Astro pack: sun / moon / Maidenhead / lat-lon display modes

[peterlewis]

Draft / proposal. A self-contained, opt-in set of GPS-derived astronomy display modes for the Mk IV. Independent of the timekeeping-fix PR — branched from master, touches no timing code. All five modes — and the astro_page_ms paging — have been built and confirmed working on real hardware; the maths is also verified natively (see Build & test). It stays a draft for your review, not for want of testing.

What this adds

Five new display modes, each disabled by default and enabled individually with a MODE_* key in config.txt, exactly like the existing modes. They follow the MODE_SATVIEW pattern: the read-out shows on the 10-character date row while the live clock keeps running on the time row, so you never lose the time to read the sky.

Mode	Date row	Example
MODE_SUN	sunrise / sunset / solar noon (local), auto-paging (default 5.5 s, set by astro_page_ms); RISE/SET/SOL labels are 4-char-padded so the time columns stay aligned	RISE 04.43 · SET 21.21 · SOL 13.02
MODE_SUN_AZEL	sun azimuth & elevation now	AZ179EL62, AZ083EL-29
MODE_MOON	phase index (0–7) + illuminated %	MOON 4 100 (full), MOON 7 16
MODE_GRID	Maidenhead grid locator	IO91xl
MODE_LATLON	latitude / longitude, auto-paging	LAT 51.48 · LON 0.00

The sun page auto-pages (default 5.5 s, set by astro_page_ms); the labels are padded to 4 characters so the
time digits stay column-aligned (note SET/SOL sit one column right of RISE):

RISE 04.43
SET  21.21
SOL  13.02

With no GPS fix the modes use fake_latitude/fake_longitude if set (handy indoors), otherwise show ----. The moon needs no fix at all.

Why these

They reuse data the firmware already has (position + disciplined UTC) for things the existing modes don't surface, and they suit the device's likely owner — a tinkerer / ham / observer: sun & moon position, sunrise/sunset, your grid square, your coordinates.

Design — keeping the maths out of the ISR

sendDate() runs inside the SysTick ISR (the 0.900 s repaint), and the L4 has only a single-precision FPU, so the double-precision astronomy would be slow soft-float in an interrupt. So the heavy maths runs in the main loop — astro_update(), gated on the active mode in exactly the way measure_vbat() is gated for MODE_VBAT — and the small result struct is swapped into a cache under a brief __disable_irq() mask. The sendDate() cases only format cached scalars: no trig, no sun_times, nothing heavy in the ISR. Position is snapshotted once per update so az/el, grid and lat/lon are always mutually consistent.

Everything is parameterised by NUM_DISPLAY_MODES (the enum append grows config.modes_enabled[] automatically), so the modes join the button cycle and config parsing with no other changes.

The maths — Core/Src/astro.c (+ astro.h)

A self-contained C99 + <math.h> unit with no firmware dependencies:

• sun_az_el() — apparent solar azimuth/elevation (NOAA/Meeus low-precision).
• sun_times() — sunrise / sunset / solar noon (+ optional civil/nautical/golden), decimal UTC hours.
• moon_phase() / moon_illuminated_fraction() / moon_phase_index().
• equation_of_time().
• maidenhead() — 6-char locator.

Because it has no firmware ties, it unit-tests on a host. test/test_astro.c checks it against 45 reference vectors (four locations spanning hemispheres/seasons incl. a polar-winter case, plus fixed Maidenhead known-answers like Trafalgar Sq → IO91wm, ARRL HQ → FN31pr):

$ cd mk4-time/test
$ cc -std=c99 -O2 -Wall -Wextra -I ../Core/Inc ../Core/Src/astro.c test_astro.c -lm -o test_astro && ./test_astro
...
45/45 passed, 0 failed

Agreement with the reference is ~1e-5 (well under the whole-degree / whole-minute the display shows). astro.c is added under Core/Src, which the .cproject compiles automatically; test/ is not a source path, so the test never enters the firmware build.

