L1C (CNAV-2) TOW is 7200 s short in the last frame of each ITOW interval — L1CD loses TOW every 2 hours

[tomojitakasu]

Summary

In decode_CNV2() (src/sdr_nav.c) the CNAV-2 TOW is computed as

update_tow(ch, getbitu(data, 22, 8) * 7200.0 + (toi - 1) * 18.0 + TOFF_L1CD);

which assumes TOI counts 1...400. The TOI actually carried in subframe 1 is
(frame index + 1) mod 400, so the last frame of every ITOW interval carries
TOI = 0, not 400. (toi - 1) * 18.0 then evaluates to -18 and the resulting
TOW is one ITOW interval (7200 s) short. Every L1C channel therefore fails the
TOW consistency check and loses its TOW every 2 hours.

Observed

TOW MISMATCH <propagated> -> <decoded> at each 2-hour boundary
(pocket_web continuous log, 2026-08-16, QZSS):

pocket_20260816_0200.log  J03/J04/J07/J08 L1CD  TOW MISMATCH  7200.520 ->     0.520
pocket_20260816_0400.log  J02/J03/J04/J07/J08  14400.520 ->  7200.520
pocket_20260816_0600.log  J02/J03/J04/J07/J08  21600.520 -> 14400.520

The propagated value on the left is the correct one. ch->tow is invalidated
and all L1CD observations disappear until the TOW is re-latched and confirmed -
about 37 s (two CNAV-2 frames) each time:

155477.540  J04 1S present    <- last epoch before the ITOW boundary
155478.540  J04 1S gone       <- TOW MISMATCH
155514.540  J04 1S present    <- back after ~37 s

L1CP (1L) is unaffected because the pilot uses the tow_v == 2 path.

Evidence for the TOI convention

Decoding the raw $NAV frames of J04 L1CD around the 2026-08-16 week boundary
(WN = bits 9-21, ITOW = bits 22-29, TOI = bits 0-8):

frame start (GPST)	WN	ITOW	TOI
604782 of week 2431 (last frame of the week)	2431	83	0
0 of week 2432	2432	0	1
18 of week 2432	2432	0	2

So WN / ITOW describe the current frame, and TOI is the index of the next
frame within its interval, wrapping to 0 at the interval end.

Suggested fix

// TOI = 0 means the next frame is in the next ITOW interval
double tow = getbitu(data, 22, 8) * 7200.0 +
    (toi > 0 ? toi - 1 : 399) * 18.0 + TOFF_L1CD;
update_tow(ch, tow);

At the end of the week this yields 604800.520, which has to be wrapped to
0.520 with the week number incremented.

Note that the TOI of NavIC L1-SPS in decode_IRNV1() uses a different, 0-based
convention and is correct as written (ITOW * 7200 + toi * 18 + TOFF_I1SD, no
mismatch at 2-hour boundaries in the same logs). The two should not be unified.

I have this implemented and tested locally and can open a PR if useful.

Environment

PocketSDR library 0.20 (SDR_LIB_VER), master at 03787da, Windows 11 /
MSYS2 UCRT64 gcc 15.2.0, 8 ch front end, QZSS L1CD/L1CP.