ENGINE=qdelay · src/engine/qdelay/qdelay_engine.{h,cpp} · class QdelayEngine

A second "flavor" of the tempo-synced dual delay, inspired by qdelay. It keeps the delay engine's entire control grammar — tempo-locked musical divisions, feedback, mix, PITCH transpose, ENV tone, the mod LFO, the Freeze/Reverse pad gestures and the three stereo topologies — and swaps only the character palette from Clean/Tape/Shimmer to Clean / Diffuse / Duck. The aim is the smeared, reverb-leaning dub end of the delay spectrum rather than the warble-and-shimmer end.

Because the binding constraint on the hardware is the control surface (a handful of knobs and two 3-way switches), qdelay does not expose the desktop plugin's ~40 parameters. It picks the highest-value, lowest-cost slice — feedback diffusion and ducking — and maps them onto the existing mode switch.

Implementation, the file map, and dev notes live in docs/dev/qdelay-impl.md.

License: unlike the rest of this MIT project, the qdelay engine is GPLv3 — src/dsp/diffuser.h is a port of qdelay's GPLv3 Diffusor, so the engine and any ENGINE=qdelay binary are GPLv3. See src/engine/qdelay/NOTICE.md.

What it shares with delay

The division table, tempo sync, control smoothing, the crossfading two-head PITCH shifter, the Freeze (Play pad) and Reverse (Rev pad) gestures, and the Stereo / DoubleMono / Ping-pong topologies are identical to delay — refer there for those mechanics. Two delay lines (deck A → left, deck B → right) live in borrowed SDRAM (~6 s each). Per sample the path splits into read_color() then write_out() so ping-pong can cross the two taps' feedback and so the stereo diffuser can run on the feedback pair between the read and the write.

Control map

_Generated from docs/diagrams/controls/qdelay.json via make diagrams._

Knob meanings are fixed across characters — the mode only changes the feedback-path treatment.

Characters (ConfigId::Mode, per deck; ring tint blue / teal / amber)

• Clean — clean digital repeats; identical to the delay engine's Clean (bit-for-bit loop when ENV is up).

• Diffuse — the feedback runs through an 8-stage allpass diffuser (src/dsp/diffuser.h, a JUCE-free port of qdelay's Diffusor/TARON MiniVerb). Each repeat is smeared a little more than the last, so the tail thickens into a dense, reverb-like dub wash. The diffuser is a single stereo instance over the two taps' feedback; each deck takes the diffused copy only if it is in Diffuse, so DoubleMono can run a Diffuse deck alongside a Clean/Duck deck. SIZE sets the smear depth (longer division = looser, more spread-out smear). Its buffers (~235 KB at 48 kHz) sub-allocate from the SDRAM arena, so SRAM is unaffected.

• Duck — the heard wet ducks under the dry input. A per-tap fast-attack / slow-release peak follower tracks the input level and attenuates the wet send while you play, so the repeats stay out of the way and bloom in the gaps once you stop. The feedback loop itself is not ducked, so the tail keeps building under a busy input and is revealed when the input drops.

Topologies (ConfigId::Route)

Identical to delay: DoubleMono (two independent mono delays), Stereo (linked to deck A's controls), Ping-pong (linked + cross-feedback, echoes bounce L↔R). In Diffuse/Duck the per-deck colored feedback is what crosses in ping-pong.

Resources

• capabilities() = CapOwnDisplay | CapDualDeck; route() reports the topology for the route LED.

• Display: a division arc tinted by character + a play indicator (white = frozen, cyan = reversed, else green by input).

• Delay lines + the diffuser are sub-allocated from the engine arena (SDRAM); SRAM_EXEC sits at ~77 % (the diffuser does not touch it), comfortably a normal (non-QSPI) build.

• Host coverage: host/test_qdelay.cpp (make -C host test-qdelay) and the standalone diffuser test host/test_diffuser.cpp (make -C host test-diffuser).

Build

make engine-qdelay        # clean build + DFU flash
make -j8 ENGINE=qdelay    # compile only