Agent-Skills.md

Agent Skills: OpenNIRScap Build

Build an OpenNIRScap 24-channel fNIRS brain cap from open-source hardware — Altium conversion, BOM sourcing, PCB fab, assembly, firmware flash

UncategorizedID: plurigrid/asi/opennirscap-build

Repository

plurigrid/asi
plurigridLicense: MIT
165

Install this agent skill to your local

pnpm dlx add-skill https://github.com/plurigrid/asi/tree/HEAD/plugins/asi/skills/opennirscap-build

Skill Files

Browse the full folder contents for opennirscap-build.

Download Skill

Loading file tree…

plugins/asi/skills/opennirscap-build/SKILL.md

Skill Metadata

Name
opennirscap-build
Description
Build an OpenNIRScap 24-channel fNIRS brain cap from open-source hardware — Altium conversion, BOM sourcing, PCB fab, assembly, firmware flash

OpenNIRScap Build

Open-source 24-channel fNIRS wearable brain cap. Kim et al. 2025, University of Toronto.

Source

  • Repo: https://github.com/tonykim07/fNIRS
  • Paper: https://arxiv.org/abs/2505.20509
  • License: BSD-3-Clause
  • Hardware: Altium Designer files (PcbDoc, SchDoc)
  • Firmware: STM32 C (STM32L476RET6)
  • Software: Python Flask-SocketIO GUI

Architecture

  • 24 detector modules (VBPW34S photodiode + AD8618 TIA, 25mm circular PCB)
  • 8 dual-wavelength source emitters (VSMD66694: 660nm + 940nm)
  • 1 ECU (STM32L476 + 8x TMUX1104 analog MUX + PCA9685 PWM, 112x112mm PCB)
  • 35mm source-detector separation (cortical depth sensitivity)
  • 1 kHz effective sampling, 52.3 dB SNR, 12-bit ADC
  • USB isolated (ADUM4160), LiPo powered (1100mAh, >5hr)

Blocker: Altium to Gerber Conversion

Repo has Altium .PcbDoc source files, no exported gerbers or CSV BOMs.

PCBs to convert

| Project | File | Description | |---------|------|-------------| | fNIRS_sensor_module | SensorModulePCB.PcbDoc | 25mm circular, combined emitter+detector | | fNIRS_ECU | fNIRS_PCB.PcbDoc | 112x112mm main control board | | DetectorOptode | DetectorOptode.PcbDoc | Detector-only variant | | SourceOptode | SourceOptode.PcbDoc | Source-only variant | | STLINK_Breakout | STLINK_Breakout.PcbDoc | Programming adapter |

Conversion approaches (in order of preference)

  1. KiCad 8 built-in importer — File > Import > Altium .PcbDoc. Best fidelity. Needs KiCad installed.
  2. altium2kicad (Perl, 933 stars) — perl convertpcb.pl file.PcbDoc. Repo: thesourcerer8/altium2kicad. Needs unpack.pl first.
  3. Altium 365 Viewer (free web) — View online, manual export. No CLI.
  4. GitHub Issue — Ask authors to export gerbers. Repo is active (last push 2026-03).

Attempted and failed

  • altium-cli (Rust, akiselev) — broken workspace deps, won't build
  • pyaltium — crashes on Python 3.14 (missing dateutil dep)
  • altium2kicad unpack.pl worked but convertpcb.pl couldn't find unpacked files

Next step

Install KiCad 8 (brew install --cask kicad or nix shell nixpkgs#kicad), import PcbDoc, export gerbers + BOM + CPL from KiCad.

BOM — Semiconductors (DigiKey)

| Part | MPN | Qty | DigiKey URL | |------|-----|-----|-------------| | Photodiode | VBPW34S | 24 | https://www.digikey.com/en/products/detail/vishay-semiconductor-opto-division/VBPW34S/4073399 | | Quad op-amp | AD8618ARUZ | 12 | https://www.digikey.com/en/products/detail/analog-devices-inc/AD8618ARUZ/1993937 | | Dual LED 660/940nm | VSMD66694 | 8 | https://www.digikey.com/en/products/detail/vishay-semiconductor-opto-division/VSMD66694/7681025 | | N-ch MOSFET | BSD840NH6327XTSA1 | 16 | https://www.digikey.com/en/products/detail/infineon-technologies/BSD840NH6327XTSA1/5409567 | | MCU | STM32L476RET6 | 1 | https://www.digikey.com/en/products/detail/stmicroelectronics/STM32L476RET6/5177041 | | Analog MUX | TMUX1104DBVR | 8 | https://www.digikey.com/en/products/detail/texas-instruments/TMUX1104DBVR/9685876 | | PWM/IO | PCA9685PW,118 | 3 | https://www.digikey.com/en/products/detail/nxp-usa-inc/PCA9685PW-118/2034325 | | USB isolator | ADUM4160BRWZ | 1 | https://www.digikey.com/en/products/detail/analog-devices-inc/ADUM4160BRWZ/1861340 |

