I scrapped it to salvage the parts.

At the beginning of February this year (2020) I started thinking again about that trashed project, also recalling the excellent experience with the USB Mini DAC. I decided to merge the two ideas using a more advanced DAC board and found on eBay, for about €30, a small DAC based on the Sabre 9023P — asynchronous USB, 24-bit/96 kHz.

When it arrived I tested it and, even by itself, it sounded very good. Lockdown had begun and work was slow due to restrictions; since I had lots of parts on hand, I decided to start a new project based this time on the 27. I dug out the two line transformers I’d made for the 26 and built a quick lash-up on the bench. The 27 has a mu of 9, while the transformer divided by 7 — perfect to apply what I learned from Nibiru, add a hair of local NFB, and get good three-dimensionality. By ear it seemed to work well. I liked the bench build even though it wasn’t perfect; the measurements showed the following frequency response:

It wasn’t so much the –1.5 dB at 10 Hz, but rather the low-frequency phase response that wasn’t ideal, while –1 dB at 25 kHz on the top end was acceptable. Also, my line transformers were big and heavy because they were made with standard EI oriented-grain laminations.

Since I wanted to build something TOP and didn’t feel like hunting down C-cores (one area where they truly have an advantage over EI for small-signal transformers), not knowing whether I’d find good-quality ones, I — perhaps forgetting the reasons I started winding transformers years ago — took a leap of faith and bought a pair of line transformers from a very famous manufacturer. It was an older model still in production, the only one with roughly a 7:1 ratio, thus the only one from that maker I could use. The datasheet promised a flat response from 10 Hz to 40 kHz with 5 mA DC in single-ended use, so I forked out €300 and waited for the package. When they arrived I put them on the bench — and the huge disappointment arrived… 👿

I measured –1 dB at 9 kHz — unspeakably poor. Maybe I was doing something wrong. I consulted the same manufacturer via the reseller, ran test after test with different tubes and wiring schemes, but nothing. In the best case I got –1 dB at 20 kHz with a 14:1 ratio by wiring the secondaries in a way not even covered in the datasheet. I spent three days testing — one of them until about 2 a.m. — driven by the frustration of having thrown away €300 and dragged back to the anger from years ago when I bought gear that didn’t perform as claimed because “buyers can’t measure and go by ear”… Too bad I had no intention of listening to a chopped-off top end like a little Geloso PA amp!

I sent the package back and to this day I’m waiting for a response about the fault, a replacement, or a refund. With that behind me, I pushed forward using my original transformers. I had a wooden chassis and couldn’t get another — we were in the middle of a pandemic, everything was closed, nobody was working. Something inside me feared the power transformer might be too close, but the line transformers were rotated the other way and shielded by a steel box… The urge to build won out…

The third glass from the left isn’t a tube — it’s an old light bulb rated for 140 V and 15 W, probably the turntable-bay light from some radio. Powered at low voltage (20/30 V) it looked beautiful. I was going for an “industrial” aesthetic I really like, and it fit the look — but I also wanted it to serve a purpose in the circuit. I wanted fixed-bias on the 27s but didn’t want coupling capacitors in the audio path between the DAC and the tube. So I created a shunt regulator to generate 15 V to lift the cathodes of the 27s. I calculated the transformer and regulator to use this lamp as the shunt resistor. It has no “sonic” or technical purpose — a 250 ohm 3 W resistor would have done the same job — but this way I hit two birds with one stone. The lamp will also last a long time, being run at only about 3 W (and I have spares anyway).

In the power supply I used an 84 / 6Z4, a small rectifier from the same era as the UY227 — although it has a dome bulb (the pear-shaped “98” version is unfortunately unobtainable), while the 84 is fairly common. It can supply 60 mA, more than enough for the two 27s, which draw 5 mA each. The 84 is the forerunner of the 6X5G/GT; electrically they are the same, differing only in base type. The rectifier is followed by a CLC cell formed by a NOS Philips dual electrolytic 50+50 µF and a 15S55 20 H choke. Why the old capacitor? Simple: when measured, its ESR and dissipation factor are one-tenth those of typical industrial caps with comparable values. Surely there are “audio-grade” caps you can buy for around €20, but I have a sack of these new; I’ve used tons of them in radio restorations and I know they work perfectly, are reliable with excellent electrical characteristics, and they also match the aesthetic I wanted.

After the electrolytic there’s an additional RC cell with a 20 ohm resistor and a 2 µF Siemens paper-in-oil capacitor — a technique used by the great Mariani to decouple the electrolytic at mid/high frequencies and let the paper-in-oil “speak.” In the photo below you can see the little silver brick.

Assembled and finished…

The knob isn’t a volume control — it’s just the power switch. Unfortunately, on first power-up the thing I feared happened: the power transformer induced a slight hum audible from the speakers, very low but still irritating. With music playing you didn’t notice it, but it bothered me. I should have used a wider chassis to space the power transformer farther away, but I couldn’t get one during lockdown. If the small commercial line transformers had worked, I probably wouldn’t have had any problem — I would have mounted them on the left, far from the power transformer. I held out a few days; the Nibiru effect was there, but I couldn’t stand that background HUMM anymore, so in the end I decided to disconnect the two line transformers, use a CCS as the plate load, and take the signal directly from the plate — even if that meant no longer having unity gain; I could just turn down the power amp volume…

Unlike a pure resistor load, a CCS does not lose all the benefits you get from an inductive load and probably preserves more detail. I added two small trimmers to balance the channels because the feedback is obviously minimal (as explained in the Nibiru project) and can’t perfectly equalize channel levels. The UY227s are nearly a century old, and expecting a perfect match is, frankly, wishful thinking — solvable only by spending a lot (with no guarantee the match will hold). But I had these two at home; the channel difference was only about 0.2 dB, so adding that 2.2 M? trimmer for a micro-adjustment is nothing that would compromise sound quality. In the end, this “centenarian” DAC has a truly airy, very clean, and brilliant sound. Compared with other builds of mine, the treble is smoother — perhaps thanks to the ICEL coupling caps or the paper-in-oil in the PSU. I want to try paper-in-oil on the signal path too — I have them — but in due time (I also have to work 🙄 ). So even if it remains a box with 6 kilos of disconnected line transformers inside serving only as decoration, I’m still satisfied with the result. I won’t fail to try other approaches in the future — probably a parafeed in pure Nibiru style using my 18S100, or rebuilding the same project with better transformer spacing (since the 26 prototype didn’t have this issue).

Here is the premium schematic for those who want to build my circuit. The kit to build this DAC consists of a power transformer and a choke; the two SB-LAB line transformers are also available, but you must install them at least 15 cm away from the power transformer. You can also use other DACs or add RCA inputs to connect an external source. The circuit gain is 15 dB when configured with the CCS or a choke load.

More photos of my build

Some measurements… Below is the spectrum analysis of the Sabre 9023P DAC alone, showing virtually zero distortion:

This, instead, is the spectrum of the circuit with the UY227 at 30 Vpp output amplitude, showing 0.18% distortion…

The bandwidth of the circuit with the UY227 loaded by the CCS is 10 Hz 0 dB – 40 kHz –1 dB…