Solving the "GPS says 12 meters, I'm standing on the basket" problem.
Disc golf apps like UDisc let you drop a pin for each tee and basket, but a single GPS reading under tree cover is often off by 10-15 meters — signal bounces off trunks and canopy (multipath) before it ever reaches your phone. This app exists so I can fix that at the source: it collects raw GPS samples over time, runs them through a proper statistical pipeline the way surveyors do, and produces a coordinate accurate to sub-meter level, with the error metrics to prove it — which then gets typed back into UDisc's course editor by hand. Two things improve once that happens: players get an accurate "distance to basket" readout on the course (and correct tee-to-basket hole distances), and anyone checking the course map from home sees baskets and tee pads right where they really are instead of scattered, sloppy pins.
It's a solo/personal project built to actually re-survey a real course, not a demo.
- Raw GPS, not fused location. The app talks directly to
LocationManager.GPS_PROVIDERand theGnssStatusAPI instead of Android's fused/network location, so it sees real per-fix accuracy and satellite data instead of a smoothed, cached estimate. - A foreground service keeps sampling even with the screen off. You place the phone on the tee pad or basket, walk away, and it keeps recording at 1 Hz with a wake lock held.
- Every sample is weighted and averaged in 3D, not on the map. Naively averaging latitude/longitude is mathematically wrong (the earth isn't flat) and gets worse at higher latitudes. Instead, each sample is converted to ECEF Cartesian coordinates on the full WGS84 ellipsoid, weighted by
1 / accuracy², averaged in 3D, then converted back — seeGeoUtils.kt. - Outliers get rejected automatically. After an initial average, any sample beyond 3 standard deviations gets dropped and the position is recomputed — see
AveragingEngine.kt. - You get real error metrics, not just a dot. CEP68 / CEP95 (the radius containing 68%/95% of samples), standard deviation, sample count, and mean satellite count are all computed live while you're standing there, so you know when it's good enough to stop.
- Multiple sessions on different days get merged. Satellite geometry repeats on a schedule, so a bad reading today can share a systematic bias with tomorrow's. Re-measuring a point on a different day and merging sessions cancels out that bias — the app tracks sessions per point and lets you merge or discard any of them.
- Dual-frequency (L1+L5/E5a) detection. Modern phones can receive a second GNSS frequency that's far more resistant to multipath under tree cover. The app reads raw carrier frequency data to detect and flag when you're getting that signal.
- Course → Hole → Point hierarchy: create a course, set hole count, measure the tee pad and basket for each hole independently.
- Live measurement screen: real-time sample count, elapsed time, CEP68/CEP95, and a scatter plot showing where your samples actually landed relative to the computed center.
- Multi-session support: re-measure a point any time (different day, different conditions) and the app merges all raw samples into one refined estimate — you can also inspect or delete individual sessions.
- Full satellite diagnostics: a sky-plot dialog (elevation/azimuth), per-satellite signal strength bars, and a sortable table across GPS/GLONASS/Galileo/BeiDou/QZSS/SBAS/IRNSS, tap-accessible from an always-on status bar.
- Course overview: color-coded grid of all holes showing measurement completeness and accuracy quality (CEP95) at a glance, plus computed tee-to-basket distance and elevation change.
- Live distance-to-point: once a tee/basket is measured, walking the course shows a real-time "you are X.Xm away" readout.
- JSON export: a full export with every raw sample (for reprocessing later with different algorithms) and a summary export with just final coordinates and quality metrics, shared via the system share sheet.
- Kotlin, Jetpack Compose (Material 3), Navigation Compose
- Android
LocationManager/GnssStatusraw GNSS APIs (no Google Play Services location dependency) - Foreground
Service+ partial wake lock for background-safe sampling - Gson for local JSON persistence — no backend, no network calls, no analytics
- Custom Canvas-drawn visualizations (sky map, scatter plot) — no charting library
Everything stays on the device. There's no backend, no account, no network permission at all — data lives in the app's local storage as JSON per course, and the only way it leaves the phone is if you tap Export and share the file yourself.
Personal project, used to re-survey a real course hole by hole. Not published to the Play Store — build and sideload it from source (see below) if you want to try it.
git clone https://github.com/Nicofisi/DiscGolfLocator.git
cd DiscGolfLocator
./gradlew assembleDebug
Requires Android SDK with compileSdk 36, targets minSdk 26 (Android 8.0+). Open in Android Studio for the easiest experience — it'll pick up the Gradle config automatically.
No license file yet — all rights reserved by default. Ask before reusing.