Gps Signal Quality

# Single log
python3 scripts/gps_signal_quality.py <log-a.ulg> --output-dir <logs/log-name/>

# Compare two logs
python3 scripts/gps_signal_quality.py <log-a.ulg> <log-b.ulg> \
    --output-dir <logs/compare-name/> \
    --label-a "before-fix" --label-b "after-fix"

After running

1. Open the PDF for the user: xdg-open <output-dir>/gps_signal_quality.pdf (or gps_signal_quality_compare.pdf in compare mode).
2. Read the pages you need with the Read tool:
   • page 1: cover + summary table + how-to-read
   • page 2: satellites_used and fix_type timeseries
   • page 3: receiver eph/epv/HDOP/VDOP
   • page 4: s_variance_m_s, c_variance_rad (velocity/course accuracy)
   • page 5: noise_per_ms, jamming_indicator, AGC (environment)
   • page 6: estimator_aid_src_gnss_* innovations + rejections
   • page 7: position_drift_rate_* from estimator_gps_status (the money plot)
   • page 8: post-EKF eph/epv/evh/evv
   • page 9: d(lat,lon,alt)/dt — raw position jitter
   • page 10: pass/fail checks (single) or Δ table + verdict (compare)
3. Interpret the findings. Pull from the summary text file gps_signal_quality.txt as well — it has the same stats in plain text and is easy to quote into a PR comment.

What to look for

• Satellite count jump on the same sky = receiver tracking more signals (e.g. L5 / E5a / B2a enabled). The X20 / ZED-F9P-style driver bugs typically show up as a ~30–50% sat-count reduction when L5 is silenced.
• eph / epv drop = multi-band dual-frequency correction working — removes ionospheric delay. Expect 2–3× smaller eph on a good fix vs. L1-only.
• TTFF improvement = faster time to first 3D fix. Stored as ttff_s in the summary.
• Fix-type upgrade = 3→5 (3D → RTK-float) or 5→6 (→ RTK-fixed) once RTCM corrections are wired up.
• Stationary drift rate (page 7) is the cleanest pass/fail signal: with the aircraft sitting still, how fast does the EKF's position estimate walk? Compare medians, not peaks.
• Noise/AGC parity (page 5) is a sanity check — if these differ a lot between logs, the RF environments weren't comparable and other deltas should be interpreted with that caveat.

Compare-mode verdict wording

The script auto-generates a one-sentence verdict on page 10 / in the text file. When relaying it to the user, add context:

• Mention the satellite-count delta and the drift-rate delta together — one without the other is weaker evidence.
• If noise_per_ms differs > 20% between logs, say so: you can't cleanly attribute deltas to the driver change.
• If either log has jamming_indicator p95 > 100, flag possible RF interference.

Missing topics

Some topics may not be in the log:

• estimator_aid_src_gnss_* (pos/vel/hgt) — absent on older logs. Page 6 will show a "topic missing" banner.
• estimator_gps_status — absent on older logs. Page 7 (the money plot) will be empty.
• satellite_info — per-satellite CN0 / band breakdown. Not consumed by this script but worth mentioning to the user: if they're trying to prove L5-band reception directly, enabling this topic (SDLOG_PROFILE includes it, or custom profile) gives the strongest evidence. ARK_FPV excludes it by default.

Dual-GPS boards

If the board has two GPS receivers, both sensor_gps multi_ids will be populated. The script currently auto-picks the one with more samples and logs which it chose (as a NOTE: line). If the user cares about the other receiver, say so and we can add an --gps-instance flag.

PlotJuggler naming note

When a log has multiple sensor_gps subscriptions (common on boards that subscribe multi_id 0 and 1 regardless of which receivers are wired up), PlotJuggler shows the topic as sensor_gps.00 / sensor_gps.01. The script handles this transparently — you don't need to rename anything.

After interpreting

• Quote the verdict line and the 2–3 most impactful deltas into your response.
• If this is validating a driver PR, surface the satellite-count and drift-rate deltas together as evidence.
• Create or update logs/<log-name>/README.md per the catalogue convention in CLAUDE.md; tag the README with skills_used: - gps-signal-quality and a dated analysis-history entry.