Preserve the candidate anchor until its expected confirmation point.
Replace it only after confirmation is missed, preventing payload
correlations from repeatedly displacing a valid preamble.
Compensate the Viterbi traceback delay when timestamping decoded SBAS
messages by carrying per-sample reception timestamps through the sample
and symbol aligners (removes a ~140 ms bias), with a deterministic unit
test covering the FIFO pairing.
Protect decoder state with d_setlock in reset(), set_satellite(),
set_channel() and general_work(), following the convention of the other
telemetry decoders.
Share per-PRN EMS dump files through a per-run dump session: the first
open in a run truncates stale data, later opens append, so channel
reassignments no longer wipe previously decoded messages.
Wait until the continuous decoder has accumulated the requested traceback depth before advancing trellis iterators. Add a regression test for the 30-bit SBAS input block versus its 35-bit traceback depth.
Signed-off-by: miguekf <miguel.gomezlopez@uclm.es>
The symbol alignment term of compute_message_timestamp() used the
opposite sign convention to the sample alignment term. In both
aligners, true means the non-shifted hypothesis won (no correction)
and false means the shifted one won (-1 unit), so the symbol term
must be (symbol_aligned ? 0 : -1).
Two of the unit tests added along with that function were failing as
a result, which is what broke the check step of the CI on all three
platforms.
Also express the correction bound in whole code periods rather than
comparing doubles at an exact boundary: the worst case is 3 code
periods (1 from the sample-pairing ambiguity plus 2 from the
symbol-pairing one), not 2, and the floating-point comparison was
fragile at that boundary.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- gnss_satellite.cc: accept the full supported SBAS PRN range (120-138)
instead of a stale 6-PRN whitelist that rejected valid PRNs such as WAAS
PRN 133. Per-PRN satellite/service names are no longer hardcoded, since
that assignment changes over time and cannot be reliably kept up to date
here (fixes review comments on PRN support and outdated satellite info).
- SBAS_L1.h: cap SBAS_L1_MAX_COHERENT_INTEGRATION_SYMBOLS at 1 (1 ms). The
telemetry decoder performs its own unsynchronized sample/symbol pairing,
so extending tracking's coherent integration to 2 ms is unsafe and
corrupts that pairing.
- sbas_l1_telemetry_decoder_gs.{h,cc}: carry an absolute per-sample/per-bit
timestamp through the sample buffer and Frame_Detector, replacing the
buggy offset that anchored the message timestamp to whichever sample was
current when decoding happened, which could be off by up to a full
message duration. Add
sbas_l1_telemetry_decoder_gs::compute_message_timestamp() as a small,
independently testable helper for the residual sub-bit correction, and a
matching deterministic unit test.
- sbas_l1_telemetry_decoder_gs.{h,cc}: implement a real reset() that clears
all decoder buffers/aligners and closes the dump file, so a channel
reassigned to a different PRN never mixes messages into the previous
satellite's file.
- Remove the orphan sbas_l1_telemetry_decoder.cc adapter: it includes a
header that no longer exists and is not part of any CMake target (the
factory instantiates the GNU Radio block directly), so it silently never
compiled.
- Dump file: overwrite instead of append on each run, remove the hardcoded
GNSS-SDR version string, and document that the .ems output is an
EMS-like raw message dump, not a conformant EMS file.
- conf/File_input/SBAS/gnss-sdr_SBAS_EGNOS_rx.conf: drop the outdated
PRN/satellite table and per-channel satellite comments, correct the
coherent-integration note to 1 ms, and clarify that SBAS is not used as
a ranging source in this configuration.
Signed-off-by: miguekf <miguel.gomezlopez@uclm.es>
Carles Fernandez