A strong frame's last pilot is an up-chirp. The detector and sync can lock
onto it again within a symbol of the frame's end, the time shift showing
as a large frequency offset; nothing follows, so the header fails and a
"preamble found, header fail" was reported. Seen once in SDRangel at
+17 dB (sync 55 chips before the end, offset -22.8 kHz) and reproduced
from the recording at 150 kS/s: two such reports, syncs at -55 and +341
chips from the frame end. The existing check only dropped syncs more than
M/2 before the end.
A failed candidate whose sync starts before one symbol after the end of
the frame just decoded is now a false alarm: not counted, not reported.
Candidates there are still tried, so a real frame starting back to back is
delivered as before.
Decoded frames are identical on all captures; back-to-back frames (gap 0,
SF9 and SF7 coherent, clean to -8 dB) decode 6/6 with old and new. make
test and tools/check_c.py pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
chirpmodem_tx_cw_id smoothed the keying envelope with a direct convolution
(5 ms raised cosine, kr = 0.005 fs taps, for every sample): 6 s for 5.1 s
of CW at 600 kS/s, 16 s at 1 MS/s, slower than real time, which stalled the
SDRangel modulator. The envelope is 0/1 and its runs are at least a dot
long, so each output sample is a sum over at most a few runs of
differences of the window's cumulative sum: O(n). Output is bit-identical
(float32) to before at 125 k, 250 k, 600 k and 1 MS/s, 13 and 20 WPM;
600 kS/s now takes 0.08 s.
The stream detector's metric divided a window's peak bin by its mean
power. On exact zeros (silence in transmit files, SDRangel FileOutput and
loopback paths) that is 0/0 = NaN, and NaN passed "met < 4.0" as not
below the threshold: 583 false "preamble found, header fail" per 5 s of
zeros. An all-zero window now has ratio 1, and the threshold test is
written so that NaN fails. Decoded frames, SNRs and offsets are identical
on all captures (ota.cu8 49, live_rx 42, ota_tx, ota_cw, soak_tx,
sweep_tx, live_tx, modtx); only the spurious failures on digital silence
are gone (ota_tx 181 -> 0).
make test and tools/check_c.py pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Python reference implementation, the C library (libchirpmodem 0.1.0,
format v1), the format description and test vectors, the over-the-air and
soak test scripts and their results, as they were on 2026-11-08 before
this repository was created. Captures (.cu8, .sigmf-data, .sdriq) and build
outputs are not tracked; the SigMF metadata is.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>