Practical Astrodynamics:
Open-Source Tools for Constellation Analysis and Design

Duncan Eddy

2026 SPNT Symposium · November 12, 2026

Stanford University

GPS satellites on --, from the IGS ultra-rapid orbit --

Which GPS satellites are above this building right now?

Which GPS satellites are above this building right now?

Basic

  • Keplerian two-body orbits
  • Earth spun about its axis
  • Spherical Earth, geocentric latitude

Better

  • TLEs with SGP4
  • TEME to Earth-fixed, with Earth orientation parameters
  • Geodetic station position

Full

  • IGS SP3 orbit products, choosing the best one
  • Earth-fixed positions, time-tagged in GPS time
  • Geodetic station on the reference ellipsoid

IGS ultra-rapid orbit -- · GP elements from -- · ultra-rapid against rapid orbit on -- · scripts/three_levels.py

Which GPS satellites are above this building right now?

-- GPS satellites above --° elevation at Stanford, -- Pacific time on --

IGS ultra-rapid orbit -- at -- · snippets/skyplot.py · full analysis in scripts/skyplot.py

Improving Navigation by Removing Errors

Measurement error

Clocks, ionosphere, troposphere, multipath, receiver noise

System design changes can also reduce measurement error

Dynamics model error

Uncertainty in location from unmodeled forces, and from uncertainty and errors in capturing the true physics

Dynamics model error

Uncertainty in location from unmodeled forces, and from uncertainty and errors in capturing the true physics

  • Earth's gravity field beyond a point mass
  • Atmospheric drag
  • Solar radiation pressure
  • Sun and Moon gravity
  • Solid Earth tides and relativity
  • The time systems and frames all of these are expressed in

Much of the dynamics can be modeled, but it is hard to get right and hard to make fast.

Astrodynamics tools tend to trade these properties off against each other

Capable

Fast

Easy to use

Open

✓ yes · ~ somewhat · ✗ no

General tendencies, not a ranking of any one tool

“It has always been too much astrodynamics for software engineers, and too much software engineering for astrodynamicists.”

Juan Luis Cano Rodríguez, “Juan Luis steps down as maintainer of the project”, poliastro blog, December 21, 2022

https://www.poliastro.space/blog/2022/12/21/juan-luis-steps-down/

brahe

brahe is an open-source astrodynamics project that aims to solve this problem: to empower engineers and researchers to use the highest‑fidelity models, with the speed of simpler ones, without adding complexity.

Fast

-- core, -- bindings

Cross-platform

Linux, macOS and Windows

Free

-- licensed

pip install brahe

The modules in brahe

I have rewritten this project five times since 2014

Design choices in brahe

SI units everywhere

Meters, seconds and kilograms in and out of every function. Angles state degrees or radians.

Composable pieces

Propagators, force models, access constraints and event detectors are separate objects you combine.

Make users explicitly invoke magic

You state each model choice precisely. Once invoked, brahe handles the complexity of using it.

Performance optimization abstracted

Compiled Rust, threading and whole-array calls sit behind ordinary Python functions.

Every documentation example runs

Each example in the documentation is compiled and run on every commit, so the docs match the code.

The documentation has worked examples and a full API reference

brahe documentation example page: Visualizing GPS Satellite Orbits

Examples: worked problems, start to finish

brahe Python API reference page: Epoch class

API reference: every class and function, with examples

docs.brahe.space · brahe 1.8.1 documentation

Time

Supports

Time systems

  • UTC, TAI, GPS, TT, UT1
  • TDB, TCG, TCB
  • BeiDou (BDT) and Galileo (GST) time

Epoch

  • Nanosecond resolution, compensated sums
  • Calendar, JD, MJD, GPS week and seconds, ISO 8601
  • TimeRange grids of epochs

Earth orientation and leap seconds

  • Leap seconds from SOFA, compiled in
  • IERS EOP 20 C04 and finals2000A files
  • UT1−UTC, polar motion, dX and dY, LOD

The same instant in four time scales

snippets/mod_time.py · output and run time (imports excluded) from scripts/snippet_outputs.py · brahe 1.8.1 documentation, brahe.space

Frames and coordinates

Supports

Earth frames

  • GCRF, ITRF, EME2000, MOD, TOD, TEME

Other bodies and interplanetary frames

  • Moon: LCI, LFPA, LFME; Mars: MCI, MCMF
  • Barycentric EMBI, SSBI; synodic EMR, SER, GSE
  • IAU body-fixed frames by NAIF ID
  • Any to any: state_frame_to_frame

