Build the recorder with evidence.

Choose one complete setup

Every package combines one ADC choice, one EMG capacity, and one IMU family. Use its breadboard plan, firmware, converter, PCB, and documentation together. Mixing packages can silently change GPIOs, bus protocols, timing records, and channel rules.

One recorder, two acquisition paths

EMG and motion enter the ESP32-S3 through different hardware paths. The controller adds its timing evidence, writes the authoritative recording to microSD, and optionally forwards a best-effort copy over Wi-Fi.

Timing is evidence, not magic

ADC frames are simultaneous within the selected converter, but EMG and IMU clocks are independent. The recording preserves counters, trigger or data-ready timing, FIFO order, and read timing so alignment can be characterized later. Filtering and analog stages also introduce delay that must be measured before interpreting millisecond relationships.

Advance through ten measured gates

Build one block at a time. Later diagnostics preserve earlier stages, so a failure can be isolated without guessing across the complete system.

Open an interactive staged build ↗ Read the detailed Markdown guide

Build and analyze the matching profile

Run commands from the selected design package. Expandable packages accept two or four EMG channels; fixed-two packages accept only two. Breadboard builds use GPIO47 for SD CMD, while compact-PCB builds use GPIO38.

Build a two-channel breadboard recorder

cd designs/<selected-configuration>/firmware
idf.py -B build-breadboard-2 \
  -DEMG_CHANNEL_COUNT=2 \
  -DAFO_HARDWARE=breadboard \
  -DAFO_WIFI=1 \
  -DAFO_DIAGNOSTIC_STAGE=0 \
  -DAFO_IMU_COUNT=2 build

Convert a recording

cd designs/<selected-configuration>
python3 -m venv .venv
source .venv/bin/activate
python -m pip install -r host/requirements.txt
python host/convert.py trial.afolog --out exported-trial

Receive the best-effort Wi-Fi copy

python host/live_receiver.py --help
# Join the recorder access point, start the receiver,
# then start recording on the device.

The converter exports timestamped EMG and IMU tables, metadata, quality findings, and basic plots. Recovery mode salvages valid records from an interrupted file and reports damage; it does not invent missing samples. Keep the original binary recording unchanged.

What has passed—and what has not

The repository preserves raw JSON, logs, reports, and manifests. The summary below is intentionally split by evidence type.

Integrated recorder verification · AD7606, two channels, ICM-42688-P

Laboratory firmware 1.12 compiles in nine profiles and passes the host and native tests. All 18 integrated cases pass on its simulation image in Wokwi’s web editor (12 October 2026), checked with the production converter and live decoder. The 1.11 run found a Wi-Fi preview race (a new session’s first data packet could precede its metadata), which 1.12 fixes. The AD7606 was checked against ADI’s Rev. G data sheet. Following ADI AN-1091, the PCB now holds the converter in shutdown until the firmware releases STBY on GPIO14, and a BAT760 diode limits VDRIVE − AVCC while the 5 V rail starts. This is DRC/ERC clean but not yet built. Its ICM-42688-P setup was checked line by line against TDK’s official datasheet (DS-000347 rev 1.6) and the MIKROE carrier schematic: every register write, the 16-byte FIFO packet, the 32/30 µs timestamp tick, SPI timing and supply limits match, and the production driver runs against a register-level sensor model with no rule violations. Two IMU-only bench builds (diagnostic-imu-bench-one and -two) print PASS/FAIL verdicts for identification, sample rate, timestamp step, gravity magnitude and gyroscope bias, with rate gates at the datasheet’s ±2 % clock tolerance; the converter adds a host_time_fitted_us column that removes the sensor clock error from IMU-to-EMG alignment. These are software and datasheet checks; the first measurement on real carriers is still pending. All eighteen v1.6 cases pass actual ESP-IDF execution in Wokwi’s web simulator, with complete exports checked by the converter and live decoder. Nominal saved acquisition has no detected sample loss. The recorder runs acquisition, FreeRTOS, FatFS and UDP together with documented sensor, storage and radio substitutes. The complete visible scene passed an additional nominal run. Earlier failed or incomplete attempts remain preserved; physical performance and a current downloaded VCD trace remain unqualified. The other five designs carry the same ported fixes (firmware dual-1.3 / mpu6050-1.3, 47 profiles rebuilt); their other evidence is unchanged.

Read the integrated test report ↗ Case-by-case results ICM-42688-P setup and bench check Register-model results

Open the complete system ↗

The complete-scene pin observer now verifies 10,041 ADC conversions/reads and both 200 Hz interrupt streams. An independent saved-data audit checks 21,116 record CRCs and CSV rows across two synthetic captures; examples include volts, g, degrees/s, channel labels, timing qualifications and quality reports. The partial derived pin waveform is explicitly separate from the unavailable native full-session VCD.

Read and download the saved data ↗ Inspect digital timing

The figures below describe the earlier six-package release. They do not imply that all six setups have run the new integrated simulation.

Open release gates

• Qualify the exact AD7606 or ADS131M04 lab carrier, including power, modes, logic levels, and loading.
• Measure analog passband, clipping, baseline noise, channel crosstalk, and ADC code scaling.
• Measure IMU rate, drift, interrupt timing, orientation, pull-ups, and cable reliability under SD/Wi-Fi load.
• Complete full-card, write interruption, sustained recording, thermal, and battery-runtime tests.
• Review the compact PCB after the breadboard gates pass; ERC/DRC alone is not an order recommendation.

Detailed validation status · Per-design automated results (JSON) · Wiring validation (JSON)

Keep bench power away from participants

Human recording is a separate research activity. It requires institutional approval appropriate to the study, informed consent, exclusion and stop criteria, an electrical-safety review, a supervised procedure, and a plan for pseudonymous data handling.

The leg and AFO views illustrate placement only. They do not establish patient fit, pressure safety, comfort, biomechanics, or clinical benefit.

Find the right artifact quickly

Package-specific documentation takes priority for exact component values, terminals, build profiles, and numerical limits. Project-level pages explain the shared workflow and evidence boundary.

Detailed documents