leslie.hpp File Reference

The two mechanical pitch-modulators of the electric-organ world: a rotary speaker cabinet and the Hammond console's internal scanner vibrato. More...

#include <pulp/signal/leslie_rotary.hpp>
#include <pulp/signal/scanner_vibrato.hpp>

Go to the source code of this file.

Detailed Description

The two mechanical pitch-modulators of the electric-organ world: a rotary speaker cabinet and the Hammond console's internal scanner vibrato.

They wear similar clothes and are different physics, which is why they are one family umbrella and two focused class headers rather than one class with a mode switch:

• LeslieRotaryT is a real horn and a real bass baffle physically SPINNING inside a wooden box. The rotation produces a genuine Doppler pitch shift (the source moves toward and away from the mic), a tremolo (the horn's beam sweeps past the mic), and a timbral sweep (a horn is brighter on-axis). Two rotors at different rates and different inertias, split by a crossover, recorded by a stereo pair, inside a reflective cabinet.
• ScannerVibratoT has no moving air at all. A passive lumped LC delay line is tapped along its length and a motor-driven rotating capacitor sweeps a pickup across the taps, so the output delay ramps up and back each revolution. A time-varying delay is a pitch shift. Mix the swept output back with the dry organ and the comb interference is the Hammond chorus — a flanger built from a transmission line and a commutator in 1949.

Why a Leslie never sounds like an LFO

Nothing in it is one clean sine, and each of those is a separate structural fact rather than a voicing choice:

• The two rotors run at deliberately different rates, so their modulations beat at the difference frequency — 1.0 Hz at the tremolo defaults, 0.20 Hz at chorale. Locking them equal would remove the single most identifying feature of the sound.
• The two rotors have different INERTIA. The horn is light and arrives at a new speed in about a second; the drum is heavy and takes several, and coasts down slower than it spins up. A speed change is therefore not a ramp, it is two ramps of different lengths — the horn gets there first and the drum smears in behind it. That asymmetry IS the gesture; see set_speed.
• One cosine drives three effects. The facing angle sets the path length (Doppler), the beam gain (tremolo) and the shelf gain (brightness), so they are inseparably locked in the model exactly as they are in the cabinet. Three independent LFOs would let them drift apart, and the sound would come apart with them.
• Two mics at different angles see the horn face them at different phases, which is where the ping-pong swirl comes from — not from a stereo widener downstream.

Series law 1 does not apply here

Stated explicitly rather than omitted: neither engine contains a feedback path. Both are strictly feedforward — modulated delay, parallel reflection taps, an amplitude multiply, and a dry mix. There is no loop for a small-signal gain to accumulate around, so there is nothing to unity-compensate or bound as a loop gain. What the registry needs instead is a CONSTRUCTIVE-SUM bound: the worst case is horn + drum + reflections + dry all peaking together. That bound is measured by the suite's parameter sweep, not estimated — see kWorstCaseGain on each class.

Oversampling: none, and that is a policy, not an omission

Series law 4 asks every module to state its policy. Neither engine contains an aliasing nonlinearity: the amplitude and directivity multiplies are by band-limited control signals, and the crossover, shelves and delay reads are linear. There is no in-band alias source to suppress, so oversampling would cost cycles and change nothing. The Leslie tube amp's growl IS a nonlinearity and is deliberately out of scope — compose a saturator ahead of this node, where it can carry its own oversampling policy.

RT contract

prepare() sizes the delay lines and may allocate. set_*, process(), process_block() and reset() never allocate, never lock, and never perform I/O. All buffers are sized in prepare() from the MAXIMUM of every parameter's range, so no later set_* can ask for a read the buffer cannot serve. State is POD; zero-init is a valid stopped cabinet. The only randomness is the optional seeded mechanical drift, rewound by reset() and never automatable (series law 2).

References: Smith, Serafin, Abel & Berners, "Doppler Simulation and the Leslie", Proc. DAFx-02, Hamburg, 2002 — the Doppler-via-modulated-delay method, the rotating-source geometry, the two-delay-line stereo mic model, the angle-dependent directivity filtering, and the first-reflection filters. The speed of sound is the standard dry-air figure from the ideal-gas approximation c ≈ 331.3 + 0.606·T(°C) m/s at 20 °C. Rotor speeds, inertia times, the crossover corner, the scanner rate and the line-box tap topology are documented hardware BEHAVIOUR implemented as named design parameters — see the honest-gap notes on each constant; no service-manual numbers are reproduced.