The inverted picture — shunt regulators
When constant total current beats variable pass-tube current.
Flip everything: instead of varying the pass tube's current to absorb the difference between Vraw and Vout, vary a parallel tube's current to keep the total drawn from Vraw constant. That's a shunt regulator — and it has a few unique tricks.
A series ballast R1 drops most of Vraw − Vout. A shunt element (VR tube, triode, or a diff-amp + triode) hangs off Vout to ground, drawing whatever current the load isn't. IR1 = Iload + Ishunt stays roughly constant. Vout is wherever the shunt element clamps it.
- Bidirectional current: the shunt can SINK current — handy for inductive loads where back-EMF would force a series pass tube into cutoff.
- Constant total bus current: Vraw doesn't change with load, so PSU ripple stays predictable.
- Naturally low noise: the shunt operates at near-fixed current, so its own noise contribution is minimised.
The simplest shunt: R1 + one VR tube. The derivation below walks through the total-current invariant, R1sizing, and the VR’s worst-case dissipation.
Shunt VR-only — derivation
The series ballast R1 sets a (nearly) constant bus current. The VR tube absorbs whatever the load does not draw. That gives the cardinal invariant:
Solve for R_ballast — the only sizing knob the designer has:
A VR tube needs a minimum current to stay struck. Size I_total so the worst-case (highest I_load) still leaves the VR above I_VR,min:
Worst-case dissipation in the VR happens at NO load — the VR alone carries the entire bus current:
Ballast resistor dissipation is constant (the upstream price):
Output impedance is dominated by the VR’s dynamic resistance r_d (≈ 100–200 Ω for an 0A2):
Ripple is divided by the ballast / r_d ratio:
Takeaway: a bare-VR shunt is a low-pass filter for noise plus a hard reference for V_out, but dissipation is permanent. Always vet P_VR,max before picking a tube.
Swap the VR for a power triode (6080, 6AS7) and Zoutdrops dramatically — but the tube dissipation can shoot up at no load.
Single-triode shunt — derivation
Same invariant as the VR version: the ballast R1 enforces a (nearly) constant total current; the triode V1 absorbs the slack between I_total and I_load:
The triode bias point: cathode held at V_ref by the VR tube, anode at V_out, grid driven from a divided sample β·V_out — so V_gk stays a few volts, not the full V_ref:
Quiescent current from the triode small-signal parameters:
Output impedance — the triode acts like a current sink of internal resistance r_p; the cathode sits on the stiff VR reference, so a rise in V_out raises V_gk → more shunt current → V_out is pulled back:
If the grid follows V_ref directly (passive bias, no comparator), there is no error-amplification — the loop gain is just μ from grid to anode current:
Worst-case triode dissipation occurs at zero load (the triode swallows all of I_total):
Compare with P_diss,max of the chosen tube. For a 6080 / 6AS7 the rating is 13 W per section, ≈ 26 W per envelope.
Takeaway: Z_out drops by ~μ vs the bare VR, but with no comparator the line regulation is still modest — the diff-amp variant adds the loop gain.
Close the loop with a 12AX7 long-tailed pair. The loop gain T divides Zout and lifts PSRR by the same factor — the derivation works it all out and compares shunt vs series efficiency.
Diff-amp-driven shunt — derivation
The geometry and the bus-current invariant are unchanged — what changes is the feedback path. The diff-amp watches a divided sample β·V_out and reacts to any drift versus V_ref.
Feedback divider ratio sets the static V_out:
Open-loop gain of the diff-amp into its anode load Ra (the 12AX7 with μ = 100 is so close to a perfect amplifier that we take A ≈ μ here):
Loop gain — open-loop times the feedback ratio:
Closed-loop output impedance — the shunt triode’s native Z divided by 1+T:
Ripple rejection — passive (ballast/r_d) times the loop gain bonus:
A series regulator only burns what its load draws:
A shunt regulator pays for I_total whether the load is there or not:
Equality holds only when I_load = I_total — i.e. the shunt tube is starved. In normal operation the shunt is strictly less efficient.
With the bench numbers above:
- — all the power dissipates in R1 + the shunt tube.
Takeaway: shunt is worth it only when I_load is rock-stable. Its strengths are micro-ohm Z_out, constant upstream bus current, and current-sinking — never raw efficiency.