Earth rotation models

  • IAU 2006/2000A, CIO based, from SOFA
  • IAU 2000B option; GAST and GMST 1982 chains

Coordinates

  • Keplerian, equinoctial, geodetic (WGS84), geocentric
  • ENZ, SEZ, azimuth and elevation, RA and Dec
  • RTN, LVLH and relative orbital elements

100,000 states between frames in one call, and by frame name

snippets/mod_frames.py · output and run time (imports excluded) from scripts/snippet_outputs.py · brahe 1.8.1 documentation, brahe.space

Orbits

Supports

Elements

  • Keplerian and equinoctial elements, anomalies
  • Period, apsides, sun-synchronous inclination
  • Osculating to mean: Brouwer-Lyddane (first-order J2) or numerical averaging

Element sets

  • TLE and 3LE parsing; TLE generation
  • Alpha-5 catalog numbers
  • GP records (OMM) to SGP4 or numerical propagators

Constellations

  • Walker Delta and Walker Star generators
  • A Keplerian, SGP4 or numerical propagator per satellite

A GPS-like Walker constellation from one builder call

snippets/mod_orbits.py · output and run time (imports excluded) from scripts/snippet_outputs.py · brahe 1.8.1 documentation, brahe.space

Propagation

Supports

Propagators

  • Keplerian, SGP4, numerical orbit, general ODE
  • Many propagators in parallel in one call
  • Events: eclipse, nodes, altitude, any function

Force models

  • Gravity: EGM2008 to 120, GGM05S to 180, JGM3, ICGEM
  • Drag: NRLMSISE-00, Harris-Priester, exponential
  • Cannonball SRP, cylindrical or conical eclipse
  • Sun, Moon, planets (DE440); relativity; tides

Integrators and variational equations

  • RK4, RKF45, DP54, RKF78, RKN1210
  • State transition matrix, sensitivity, covariance

One day at GPS altitude with 80×80 gravity, Sun, Moon, relativity and the STM

snippets/mod_propagation.py · output and run time (imports excluded) from scripts/snippet_outputs.py · brahe 1.8.1 documentation, brahe.space

Access

Supports

Locations and satellites

  • Point and polygon ground locations, GeoJSON
  • Any propagator or trajectory
  • Many locations and satellites in parallel

Constraints

  • Elevation, elevation mask, off-nadir, azimuth, range
  • Local time, look direction, ascending or descending
  • All, Any, Not, and custom Python constraints

Window properties

  • Open, close, duration, azimuth, elevation
  • Doppler, range and range rate, or your own

Every KSAT ground station against one satellite for a day

snippets/mod_access.py · output and run time (imports excluded) from scripts/snippet_outputs.py · brahe 1.8.1 documentation, brahe.space

Data clients

Supports

GNSS orbit products

  • IGS products from CDDIS, Garner (SIO) and BKG
  • SP3 versions a to d read, version d write

Orbit catalogs

  • CelesTrak GP and SATCAT
  • Space-Track GP, GP history, SATCAT, CDM
  • Starlink public ephemerides

Reference data

  • NAIF SPICE kernels, DE440 and others
  • Ground station networks: KSAT, AWS, NASA NEN and more
  • IERS EOP, CSSI space weather, ICGEM gravity

Message formats

  • CCSDS OEM, OMM, OPM, CDM, APM, AEM

List and read an IGS rapid orbit from the BKG mirror

snippets/mod_data.py · output and run time (imports excluded, product already in the local cache) from scripts/snippet_outputs.py · brahe 1.8.1 documentation, brahe.space

Estimation

Supports

Estimators

  • Extended Kalman filter, Joseph-form update
  • Unscented Kalman filter
  • Batch least squares with consider parameters

Measurement models

  • Inertial and ECEF position, velocity and state
  • Azimuth, elevation and range from a ground sensor
  • Custom models in Python

Dynamics and noise

  • Any force model, with the state transition matrix
  • Process noise from a user Q matrix, scaled with time

An extended Kalman filter fed thirty position observations

snippets/mod_estimation.py · output and run time (imports excluded) from scripts/snippet_outputs.py · brahe 1.8.1 documentation, brahe.space

Download public GPS orbit elements and plot ground tracks

Ground tracks of -- of the -- GPS satellites in the public GP catalog, one per orbital plane, starting at the talk time

Public GP element sets, propagated with SGP4 · snippets/ground_tracks.py · full analysis in scripts/quick_hits.py

Propagate an orbit with two-body and full force-model dynamics

--: the two predictions are -- km apart after -- days, and drag alone moves it -- km

brahe default force model · snippets/force_model.py · full analysis in scripts/quick_hits.py

How far is the public TLE from the precise orbit for the same GPS satellite?

