RF Classifier and Workbench

01Overview

The GEOx RF Classifier turns raw Spire RF geolocation collections into a labeled, triaged RF picture. Drop one or more GeoJSON collections into the browser and the application classifies every fix, assigns a confidence tier, plots the results on an interactive map, and builds a sortable catalog you can filter, replay, interrogate, and export.

02Getting started

1. Open the application in a modern browser (Chrome or Edge recommended).
2. Drop one or more Spire RF geolocation GeoJSON files onto the Catalog drop zone in the left panel, or click the zone to browse.
3. The application classifies the fixes, populates the map and the catalog table, and displays summary statistics across the top of the map.
4. Use the left-panel filters to narrow the picture, click any table row or map marker to inspect a fix, and export when ready.

03Loading collections

The Catalog drop zone accepts multiple Spire RF collections in a single session. Each loaded file appears in the file list with its band and fix count. Fixes from all loaded collections merge into one catalog, so you can build a multi-band, multi-pass picture of an area of interest.

• Supported input: Spire RF geolocation GeoJSON. The application reads the source ground location estimates, their paired error estimate features, and any LineString collection tracks.
• Per-fix fields consumed: carrier and measured frequency, filter bandwidth, post-detection SNR and SINR, 95% error ellipse geometry (CEP95, semiaxes, azimuth), batch and solver iteration, and collection identifiers.
• Satellite attribution: the spacecraft pair behind each geolocation is parsed from the collection IDs and shown in the Satellite column.
• Clear all removes every loaded collection and resets the workspace.

04Context overlays

The Context overlays drop zone adds reference geometry to the map without affecting the catalog. Use it to bring in areas of interest, boundaries, infrastructure layers, or prior reporting.

• Accepted formats: zipped shapefile (.zip), loose .shp, GeoJSON, and KML.
• Each overlay is drawn in a distinct color and listed with a per-file clear control.
• Overlays render beneath the fix markers so they never obscure the catalog picture.

05Basemaps

Sixteen basemaps are available from the left panel, grouped by use case. Switch at any time; the fix layer and overlays persist across basemap changes.

06Reading the map

Markers

Each fix is a colored circle. Color and size encode signature class: jammers are the largest and warmest markers, clean emitters render green, and low-confidence or spurious fixes render small and gray. Hover any marker for a summary; click to open a popup and highlight the matching catalog row.

Error ellipses

Toggle Error ellipses to draw the 95% error ellipse for each fix, built from the major and minor semiaxes and azimuth in the collection. Ellipses are the honest footprint of a geolocation. A tight ellipse supports targeting-grade language; a long thin ellipse should be treated as a bearing.

Summary statistics

Six stat cards above the map track the filtered set in real time: Catalog set (fixes passing filters), Jammers, Interference, Clean emitters, Tier A, and Bands · collects. Click the Jammers card to fly the map to the strongest jammer site.

Priority banner

When swept jammer activity is present, a priority banner appears above the map identifying the highest-priority track on signature grounds: the single site with the most hits, its mean position, sweep width, and time of collect, with a Locate button that centers the map on it.

07Filtering the catalog

08The catalog table

The Emitter Catalog lists every fix passing the current filters. Click any column header to sort; click a row to select it and highlight the fix on the map. Hover a column header for its definition.

Right-click actions

• Right-click a row for a context menu: Copy record (a formatted plain-text fix report covering signal, position, error, and cluster fields), Copy raw JSON (the underlying feature), or Copy lat, lon.
• Right-click the table header to copy all records in the current filtered set at once.
• Press Esc to dismiss any context menu.

09Time-lapse replay

Toggle Time-lapse to replay the collection as it was solved. Fixes appear in batch, then solver iteration, then index order, with a running clock, so you can watch a site accumulate hits over the pass and separate persistent sites from one-off detections.

• Standard transport controls: play, pause, and a scrubber for seeking.
• A speed control adjusts replay rate.
• Replay honors the current filters, so you can time-lapse a single band or class in isolation.

10Intel Lookup

Intel Lookup generates a short, plain-language intelligence brief about one loaded collection, written for a reader who is not an RF specialist. Select a collection, optionally type a question, and run the lookup.

The brief is generated under strict grounding rules:

• Only facts present in the collection brief are used. Operator names, unit identities, and platform types are never invented.
• Every factual claim cites the data field it came from, inline in brackets.
• Analysis is explicitly labeled ASSESSED: and names the fields the assessment rests on.
• Missing fields are reported as missing. Parametric characteristics (PRI, modulation, pulse width, scan) are not described unless those fields are populated.

11Exporting data

• Export CSV writes the current filtered catalog to geox_rf_emitter_catalog.csv for use in spreadsheets, GIS, or downstream fusion.
• Formatted single-fix reports and raw JSON are available from the table's right-click menu (Section 08).
• The filtered set is what exports. Set your bands, classes, tiers, and thresholds first, then export.

12Classification reference

The application ports the GEOx build pipeline classifier into the browser. It runs in two stages: confidence tiering from solution quality, then signature classification from spectral and spatial behavior.

Confidence tiers

Signature classes

A site is flagged as swept when three or more fixes fall within 3 km of one another, the cluster's median SNR−SINR gap is under 3 dB, and the carrier is swept across more than 150 kHz. Sweeping drives the two highest-interest classes.

13Analyst tips