Synthetic datasets

The generator draws objects, attitude states and geometries, renders their light curves with the GPU integrator and adds the noise of a real sensor. Every curve carries its true labels and the class found by LightCurve’s attitude classifier.

Command line

npm run dataset -- -n 1000 -o dataset/ --duration 600 --cadence 2 --sensor rasa11-qhy600 --seed 1

Interface

Synthetic dataset section of the Satellite tab: number of curves, length, step and attitude states, then Generate and download.

How a curve is drawn

Element	Draw
Model	A satellite or rocket body of the catalogue
Attitude state	Stabilised; rotation about a fixed axis, period log-uniform from 4 steps to a third of the duration; free tumbling, random inertia and an initial axis at least 25° from every principal axis
Geometry	Starting phase ±(20–110)°, drifting by up to ±30° along the pass; phase-plane tilt; range 0.7 to 1.5 × the typical range of the model
Materials	Albedo factor, ±20 %
Noise	Photon noise of the object and the sky, read noise (SNR budget of the chosen sensor), plus a 0.01 mag floor

The Earth’s shadow is ignored: the curves describe the object, not the night.

Options

Option	Default
-n	100 curves
-o, --out DIR	dataset/
--seed	1
--classes	stable,spin,tumble
--presets	every satellite and rocket body
--duration, --cadence	600 s, 2 s
--sensor, --exposure	the interface’s sensor
--floor	0.01 mag
--resolution	96 px

About 0.2 s per 300-point curve.

Output

id,preset,class,period_s,spin_axis_az_deg,spin_axis_el_deg,inertia_x,inertia_y,inertia_z,albedo_factor,range_km,phase_start_deg,phase_end_deg,points,predicted_class,predicted_confidence,predicted_period_s
bs-00001,geo,stable,,,,,,,1.357,37280.1,63.97,37.58,301,stable,0.950,
bs-00002,falcon9-s2,tumble,45.457,88.5,-59.1,1.000,6.130,2.739,1.226,785.6,-28.39,-53.89,301,spin,0.919,28.746
bs-00003,ariane5-esca,spin,10.625,42.2,24.6,,,,1.146,10048.4,-34.25,-32.12,301,spin,0.970,10.626

id,t_s,mag,mag_err
bs-00003,0.000,7.4803,0.0101
bs-00003,2.000,7.3961,0.0101