Roadmap

PedalKernel is under active development. This page tracks the significant work items we want to take on next. It is intentionally short — the modeling limits page is a fuller accounting of known gaps.

Modeling frontier

These are the largest open questions in the WDF engine.

Op-amp closed-loop dynamics inside the WDF tree. Feedback topology and gain are now derived from the circuit graph (see the modeling limits page), but gain-bandwidth product, slew rate, and rail saturation still run as a post-processing layer on top of each stage rather than as scattering matrix entries. For most guitar circuits this is inaudible; for precision topologies (servo loops, active filters where GBW is a design variable) it is the next accuracy win available.

ADAA-LUT: antialiased nonlinear roots. The K-method tables currently store a_root = f(b_tree) directly, so they alias the same way Newton-Raphson does at large signal levels. Antiderivative antialiasing (ADAA) reduces that aliasing by integrating the nonlinearity over the sample interval rather than sampling it pointwise — and the LUT variant precomputes both f and its antiderivative F so the per-sample cost is two table lookups and a divide. Layering ADAA-LUT on top of the existing K-method pipeline would let high-gain stages run at audio rate without the 4× oversampling we currently fall back to.

Power supply sag across a pedalboard. Single pedals can opt into a supply model (see the supplies block in the DSL). A shared 9 V supply sagging across several pedals drawing current together is not yet modeled.

Speaker and cabinet physics. The chain currently ends at the pedal output. Thiele-Small driver parameters, voice-coil compression, cone breakup, and cabinet resonance are all missing, and matter for full-rig simulation.

Oversampling latency reporting. Oversampling adds latency that a DAW host needs to know about for dry/wet alignment. The oversampling module does not yet report it.

3D K-tables for pentodes. Pentodes report (true, 3) from k_method_candidacy(), but the table generator currently emits 1D and 2D only. Adding the 3D path (b_tree × Vg1k × Vg2k) would extend the lookup fast path to power tubes.

Bidirectional cross-network coupling. The cross-network coupling layer currently injects donor-stage voltages into recipient-stage incident waves one sample late — back-action from the recipient to the donor lags by 1/fs. For most audio circuits this is inaudible, but for tight servo loops (high-feedback opto-compressors, push-pull pairs that share a phase inverter) the next refinement is delay-free bidirectional coupling — likely a Newton iteration across the network boundary inside one sample.

Transformer-flux metering. The metering transformer_flux[4] field has been a stub since the metering layer landed. The Jiles-Atherton transformer model now carries M (magnetisation) and B (flux density) per sample as part of JaState. Wiring those into the metrics ring buffer for a transformer-saturation indicator is mechanical and shovel-ready.

Faithful LA-2A and Pultec EQP-1A. Both circuits exist in the tree but at different fidelity levels. The LA-2A 1:1 rebuild has the right topology (faithful 4-terminal transformers, detector-tap de-fusion). The Pultec EQP-1A has an ngspice baseline for its passive EQ section. Closing the remaining gaps — feedback-loop convergence on LA-2A masks, validated harmonic spectra on Pultec — is the next milestone for the outboard-gear coverage line.

Coverage

More circuits. There is no fixed target list — we merge new .pedal files whenever they pass SPICE validation. Contributions of well-documented vintage circuits are especially welcome.

Hardware specs (.pedalhw). Voltage-rating and part-number metadata exists for some circuits but is not systematically applied across the library.

Non-goals

A few directions we have deliberately chosen not to take:

Reviewed against commit 222212f .