RMS difference over -- hours
--km
Largest difference
--km

-- (NORAD --): GP elements from -- (epoch --) with SGP4, against the IGS ultra-rapid orbit -- · snippets/tle_vs_sp3.py · full analysis in scripts/tle_vs_sp3.py

Move the receiver from Stanford to the ISS by changing one argument

The -- at the talk time, about -- km up, with lines of sight to the -- GPS satellites it sees above a --° mask

ISS GP elements from -- with SGP4 · IGS ultra-rapid orbit -- · drawn to scale · snippets/leo_receiver.py · full analysis in scripts/leo_receiver.py

GPS satellites in view of the ground station and the ISS over the next -- minutes

With the same --° mask and a zenith antenna, the ISS sees slightly fewer GPS satellites than the ground station: -- against -- on average.

ISS GP elements from -- with SGP4, station on the Stanford campus, one sample every -- s · scripts/leo_receiver.py

Can you use GPS from above the GPS constellation?

Distance from Earth's center
--Earth radii
GPS satellites in the main lobe
--
In the main or side lobes
--

Geometric visibility, no link budget

-- over one orbit, --, to scale with the Earth and the GPS orbit radius

-- GP elements from -- with SGP4; GPS from IGS rapid orbits; transmit pattern after Donaldson et al. (NAVIGATION, 2020; ION GNSS+ 2018) · scripts/mms_visibility.py

GPS satellites usable by -- against distance from Earth's center

Share of orbit with -- or more in the main lobe
--%
Share of orbit with -- or more in main or side lobes
--%

Geometric visibility, no link budget

-- over one orbit, --, one sample every -- minutes · scripts/mms_visibility.py

Generate a Walker constellation and propagate every satellite in parallel

-- satellites in -- planes at -- km and --°, propagated for one orbit (-- minutes) with --

Serial loop
--s
Parallel
--s
Threads
--
Speedup
--×

Integrator -- · wall-clock times on the build machine · snippets/walker.py · full analysis in scripts/walker_pnt.py

Mean satellites in view by latitude, one Walker shell against GPS

The --° shell has more satellites in view than GPS between about --° and --° latitude and fewer elsewhere: -- against -- at the equator and -- against -- at --°.

--

Time per gravity evaluation against degree and order, serial and parallel

Degree --, serial
--µs
Degree --, parallel
--µs
Degree --, serial
--µs
Degree --, parallel
--µs

From degree -- up, brahe splits the per-order sums across threads automatically.

Measured through brahe's numerical propagator with a gravity-only force model (--), -- to degree --, -- kernel, parallel mode forced off and on. Each value includes about -- µs of integrator and frame-rotation overhead. Lines are medians of -- passes of -- runs per degree; bands span the middle half.

--, brahe thread pool of -- · scripts/gravity_crossover.py

Median run time of common tasks in other libraries, relative to brahe

brahe from Python takes -- times its Rust time on the -- tasks above -- µs and -- times on the -- microsecond-scale tasks. Another library is faster than brahe (Rust) on -- of the -- tasks in the run, listed in the backup.

brahe comparative benchmark suite, run -- · -- · -- · median of -- iterations · -- · -- (open circle) starts one process per call, about -- s, and is shown where its time is at least -- times that · scripts/benchmarks.py

The modules in brahe, after these examples

Organizations that use brahe

Planned work in brahe, and how you can get involved

On the roadmap

  • GNSS orbit determination support, reading RINEX observation and navigation files
  • Estimation with a choice of process noise models
  • GPU acceleration: Astrojax, an experimental GPU-accelerated framework in JAX. SGP4 at -- billion propagations per second on -- GPUs, about -- times the Rust sgp4 crate on -- CPU threads

How you can get involved

  1. Share data: sources of orbital observations that can be shared for development and testing
  2. Request features that your work needs
  3. Contribute code, examples or documentation

SGP4 rates: one satellite, one year at 1 s steps, float64, two NVIDIA H100 NVL GPUs and an AMD EPYC 9554, from github.com/ATTron/astroz/pull/80 · github.com/duncaneddy/astrojax

Every figure in this talk comes from a short script that uses brahe

Duncan Eddy · [email protected]