MCU SPI

shm-edge-sensor

Event-triggered structural health monitoring pod. A geophone and an accelerometer watch for impact events; an FPGA buffers the waveform while the microcontroller sleeps, and the MCU wakes only when something happens, classifies the event, and reports it.

Layout

KiCAD-MCP-Server/ may be present locally. It is a clone of a third-party tool and is deliberately gitignored — it is not part of the build.

The board in one paragraph

MCU_SPI is a 3.3 V board carrying an ADXL362 accelerometer and an ADS1220 24-bit ADC on one SPI bus, a DS3231M RTC on I2C, and a geophone front end. Its distinguishing feature is a discrete wake latch: a 74LVC1G00 NAND (U4) fed by the accelerometer interrupt and the RTC alarm clocks a 74LVC1G74 flip-flop (U5), whose output both signals WAKE_STATUS to the MCU and switches the +3V3_SW rail through a DMG3415U P-FET (Q1). One event latches a wake and powers the ADC; the MCU clears the latch to go back to sleep.

MCU_SPI/configa.cpp is the machine-readable description of all of that — pinout, bus topology, timing, and a behavioural model of the latch.

Verifying the hardware description

cd shm-sensor-pod

# schematic still matches configa.cpp?
python3 tools/netlist_from_sch.py --regen

# configa.cpp's own arithmetic and pin-map guards
c++ -std=c++17 -Wall -Wextra -DCONFIGA_SELFTEST MCU_SPI/configa.cpp -o /tmp/ct && /tmp/ct

# board glue running in Renode
cd renode && renode wake_demo.resc

The checker exists because the netlist exports in this project silently went stale once already, describing a board revision that no longer existed. Run it after any schematic edit.

Known issues

1. INTMAP1 bit 7 (INT_LOW) is mandatory on the ADXL362. The NAND acts as a negative-logic OR and needs both inputs idle-high. The ADXL362 defaults to an active-high, idle-low interrupt, which pins the NAND output high and stops the RTC alarm from producing any clock edge at all — the node never wakes. renode/wake_demo.resc reproduces this.
2. ~34.5 ms blind window after every wake. VMID is biased from the switched rail, so C13 recharges through R6‖R7 on each wake. That is longer than the transient being measured. Fix is to bias VMID from the always-on rail.
3. Anti-alias filter is far too wide. ~75.8 kHz gives only ~11 dB at the ADS1220 modulator frequency. C19 → 33 nF and C20/C21 → 3.3 nF puts the corner near 2.3 kHz and buys ~30 dB.

Open question: firmware target

mcu-firmware/ targets an ESP32 (platformio.ini, main_esp32.ino, WiFi + TinyML). renode/ models the board against an nRF52840 (Arduino Nano 33 BLE), because Renode has no ESP32-S3 SoC model — only the Xtensa ISA — whereas the nRF52840's SPI, I2C and GPIO are all properly modelled. Pick one before writing much more firmware.