SAK AUDIO · ENGINEERING RECORD

Finding the missing low end

The original circuit imposed a modest passive bass loss. Its drive mapping then pushed the core into much larger low-frequency compression. The revision corrects those two causes separately. No compensating shelf EQ is used.

Correct the circuit

Remove unsupported 2–5 µm joint gaps; use an explicit 50 Ω line source. Copper, winding geometry, hysteresis, and magnetic losses remain.

Rebalance excitation

Most knob travel now covers reasonable transformer voltage. The extreme endpoint still drives the same core into saturation.

Keep compression

Static calibrated gain replaces OUTPUT. It applies no envelope tracking and cannot restore a transient that the magnetic circuit compressed.

Before and after

Separate small-signal response from magnetic compression

The linear probe holds the internal source at 0.5 mV peak at every IRON setting. The other sweeps use a fixed digital input level. A 96 kHz host and 4× oversampling allow measurements through 40 kHz.

Input level

Original engine frequency response — Original engine, including its old 0–12 dB drive makeup.

Revised engine frequency response — Revised production engine, including static automatic compensation. Broadband offsets are level calibration; changes in relative frequency response come from the circuit and magnetic compression.

Revised engine THD versus frequency — THD includes only representable harmonics below Nyquist. Missing high-frequency points cannot be interpreted as zero distortion.

Nonlinear compression relative to the small-signal response — Large-signal fundamental gain minus the linear probe, at the same IRON setting. This directly separates compression from passive LF attenuation.

Passive loss: the controlled experiment

20 Hz, 0.5 mV source peak	Nickel	Alloy	Steel
Original JA circuit	−0.102 dB	−0.430 dB	−0.813 dB
Replace core with linear initial Lm	−0.102 dB	−0.430 dB	−0.813 dB
Remove gap only	−0.028 dB	−0.164 dB	−0.255 dB
Remove gap + use 50 Ω source	−0.011 dB	−0.063 dB	−0.095 dB

Nominal resistive insertion compensation only; no autogain scalar. Initial inductance changes from 11.98 / 6.08 / 5.70 H to 25.45 / 10.18 / 12.01 H. The retained steel eddy term contributes a small broad loss.

Analytical response for the original circuit and controlled modifications — Independent continuous-time circuit calculation. These predictions are checked against the exact C++ linearized and JA circuits.

View every component substitution

Source, load, gap, copper, leakage, capacitance, eddy, ideal-source and turn-count experiments

Original component data · Revised component data

Magnetizing impedance and incremental inductance

Incremental inductance across magnetic loops

Winding voltage, magnetizing current and hysteresis

FIR response and aligned dry/wet path

Oversampling filter gain and dry wet response

The FIR-only 5–100 Hz error is negligible. Its response does not explain the original bass shelf.

THD versus input level

TUBE and BUS regressions

TUBE, BUS, and the shared JA kernel were left unchanged. Their frozen/current audio is identical. BUS already retains flat bass; TUBE's wider amplifier and speaker circuit has a different bandwidth objective. The existing steel minor-loop fit failure and BUS stored-baseline discrepancy remain documented.