- Restore true chip semantics for L1: QZSS_L1_CHIP_RATE is 1.023 Mcps and
QZSS_L1_CODE_LENGTH is 1023 chips again. The replica carries
QZSS_L1_SAMPLES_PER_CHIP = 2 samples per chip (L1 C/A chips duplicated,
L1 C/B sinBOC(1,1) subchips), following the Galileo E1 approach. The
generated waveform is unchanged, but Tracking_1J.early_late_space_chips
is expressed in chips again, consistently with all other signals.
- Normalize the DLL discriminator with the sinBOC(1,1) correlation
function when the tracked PRN broadcasts L1 C/B, selected per PRN in
start_tracking(); warn if the configured correlator spacing falls
outside the BOC(1,1) correlation main peak (> 0.33 chips).
- Attribute L1 C/B observables and ephemerides (PRN 203-206) to the PRN
of the satellite's nominal PNT signals (RINEX 4.00, Table 6), so RINEX,
RTCM and rtklib identify the transmitting satellite correctly and
multi-frequency observations of the same satellite are paired instead
of counted twice.
- Do not search QZSS L5 signals for PRNs above 202, which have no L5
PRN code assigned in IS-QZSS-PNT.
- Remove reserved, non-operational PRNs 198 and 202 from the default
QZSS search list; they remain selectable via the QZSS.prns option.
- Fix a potential std::out_of_range in
GNSSFlowgraph::priorize_satellites() when assisted acquisition
prioritizes satellites of a constellation with only a subset of its
signals configured.
The Monitor block is fed from the Observables output, which only
produced data for channels with a decoded time of week. Galileo E6
channels cannot obtain the TOW from HAS pages alone, so E6-only
configurations never produced any Monitor output. When
Monitor.enable_monitor=true, the Observables block now emits every
epoch, filling channels without valid observables with their latest
tracking data (C/N0, Doppler, carrier phase), flagged as invalid so
that the PVT engine and other consumers keep ignoring them.
Fixes#1018
Results are bit-for-bit equivalent to the previous two-correlator
implementation, verified over 3- and 5-tap configurations with both the
normal and high-dynamics resampler paths.
When tracking the pilot component, dll_pll_veml_tracking previously ran
two separate carrier wipe-off passes over the same input samples: one
for the pilot correlator taps and a second, redundant pass through a
dedicated Cpu_Multicorrelator_Real_Codes instance to compute only the
data prompt. The carrier rotation dominates the per-sample cost, so it
was effectively being computed twice over identical data.
Extend Cpu_Multicorrelator_Real_Codes with optional data-prompt support
(init with has_data_prompt, set_data_code_and_prompt_tap, and
Carrier_wipeoff_multicorrelator_resampler_with_data_prompt) that
resamples one extra local-code buffer for the data prompt and computes
all pilot taps plus the data prompt under a single carrier rotator via
one VOLK_GNSSSDR xN kernel call. Scratch buffers are allocated at init,
so there are no per-call heap allocations in the hot path. Both the
normal and high-dynamics resampler/rotator paths are preserved, and the
second correlator instance is no longer used for CPU VEML tracking.
Results are bit-for-bit equivalent to the previous two-pass
implementation.
This cuts tracking-loop time by ~33% (about 1.5x throughput) for
pilot-tracked signals (GPS L5, Galileo E1/E5a/E5b/E6, QZSS L5). Data-only tracking
(track_pilot = false) is unchanged.
Store PCPS Doppler wipeoffs and magnitude grids in flat, stride-aligned buffers instead of nested vectors. This reduces per-row indirection, improves cache locality, and makes the Doppler-bin data layout suitable for future batch FFT planning.
Move QuickSync acquisition temporary buffers out of general_work() and
initialize them once in the block constructor. Clear the FFT input buffer
directly with std::fill_n() for each Doppler bin instead of recreating and
copying d_signal_folded.
... when Acquisition_5X.coherent_integration_time_ms is greater than 1.
In this case vectors codeQ_, codeI, codeQ are constructed with
code_length size, but written with vector_length which results in
arbitrary memory overwriting and crash.
Construct these arrays with vector_length sizeto make sure that there
will be enough space for all primary code replicas and zero padding.
Signed-off-by: Vladslav P <vladisslav2011@gmail.com>
Carles Fernandez