BOM — Passives (0603 SMD, from schematic)

Core known values (exact values need Altium schematic export):

  • 60.4k 1% x24 (TIA feedback)
  • 200pF x24 (TIA bandwidth, fc~13kHz)
  • 100nF x60 (bypass, 2 per board + ECU)
  • 4.7k x4 (I2C pullups)
  • LED current limit resistors x16

BOM — Amazon

| Item | URL | |------|-----| | 3.7V 1100mAh LiPo JST | https://www.amazon.com/Qimoo-Battery-Rechargeable-Connector-Electronic/dp/B0CNLNGBT4 | | ST-LINK V2 programmer | https://www.amazon.com/ST-Link-Programming-Emulator-Downloader-Random/dp/B08YZ4K3Z5 | | 8-pin ribbon cable IDC | https://www.amazon.com/DMiotech-Ribbon-Digital-Cameras-Computers/dp/B0CJYV768J | | PLA filament 1.75mm | https://www.amazon.com/ELEGOO-PLA-Filament-1-75mm-Printers/dp/B0C14PXRZH | | Velcro strips | https://www.amazon.com/VELCRO-Brand-Sticky-Fasteners-Perfect/dp/B00006IC2L | | 3M Micropore tape | https://www.amazon.com/Micropore-Medical-Dispenser-Friendly-NonSterile/dp/B00PZ2F1Q6 | | Neoprene headband | https://www.amazon.com/Headband-Neoprene-Hairband-Non-Slip-Snorkeling/dp/B0FB8Y968R |

BOM — PCBs (JLCPCB, after gerber export)

  • 24x sensor module (25mm circular)
  • 1x ECU (112x112mm, 4-layer)
  • 1x STLINK breakout

Cost Summary

| Category | Cost | |----------|------| | Semiconductors | ~$130 | | Passives | ~$15 | | PCBs | ~$13 | | Amazon | ~$150 | | Total | ~$310-420 |

Assembly

  1. Export gerbers from KiCad (after Altium import)
  2. Order PCBs from JLCPCB
  3. Order DigiKey + Amazon parts
  4. Solder sensor modules (24x, SMD 0603 + photodiode + LED)
  5. Solder ECU (QFP-64 MCU, SOT-23 MUXes, TSSOP PWM)
  6. Wire cable harnesses (8 groups of 3 sensors)
  7. 3D print sensor capsules (STL from repo or design in OpenSCAD)
  8. Flash firmware via SWD (ST-LINK)
  9. Connect USB, run Python GUI
  10. Calibrate on forehead, verify heart rate waveform first

bci.horse Integration

  • Tree: bcf-0036 in plurigrid/horse
  • Inventory ID: planned, not yet ordered
  • Role: walk-up fNIRS fleet ($419/unit vs $50k NIRSport 2)
  • Calibrate against Artinis Brite Lite (if purchased)

Firmware

Located in firmware/STM32/fNIRS/. STM32CubeIDE project.

  • MCU: STM32L476RET6, ARM Cortex-M4 80MHz
  • ADC: 12-bit, DMA, 5kHz raw sampling, multiplexed to 1kHz effective
  • USB: CDC serial via ADUM4160 isolator
  • PWM: PCA9685 drives LED sources, alternating 660/940nm
  • SD: SDMMC logging

Software

Located in software/. Python 3.

