Tube Amplifier
This is an ongoing build of a stereo OTL (output-transformerless) cathode-follower headphone amplifier. The signal path uses a 6922 dual triode as the voltage amplifier and a 6AS7 dual triode as the cathode-follower output. Rectification is handled by a pair of 6Z4 vacuum tube rectifiers, and an Amperite 6NO45 thermal time-delay relay holds off B+ for 45 seconds at power-up so the rail doesn't hit the circuit while the cathodes are still cold. The version 1 prototype is built on perfboard, and works well, full bench measurements below. A substantially redesigned version 2 exists in KiCad and is covered at the end of this page.
View of prototype amplifier build on perfboard
The input runs through a 100k log dual-gang volume pot and a 33 Ω grid stopper into a common-cathode 6922 stage (47k plate load, 820 Ω unbypassed cathode), which provides all of the voltage gain and inverts the signal. From there it's direct coupled, no interstage capacitor, into the 6AS7 cathode follower, which provides no voltage gain but the low output impedance that lets this amp drive headphones with no output transformer. The follower's cathode sits on a 3.3 kΩ 20 W resistor, and the only capacitor the signal ever passes through is the output cap, a 220 µF electrolytic bypassed with a 0.22 µF film cap, with a 10k bleed resistor behind it. The 6AS7 was originally designed as a series-pass regulator tube, it has low plate resistance (about 280 Ω per section) and a rugged heater-cathode structure, but has high heater current at 2.5 A.
A 369AX power transformer provides the 250 V center-tapped secondary for the 6Z4 rectifiers plus a 6.3 V @ 2 A heater winding. That 2 A winding is nowhere near the full heater budget, the 6AS7 (2.5 A), 6922 (~0.4 A), and the two 6Z4s total roughly 4.5 A, so a second transformer, a Hammond 167S6 (6.3 V @ 10 A), carries the heavy heater load, with all non signal tube heaters running on AC. B+ smoothing is a CLC pi filter with a Hammond 193B choke (12 H @ 100 mA) between the capacitors, and the 6NO45's 3 A contact sits in the B+ line, closing 45 seconds after power-on. Each channel's 6922 plate supply gets an additional RC decoupling stage (4.7k 5 W into 220 µF) off the main rail. This all works and, as the measurements show, it's quiet, but it's heavy, expensive, and eats a massive amount of chassis real estate, which is what's driving the version 2 redesign at the bottom of this page.
I characterized the prototype using the scope's frequency response analyzer (0.4 Vpp stimulus; right channel swept 100 Hz–20 kHz and again out to 10 MHz, left channel 100 Hz–20 kHz), plus FFT and time-domain captures with the generator at 1.5 Vpp. All measurements were taken with the output unloaded, scope probe and the internal 10k bleed only, loaded measurements and output impedance are on the to-do list.
The response is flat across the audio band with the −3 dB point way out at 270 kHz, with no output transformer and no interstage coupling cap, there's very little in this circuit to roll off either end. The channels currently sit 0.49 dB apart, which is enough to hear as a slight image shift. The gang-to-gang tracking error is a prime suspect, alongside section-to-section gain variation in the 6922.
Distortion at 1 kHz with the output swinging 13.9 Vpp (4.8 V rms, unloaded) measures 0.6% THD. For scale, that swing corresponds to about 76 mW into 300 Ω-class headphones, several times more than high-impedance headphones need, though delivering it into a real load is exactly what the upcoming loaded tests will verify. The spectrum is what a single-ended triode chain should produce, second harmonic dominant at −46 dBc, with the third and everything above it below −55 dBc. That even-order-dominant profile is the "tube sound".
At idle the output noise is 4.5 mV rms unweighted, for a signal-to-noise ratio of about 61 dB referenced to the 4.8 V rms test level. Mains-related hum is nearly absent, 60 Hz measures ~0.9 mV and 180 Hz ~0.7 mV at the output. DC offset at the output is −17 mV. There's a small cluster of components around 3.1–3.3 kHz at ~0.6 mV that I suspect is pickup rather than anything in the audio chain.
The most annoying part of this project has been the interdependency of the power supply design decisions. Changing the transformer changes the available secondary voltages, which changes the B+ rail, which affects the operating point of every tube stage, which changes the current draw, which feeds back into transformer and filter component selection. This is the same trap as the liquid rocket engine design, where chamber pressure, injector geometry, and mass flow rate are all codependent, you pick starting values, iterate, and refine.
Version 2 replaces both Hammond transformers with a single custom toroid (Antek) carrying a 250 V CT @ 550 mA secondary for the rectifiers and a 6.3 V @ 10 A winding rated for the full heater budget. Toroids bring a lower stray field, less weight, and a smaller footprint, but much higher inrush current at startup — handled here by a 5 Ω Ametherm SL32 NTC thermistor in the primary. The two 6Z4s stay, now with 180 Ω series resistors on their plates.
The 193B choke is gone, replaced by a capacitance multiplier, the winner over the other candidate, a board-mount inductor, which turned out to be a dead end, multi-henry parts that can carry the DC bias current are either far under-rated or nearly Hammond-sized anyway. The v2 implementation is an IRFBC40 MOSFET source follower, the gate sees the raw rail through a 100k/220 µF RC filter, so the source, the B+ rail, follows a heavily smoothed version of it, with a 15 V zener protecting the gate and a 220k bleed. The gate RC is a ~20-second time constant, so B+ doesn't step in, it ramps in. The penalty is voltage dropped across the MOSFET, dissipated as heat which is manageable at this amp's modest B+ current.
The PCB layout rules were one of the more involved parts of the KiCad work. This is a board where the B+ rail, 120 V mains connections, heater lines, and millivolt-level audio signals all coexist, and the clearance requirements between those nets are very different.