DSP TOOLBOX · REVERB

Algorithmic Reverb Laboratory

The same small DSP building blocks can be connected in different ways to produce different reverbs. Compare three simple architectures: Schroeder, Moorer-style, and a small 4×4 feedback-delay network (FDN).

The algorithm is the design. A reverb algorithm is the particular rule-set/topology deciding how delays, feedback, filters, diffusion and summing are connected.

1. Choose the algorithm

Parallel feedback combs create decaying echo patterns, then serial all-pass sections increase diffusion.

Animated pulses are schematic: they show energy moving through the structure, not sample-accurate metering.

2. Shared musical controls

These are the sorts of controls a musician sees on a reverb plugin. Here we expose what they change internally.

Size ≠ Decay. Size mainly changes internal delay relationships; RT60 describes how long reverberant level takes to fall by 60 dB. A large room can be dead. A small reflective space can ring for a long time.
Feedback safety. This teaching model deliberately clamps internal feedback gains below unity. As loop gain approaches 1, energy can persist for a very long time; at or above an unstable loop gain the network can build instead of decay. Real reverb designs also have to manage stability and gain around feedback loops.

3. Watch the internal delay times move

Current delay set

Size
Scales the relationships between internal delay times.
Decay / RT60
Controls how long energy remains in the network; internally this becomes feedback gain.
Diffusion
Controls how quickly discrete reflections become a dense field.
Damping
Removes high-frequency energy in the reverberant path so highs can decay faster than lows.

4. Presets are targets — algorithms are structures

Watch the controls move when you choose a target sound.

Important: “Hall”, “room” and “plate” are not necessarily algorithms. The same broad target can be approached using different computational structures.

5. Three simple architectures

Schroeder

┌→ COMB ↻ ─┐ ├→ COMB ↻ ─┤ INPUT ────┼→ COMB ↻ ─┼→ Σ → AP → AP → OUT └→ COMB ↻ ─┘

Parallel feedback combs generate decaying repeats; all-pass sections increase density.

Moorer-style

EARLY REFLECTIONS ↓ DAMPED COMBS ↻ ↓ ALL-PASS ↓ OUT

Adds explicit early reflections and frequency-dependent damping.

4×4 FDN

D1 ─────── D2 │ ↖ ↗ │ │ × │ │ ↙ ↘ │ D3 ─────── D4 feedback matrix

Several delays feed one another; a matrix decides where returning energy goes.

6. FDN: matrix without the fear

delay outputs → feedback matrix A → delay inputs

A matrix here is simply a table of numbers deciding how much of each delay output is sent to each delay input. Increase Cross-feedback and the network moves from mostly self-feedback toward a normalized mixing matrix, spreading energy between paths without simply adding more feedback gain.

7. What have we actually learned?

DELAY / STOREGAIN / MULTIPLYSUM / ADDFEEDBACKFILTERALL-PASS / DIFFUSION

Same families of operations. Different topology, delay relationships, feedback structure and filtering. Different reverb algorithm.

Next: instead of designing a reverberant space, what if we measure one? That takes us to impulse responses and convolution.