Accuracy & legibility (honest notes)

• Low-precision algorithms: sun position good to a fraction of a degree, event times to a minute or two — appropriate for a 7-seg read-out.
• 7-seg letters are approximate (S→5, I→1, O→0, Z→2), the same as the existing GPS/bat/ttff/rtc labels. Maidenhead's lowercase subsquare (a–x) renders best-effort and is in-bounds for lut_7seg.
• Moon phase is shown as the standard 8-step index because the names don't spell on 7-seg; the legend (0 new … 4 full … 7 waning crescent) is documented in config.txt.

Config

config.txt gains the five mode keys (disabled by default) with comments and the moon-index legend, plus astro_page_ms — the dwell (ms) each sub-screen of the paged modes (SUN, LATLON) shows before flipping (default 5500, floored at 250):

MODE_SUN = disabled
MODE_SUN_AZEL = disabled
MODE_MOON = disabled
MODE_GRID = disabled
MODE_LATLON = disabled
astro_page_ms = 5500
#fake_latitude  = 51.48
#fake_longitude = -0.01

Build & test

Build mk4-time in Release (Debug links a non-bootloader script, so a Debug fwt.bin won't flash via the bootloader) — no project-file changes are needed (astro.c auto-compiles). All five modes — and the astro_page_ms sub-screen paging, its cadence tuned on the unit itself — have been built and confirmed working on real hardware (from a single location); cross-location and edge-case behaviour — hemispheres, the antimeridian, polar day/night — is covered by the 45 native reference vectors above. It stays a draft for your review, not for want of testing.

Review & verification

Before this PR the change was put through an adversarial review across three independent lenses, each trying to break it rather than confirm it:

• Firmware integration / ARM build — enum + cycle wiring, config.modes_enabled[NUM_DISPLAY_MODES] sizing, buffer bounds, switch-case correctness, and that the new modes take the normal COUNT_NORMAL live-clock path.
• Math & port fidelity — re-ran the native test (45/45) and checked edge cases: poles / polar day-night, the antimeridian, azimuth wrap, moon-phase index wrap, maidenhead clamping, fmod sign.
• 7-seg display strings — every mode's output width across the full value range (≤10 chars, no overflow) and glyph legibility.

Verdict was ship-with-fixes; the findings were addressed in this branch:

• ISR snapshot bug (fixed) — MODE_LATLON formatted the live latitude/longitude inside sendDate() (which runs in the SysTick ISR) instead of the IRQ-mask-swapped astro cache, so the shown coordinates could disagree with the cached have_pos/grid in the same frame. Lat/lon are now snapshotted into the cache with everything else.
• Negative zero (fixed) — on the prime meridian / equator %.2f printed LON-0.00; switched to integer-hundredths so it reads LON 0.00.
• Glyph set (confirmed, no change) — the date board's lut_7seg renders the full printable-ASCII set, so the labels and lowercase Maidenhead subsquares display fine; maidenhead() is left faithful.
• ISR safety (confirmed) — the double-precision trig runs once per second in the main loop (mirroring measure_vbat); only cached scalars are formatted in the ISR, so there's no soft-float in the interrupt.

Possible follow-ups

• Render sun-event times on the big time row (needs the SHOW_OFFSET-style segment-buffer path) for nicer HH:MM:SS.
• Civil/nautical twilight and moonrise/moonset (the maths is already present in astro.c).

Files

• mk4-time/Core/Src/astro.c, mk4-time/Core/Inc/astro.h — new, self-contained maths.
• mk4-time/test/test_astro.c — new, native test (not built into firmware).
• mk4-time/Core/Src/main.c, Core/Inc/main.h — enum, cache, astro_update(), 5 sendDate cases, 5 config keys, main-loop hook.
• qspi/config.txt — documented keys + legend.

🤖 Generated with Claude Code