- README, CHANGELOG, RELEASE_NOTES, THIRD_PARTY_NOTICES and docs/*.md translated to English; cross-markdown anchors updated - C/C++/Python/shell comments and user-facing messages translated - identifiers, protocol names, commands and literals unchanged - native tests re-verified: 24/24 PASS; tpms_frame.py self-test PASS
5.6 KiB
Validation
One command without hardware
bash tools/check_all
The script runs the Python self-test, native tests, the full ESP32 build,
cppcheck, fetching the pinned sources (on first run), and CU8 decoding by the
real rtl_433. Logs: artifacts/validation/. Every stage has pipefail,
and an error ends the script with a non-zero code.
A reproducible alternative:
docker build -t tpms-rc .
docker run --rm -v "$PWD/artifacts:/work/artifacts" tpms-rc
Results from 2026-10-28
Environment: Linux, Python 3.14.4 on the host test; PlatformIO 6.1.18;
espressif32 8.0.0; Arduino ESP32 3.3.12; SmartRC CC1101 commit
de6757ef633b3e626dae5c7835cdb66ab53f9a1d; rtl_433 commit 406f9a3…;
rtl_433_tests commit 1aa8c29…. Full SHAs are in PROTOCOL.md.
| Command | Result |
|---|---|
pio test -e native |
24/24 PASS — frame, four independent vectors, negative tests, NVS, console parsing (NaN/ranges), fake clock/radio, Clone IDs, LEARN, SINGLE, ACC, restart |
pio run -e esp32s2mini |
PASS, own warnings treated as errors (-Werror) |
pio check -e native ... --fail-on-defect=low |
PASS, 0 defects after the struct initialisation fix |
python3 tools/tpms_frame.py --self-test |
PASS, capture + 500 frames |
tools/verify_rtl433.sh |
PASS, 64/64 reference records; 3/3 synthetic; 3/3 negative rejected |
git diff --check |
PASS |
bash tools/package_release.sh |
PASS, ESP32-S2 image with the bootloader at 0x1000 and a SHA-256 manifest |
Reference data: each of gfile001…gfile004 produced exactly 16 matching
records, with the expected ID/flags/pressure/temperature. The synthetic frames
2FC871/260 kPa/73 F, 3495AC/220 kPa/23 F and ABCDEE/262.5 kPa/75 F were
decoded by an independent rtl_433. Damaged Manchester, a wrong sum, and
truncation produced no EG53MA4 record.
Coverage matrix
| Layer | Automated | Physical RTL-SDR | Physical CC1101 | Car |
|---|---|---|---|---|
| fields, checksum, quantisation | done | not needed for the field | not needed | confirmation of ECU interpretation |
| Manchester/preamble/length | done | not done | not done | not done |
| real recordings → rtl_433 | done, 64/64 | n/a | n/a | n/a |
| synthetic CU8 → rtl_433 | done | n/a | bypasses hardware | n/a |
| CC1101 registers/tail/power/frequency | code audit | required | required | required |
| NVS, IDs, intervals, modes | done; NVS emulated with a blob | a restart test is required | required | required |
| LEARN FL→FR→RR→RL | state machine PASS | auxiliary | required | not done |
| consumption/brownout/ACC | ACC in the state machine; no measurements | n/a | required | required |
There is no basis for a "validated" label. Gitea Actions has a workflow, but the runner state and the remote CI result must be checked; a local PASS is not pretend CI.
HIL checklist — requires a user and hardware
CC1101 + analyser/RTL-SDR
- Read the registers after initialisation: OOK, 433.92 MHz, fixed 30 B,
SYNC_MODE=0, no CRC/Manchester/whitening/FEC/append. - On GDO0/ANT confirm exactly 30 B, a ~120 µs chip, a full last chip, and no hardware preamble/sync/CRC/extra bytes.
- Record TX-ERR/underflow; there should be none on a correct RF path.
- Measure the frequency and relative power at −10 dBm; do not assume the calibration of a cheap module. Do not raise the PA unnecessarily.
send 1…send 4: RTL-SDR decodes the correct IDs, 260 kPa and flags.- Repeat after a reset and 10 deep sleep cycles.
NVS and inputs
- Set your own IDs/pressures/RF, disconnect power for >1 min, and re-check
infoand transmission. - Verify rejection of a duplicate ID, a bad power, and a bad interval.
factory YESrestores 260 kPa and the default RF.- Brownout/crank on a protected supply: no hang or configuration damage; write settings only after the supply stabilises.
- If ACC is installed: measure the GPIO voltage for OFF/ON/transients, check the intervals and automatic resumption after ACC ON. Skip without ACC.
The specific Opel Astra J
- Confirm the profile via VIN/markings/diagnostics before transmitting.
- Clone IDs: reception of the four known IDs without assuming learning will vanish.
- New IDs: full learning FL→FR→RR→RL with one active wheel.
- DTC and TPMS data read before/after; record the ECU version and procedure.
- Position test: centre console/behind the glovebox, possibly the boot; while parked, without guessing the receiver location.
- A tyre gauge check of all wheels before and after. Do not perform the test as a road drive with the simulator in place of a working TPMS.
Artifacts and flashing
bash tools/package_release.sh
The artifacts/release/ directory:
| File | Address |
|---|---|
bootloader.bin |
0x1000 — the address for ESP32-S2, not 0x0 as in ESP32-C3 |
partitions.bin |
0x8000 |
boot_app0.bin |
0xE000 |
firmware.bin |
0x10000 |
firmware-full.bin |
0x0 — the full 4 MB image |
firmware.elf |
debug/symbols, do not flash it |
SHA256SUMS, BUILD-INFO.txt |
integrity and version check |
The full image erases the previous flash while writing the whole area; NVS in
that partition may be lost — write down your IDs before updating with the
full image. Updating only the application firmware.bin at 0x10000
usually preserves NVS, but before a release the user should test it on their
own module.
PORT=/dev/ttyACM0 bash tools/flash_release.sh
(cd artifacts/release && sha256sum -c SHA256SUMS)
Do not flash an image from another toolchain blindly. The RC contains no physical update test or car test.