Skip to content
FOURBILLION
← Work

4B Core

An embedded audio compute platform for digital pedals — specified, reviewed and taken to schematic.

A purpose-built embedded audio board. Designed against the actual constraints of a pedal — enclosure size, power, latency, cost, and the fact that it will be stood on.

Rev A layout target. Codec top left, MCU centre, analogue I/O along the bottom edge.
Block diagram — two power domains, true-bypass relay, STM32H750 with external SDRAM and QSPI flash.
Fitted in a 125B with room for the jacks and a 9V supply.

The brief

Take studio-grade digital effects, the kind that normally live in a plugin, and put them in a pedal that fits a 125B enclosure. It runs off a standard pedalboard supply, it gets stood on, and it can't lose the dry signal when the power drops. Development kits were never meant to ship in a product. Closed modules make you design your product around someone else's decisions. So this is the board we built instead.

We ran it the way a client engagement should run: spec first, decisions written down, every part checked against primary sources, and an independent review before any money goes on a board spin.

Architecture

  • STM32H750IBT6 (LQFP176): a 480 MHz Cortex-M7. We moved up from an LQFP144 after the datasheet showed the smaller package doesn't break out the SDRAM controller. Checking the datasheet itself, not the pin count, caught it before pin mapping.
  • 32 MB SDRAM + 8 MB QSPI flash: about 170 seconds of stereo float buffer at 48 kHz, enough for long reverb tails and pitch buffers. The flash part matches Daisy Seed so we can reuse an existing, proven programmer loader.
  • WM8731 codec with an OPA1656 instrument-level front and back end.
  • Relay true bypass, fail-safe by hardware. The relay coil runs from the 9V analog domain, not the 3.3V logic rail. When 9V drops, the pedal falls back to bypass on its own, whatever the firmware is doing.
  • Two power domains. USB can power logic for development, but never the analog path. Development hardware behaves like production hardware.

Review before fabrication

A Rev A schematic review found three blockers:

  1. A bias capacitor on the guitar input was shunting the signal to ground.
  2. The bypass relay wiring didn't actually give true bypass.
  3. The codec's DCVDD requirement had been misread.

It also found several lower-severity issues. All of them were fixed on paper, in Rev 0.1, before any board was ordered.

Fitness-for-purpose: porting a real product

To test whether the platform could carry a real product, we wrote a port specification for Stairwell, a GOREBANG reverb plugin, against the board as designed. The DSP ported cleanly: JUCE-free C++17 with 14 MB of delay lines against 32 MB of SDRAM. The spec also turned up two hardware blockers that the platform review had missed:

  • Stereo relay drive: two relay coils overdraw their regulator, so the coils never reach must-operate voltage.
  • Power jack polarity: the board was specified centre-positive, but pedalboard supplies are centre-negative.

It also recorded a CPU estimate that still has to be measured on target before anyone relies on it. Finding problems like these on paper is the point of doing the spec.

Status

Architecture frozen at spec v0.6. Schematic capture is under way in KiCad, sheet by sheet, each with a written design note and bill of materials. Hardware design files will be published under an open hardware licence.