CubeSat ADCS simulation (side project)

A small attitude determination and control simulation for a CubeSat, in MATLAB and Simulink. I built it up step by step to learn the basics of ADCS. It is a personal learning project, not flight software. NOTES.md is the log of what went wrong and what I changed along the way.

What is ADCS (short version)

A satellite has to point somewhere (a camera at the Earth, an antenna at a ground station, solar panels at the sun). ADCS is the part that does this:

• Determination: work out which way the satellite is pointing, from sensors (sun sensor, magnetometer, gyro).
• Control: turn it to where it should point, here with reaction wheels (spinning one wheel one way turns the satellite the other way).

What it simulates

• A 3U-size CubeSat in a 500 km circular orbit, going through eclipse
• Three reaction wheels with torque and momentum limits
• Sun sensor, magnetometer and gyro (with noise and a drifting bias)
• TRIAD to get a first attitude, then a MEKF that estimates attitude and gyro bias
• A PD controller that points the satellite using only the estimated attitude
• A requirements check (PASS/FAIL) and a 50-run Monte Carlo
• A Simulink version of the control loop, compared with the MATLAB loop

Results (one orbit, run_adcs, MATLAB R2024b)

Requirement	Result	Limit
REQ-01 Pointing error (sunlit)	0.47 deg	< 2 deg	PASS
REQ-02 Body rate (sunlit)	0.0013 rad/s	< 0.01 rad/s	PASS
REQ-03 Peak wheel momentum	36 %	< 80 %	PASS
REQ-04 Sun sensor error (RMS)	0.41 deg	< 0.5 deg	PASS
REQ-05 Magnetometer error (RMS)	0.81 deg	< 1 deg	PASS
REQ-06 Attitude knowledge (sunlit)	0.35 deg	< 1 deg	PASS
REQ-07 Bias estimate error (sunlit)	2.7e-4 rad/s	< 5e-4 rad/s	PASS

In eclipse there is no sun sensor and the attitude drifts by several degrees until the sun comes back. That is why some requirements are only checked in sunlight (see docs/requirements.md).

Repository

matlab/              the simulation (start from run_adcs.m)
simulink/            Simulink version of the control loop
tests/               quick checks for the quaternion functions and TRIAD
docs/requirements.md what the simulation should achieve
docs/assumptions.md  what is simplified
docs/code_guide.md   how the code works, in which order to read it, glossary
plots/               saved figures
adcs_simulation.py   first Python version (control only)
ekf_estimation.py    first Python version (1-axis EKF)
NOTES.md             bugs, surprises, decisions

How to run

In MATLAB (the scripts use relative paths, so cd into the folder first):

cd matlab
run_adcs        % one orbit (~95 min simulated, ~40 s to run), prints the requirements check
monte_carlo     % 50 short runs with random start conditions, takes a few minutes

Simulink (needs Simulink installed):

cd simulink
build_adcs_model   % creates adcs_model.slx (already in the repo, only needed after changes)
run_simulink       % runs 300 s and compares with the MATLAB loop

Tests:

cd tests
test_quaternions

The first Python versions need numpy and matplotlib: python adcs_simulation.py and python ekf_estimation.py.

Status

• Requirements and PASS/FAIL check
• Reaction wheels
• Sensors
• TRIAD + MEKF with gyro bias
• Monte Carlo
• Circular orbit, field direction, eclipse
• Simulink version of the control loop
• Sensors and MEKF in Simulink
• Monte Carlo with the orbit (only run before the orbit was added)