  • Flask-SocketIO backend for real-time streaming
  • Web GUI: Plotly.js + BrainNet Viewer 3D brain mesh
  • Modified Beer-Lambert Law processing (NIRSimple library)
  • Bandpass 0.01-0.5 Hz for cortical hemodynamics
  • Correlation-based signal improvement for HbO/HbR

Related Skills

  • forester — bci.horse forest publishing (bcf-0036 is this device's tree)
  • reverse-engineering — MCP servers for Ghidra/radare2/IDA if binary analysis needed
  • binary-triage — systematic binary survey workflow
  • performing-firmware-extraction-with-binwalk — firmware extraction from embedded devices
  • protocol-reverse-engineering — reverse engineering communication protocols
  • radare2-hatchery — radare2 ecosystem tools

Altium Conversion Tools

  • altium2kicad (Perl, 933 stars): https://github.com/thesourcerer8/altium2kicad
  • KiCad 8 built-in Altium importer: https://www.kicad.org/blog/2020/04/Development-Highlight-Altium-Pcb-Importer/
  • altium-cli (Rust, broken): https://github.com/akiselev/altium-cli
  • KiCad Forum thread on CLI import: https://forum.kicad.info/t/how-to-import-altium-pcbdoc-via-commandline-tools/50893

Why This Build

The problem

Research-grade fNIRS (NIRx NIRSport 2) costs $50k-150k. PLUX biosignalsplux fNIRS sensor is $789 for 1 channel with short source-detector separation — measures superficial scalp hemodynamics, not cortical activity.

Source-detector separation is everything

Light follows a banana-shaped path through tissue. Separation controls measurement depth:

  • <10mm: scalp blood flow (artifact)
  • 30mm: gray matter (cortical hemodynamics — what we want)

  • 50mm: too much attenuation

Per Strangman et al. (PLOS ONE 2013): every 10mm increase up to ~45mm adds ~4% gray matter sensitivity. Depth sensitivity decays as S = 0.75 * 0.85^depth.

PLUX has emitter+detector in one housing = short path = scalp oximetry. NIRSport 2 uses separate optodes at 30mm = cortical imaging = $50k+. OpenNIRScap uses separate optodes at 35mm = cortical imaging = $419.

Why OpenNIRScap specifically

  • 24 channels (vs PLUX 1ch, vs DIY-fNIRS headband 4ch)
  • 35mm separation (proper cortical depth)
  • 52.3 dB SNR, 1 kHz sampling
  • Validated hemodynamic response during cognitive tasks (paper Fig. 10)
  • $419 total BOM from off-the-shelf parts
  • Fully open source (BSD-3, hardware + firmware + software)
  • Published May 2025, repo active March 2026
  • Comparison from paper (Table II): 24ch/$419/open vs 4ch/$215/open vs 1ch/$789/proprietary

The strategy for bci.horse

Build 3x OpenNIRScap ($1,257) as the walk-up fNIRS fleet. If/when a commercial reference is acquired (Artinis Brite Lite ~$6k or Cortivision Spectrum C23), use it to validate OpenNIRScap signal quality. The AFE4404 in the earlier DIY spec (bcf-0012) uses the same modified Beer-Lambert deconvolution as NIRx's proprietary pipeline — the physics is identical, only the engineering precision differs.

Key references

  • OpenNIRScap paper: https://arxiv.org/abs/2505.20509
  • OpenNIRScap repo: https://github.com/tonykim07/fNIRS
  • ninjaNIRS (200 optodes, openfnirs.org): https://openfnirs.org/hardware/ninjanirs/
  • Strangman depth sensitivity: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0066319
  • NIRDuino (Arduino fNIRS): https://www.medrxiv.org/content/10.1101/2024.12.20.24318425v1
  • FlexNIRS (open-source): https://www.sciencedirect.com/science/article/pii/S1053811922003408
  • DIY-fNIRS headband: https://www.hardware-x.com/article/S2468-0672(21)00033-X/fulltext
  • Lieberman Lab UCLA (NIRx NIRSport 2 user): Miao, Lieberman, Pluta 2025 hyperscanning review
  • Artinis Brite Lite: https://www.artinis.com/brite-lite
  • Cortivision Spectrum C23: https://www.cortivision.com/spectrum-23/
  • PLUX fNIRS sensor datasheet: https://support.pluxbiosignals.com/wp-content/uploads/2021/10/biosignalsplux-FNIRS-Datasheet.pdf
  • bci.horse source-detector physics tree: bcf-0034 in plurigrid/horse
  • bci.horse fNIRS comparison: bcf-0032 (NIRSport), bcf-0035 (Brite Lite), bcf-0036 (OpenNIRScap)