Satellite count drops for ~28 s at every GPS week rollover (ch->week is not advanced with ch->tow)

[tomojitakasu]

Summary

Running pocket_trk / pocket_web across a GPS week boundary (Sunday 00:00:00
GPST), the satellite count in the PVT solution drops from ~33 to ~5 and takes
about 28 s to recover.

update_tow() in src/sdr_ch.c wraps ch->tow at the end of the week but does
not advance ch->week, which is only updated when a subframe carrying WN is
decoded. gen_prng() in src/sdr_pvt.c then adds a spurious one week to every
pseudorange, so tau becomes ~604800 s and all observations are rejected by the
tau < -0.02 || tau > 0.167 check.

Observed

$GNGGA satellite count (UTC = GPST - 18 s), identical on 2026-08-02, 08-09 and
08-16, so fully deterministic:

UTC	23:59:42	23:59:43	23:59:49	23:59:57	00:00:10
GPST	00:00:00	00:00:01	00:00:07	00:00:15	00:00:28
NSat	33	5	15	19	32

$CH records show the TOW wrapping while the week stays behind:

148277.540 G07 L1CA tow=604799927 week=2431   <- GPST 00:00:00
148278.540 G07 L1CA tow=      927 week=2431   <- TOW wrapped, week did not
$LOG,148278.540,G07,L1CA,7,PSEUDORANGE OUT OF RANGE (604800.073022)
148284.540 G07 L1CA tow=   6927   week=2432   <- recovers when SF1 is decoded
148305.540 E03 E1B  tow=  27924   week=2432   <- Galileo recovers at +28 s

The staggered recovery matches the WN repetition period of each signal:
GPS/QZSS LNAV SF1 (+7 s), BDS D2 (+15 s), Galileo I/NAV word 5 (+28 s).
GLONASS is unaffected because ch->week stays 0 and the tow_v == 2 path in
gen_prng() does not use the week - the 5 surviving satellites are exactly the
GLONASS ones.

Suggested fix

static void update_tow(sdr_ch_t *ch, double sec)
{
    if (ch->tow < 0) return;
    ch->tow += (int)floor(sec / 1e-3 + 0.5);

    // the week has to follow the TOW rollover until the next week is decoded
    if (ch->tow >= 86400 * 7 * 1000) {
        ch->tow -= 86400 * 7 * 1000;
        if (ch->week > 0) ch->week++;
    } else if (ch->tow < 0) {
        ch->tow += 86400 * 7 * 1000;
        if (ch->week > 0) ch->week--;
    }
}

BeiDou needs one more change: SOW + GPST_BDT (14 s) exceeds 604800 for the
14 s after the GPS week end (the BDT week ends later), raising
TOW MISMATCH 0.016 -> 604800.016 and forcing a TOW re-latch. The week has to
be advanced together with the TOW there, since GPST week = BDT week + 1356
alone leaves BeiDou one week behind. NavIC L1-SPS has the same overflow.

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, GPS/GLONASS/Galileo/BeiDou/QZSS.