The Dynamics of Return

Analytical analysis of subsonic instabilities and phase rotations in the negative feedback loop under complex reactive loads

by Andy Roman

In the design of thermionic tube amplification systems, the standard academic approach commits a systematic modeling error: treating the electroacoustic transducer (loudspeaker) as a purely resistive load. In linear simulations and traditional test benches, the speaker is reduced to a pure scalar value (for instance, a dummy load of 8 or 16 Ohm used as a laboratory standard).

In real physics, the loudspeaker is a highly reactive load governed by a time-varying complex impedance Z(s), where s = σ + jω represents the complex frequency in the Laplace domain:

Z(s) = Rdc + sLe + Zmotor(s)

The motional component (Zmotor) is the direct result of the Inverse Electromotive Force (Inverse EMF) generated by the movement of the voice coil immersed in the magnetic field. This accumulated kinetic energy converts back into electrical energy and is projected straight into the secondary of the Output Transformer (OT). This interaction mutates the plate load line of the power tubes frame by frame, invalidating any static calculation.

The Illusion of the Resistive Dummy Load

The fetishism of the 1 kHz square wave displayed on an oscilloscope through a bank of chassis-mounted power resistors is a methodological fallacy. A purely resistive load possesses no mechanical inertia, generates no pneumatic counter-pressure, and returns no magnetic energy to the circuit.

Achieving perfect wave linearity on a fixed resistor is an elementary textbook exercise that completely ignores dynamic phase rotations. Anyone who relies exclusively on this test is optimizing the transfer behavior of the circuit to power an electric heater, not to reproduce the complex transient of a musical instrument.

Polarized Approaches and the Loop Gain Equation

The current state of the art is polarized between two opposing schools of thought, both suffering from structural limitations:

1. Hyper-Feedback (Overdamping)

Driven by the obsession with stabilizing the response on a dummy load, many technicians massively increase the feedback ratio (β). Examining the closed-loop transfer function:

Af(s) = A(s) / [1 + A(s) * β(s)]

Linearly increasing β to depress the output impedance and artificially raise the Damping Factor leads to dynamic overdamping. This approach crushes the natural inter-electrode elasticity of the tubes and degrades the rise time (tr). When the loop is stressed by the actual transients of a real speaker, the excessive feedback ratio saturates the input nodes of the phase inverter, resulting in a flat, lifeless, compressed sound that lacks temporal micro-contrast.

2. Open NFB and Uncompensated "Raw" Controls

At the opposite extreme, completely removing the negative feedback or inserting a linear Raw control (typically a 50 kOhm potentiometer in series with the feedback resistor to parzializza the loop) introduces destructive systemic vulnerabilities.

At low signal levels, the open-loop system exhibits an apparent sonic openness. However, as soon as the overall gain increases and the power stage leaves its linear zone, the loop gain function becomes strictly dependent on a vectorial phase shift term:

A(jω) * β(jω) = |Aβ| * e^(jθ(ω))

The Subsonic Collapse Mechanism at High Volume

When the signal amplitude approaches saturation, the parasitic inductances and inter-winding capacitances of the output transformer combine with the reactive behavior of the speaker. In open-loop or uncompensated Raw network systems, this phenomenon shifts the system poles toward the right half-plane of the Laplace complex plane (s), violating the Nyquist Stability Criterion.

The onset of this dynamic instability is not limited to classic motorboating (low-frequency relaxation oscillation). The real hidden issue is the appearance of a fixed, ultra-low frequency noise, which is a parasitic subsonic saturation (often confused with phase inverter blocking distortion).

When the guitarist strikes a hard note at high volume, the returning EMF energy finds no temporal alignment in the loop. The potential at the feedback node fluctuates uncontrollably, asymmetrically saturating the grids of the phase inverter (Long-Tail Pair). The amplifier's transient response time instantly collapses: the attack "sits down," the wavefront is destructively compressed, and spectral separation becomes muddy and indistinct.

The Atelier Solution: Dynamic Phase Tuning

Our design philosophy resolves this breaking point by rejecting both the rigidity of industrial overdamping and the unstable anarchy of conventional Raw controls.

Instead of configuring the NFB network as a simple static amplitude attenuator, we have implemented a temporal and vectorial linearization network. By acting on the coupled time constants at the output of the Long-Tail Pair and integrating reactive compensation into the return loop, the phase vector is forced to remain confined within an extremely tight, controlled envelope around the original impulse.

This stability prevents the shifting of poles in the Laplace plane, even when the plates operate under maximum current excursion and the output transformer approaches core saturation.

Conclusions: Pure Sag and the Preservation of Attack Time

By isolating phase behavior through a rigid, stiff power supply, the sag effect ceases to be a chaotic factor of electrical instability. The controlled sag of the plate voltage manifests as it should: a purely acoustic, smooth, organic compression shaped by the musician's touch.

The true revolution introduced by this physical approach lies in the preservation of transient response time. The amplifier suffers no delay or subsonic collapse at high volume; it maintains maximum attack speed and ruthless dynamics, ceasing to fight against the speaker's reactivity and fully embracing the physics of the audio signal.