Skip to content

Satellite Development

Last updated: 2026-04-10

Last updated: 2026-04-11

Focus: small satellites (CubeSat, microsat), LEO missions, on-board software.

Recent Finds

Model-Based ADCS Design-to-Verification for INHA RoSAT 3U CubeSat (MDPI Aerospace, Feb 2026)

Reusable end-to-end ADCS pipeline: algorithm design in Simulink → simulation → automatic C code generation → integration into flight software, bypassing manual C coding entirely. Validated on INHA RoSAT, a 3U CubeSat with rollable solar panels to overcome the chronic power constraint of standard body-mounted cells. The paper's contribution is a verified MBD (model-based design) framework applicable to any team building ADCS from scratch — the verification framework is the deliverable, not just the algorithm. Key practical note: the Simulink-to-C path requires careful fixed-point arithmetic configuration to avoid numerical drift on embedded targets.

LeLaR: First In-Orbit AI Attitude Controller via Deep Reinforcement Learning (arXiv 2512.19576)

Landmark result: a DRL controller trained entirely in simulation was deployed to InnoCube (3U nanosatellite, launched Jan 2025) and successfully commanded reaction wheels in orbit on October 30, 2025 — the first confirmed in-orbit AI attitude control demonstration. Sim2Real transfer worked without any fine-tuning in orbit. The training domain randomized over inertia tensor uncertainty, actuator noise, magnetic disturbances, and thruster faults. Implication: DRL-trained attitude controllers may eventually replace hand-tuned EKF + PID stacks for ADCS in CubeSats, especially for missions with uncertain or time-varying inertia (deployable structures, fuel slosh). Open question: how does it perform during detumble (high angular rates post-deployment), where classical B-dot controllers are most reliable?

CubeSat Standard

Unit (U) = 10×10×10 cm, ~1.33 kg max. Common form factors: 1U, 2U, 3U, 6U, 12U.

1U CubeSat mass budget (~1.3 kg):
- Structure + panels:  0.3 kg
- EPS + batteries:     0.3 kg
- OBC + comms:         0.2 kg
- ADCS:                0.2 kg
- Payload:             0.3 kg

Subsystems

Subsystem Acronym Function
Electrical Power System EPS Solar panels, battery, power regulation/distribution
On-Board Computer OBC Central processor, data handling, software
Communications COMMS RF transceiver, antenna deployment
Attitude Determination & Control ADCS Orientation sensing + control (magnetorquers, reaction wheels)
Thermal Control TCS Passive (coatings) or active (heaters)
Structure STR Mechanical frame, deployment mechanism
Payload PL Mission instrument (camera, spectrometer, SDR...)

EPS Architecture

Solar panels → MPPT → Battery → Power bus (3.3V / 5V / 12V / unregulated) → subsystems
  • MPPT (Maximum Power Point Tracking) — extracts max power from solar cell I-V curve
  • Battery: typically Li-ion or LiFePO4 — must survive LEO thermal cycles (-40°C to +80°C)
  • Power budget: 1U gets ~2–5W average orbital power

ADCS Basics

Determination (where am I pointing?): - Magnetometer → magnetic field vector - Sun sensor → sun direction - Gyroscope → angular rate - Star tracker → precision pointing (expensive)

Control (change my pointing): - Magnetorquers — coils generate torque against Earth's B-field. Simple, low power, slow. - Reaction wheels — momentum exchange. Faster, needs desaturation via magnetorquers. - Thrusters — active orbit control, rare in CubeSats (cold gas, electrospray)

Modes: 1. Detumble — kill post-deployment tumble with B-dot controller 2. Nadir pointing — always face Earth 3. Sun pointing — maximize power during eclipse recovery 4. Target pointing — point payload at target


On-Board Software

Architecture Patterns

Monolithic RTOS (most CubeSats): - FreeRTOS / ChibiOS / RTEMS - Tasks: EPS monitor, ADCS loop, comms handler, payload manager - Simple, deterministic, easy to certify

Linux-based (larger small sats, payload-heavy): - Yocto/Buildroot custom image - Run ML inference (TFLite/NCNN) on payload processor - More complex reliability story (kernel panics, filesystem corruption)

Fault Tolerance Patterns: - Watchdog timers — hardware kicks OBC if software hangs - Safe mode — minimal power draw when anomaly detected - Command & Data Handling (C&DH) — centralized state machine - Triple Modular Redundancy (TMR) — vote among 3 results for critical decisions

Key Open Source Flight Software

Framework Language Notes
NASA cFS (core Flight System) C NASA open-source, POSIX + RTOS, production-used
FPrime C++/Python JPL-developed, component-based, used on Mars Ingenuity
OpenSatKit Ruby/C cFS-based toolkit with GCS
LibCSP C CSP protocol stack, runs on bare metal + Linux
SatOS (EnduroSat) C Commercial, free for non-profit

NASA cFS is the gold standard open-source FSW framework. Steep learning curve but production-proven.

Common OBC Hardware (Open Source / COTS)

Board CPU OS Notes
GomSpace NanoMind ARM FreeRTOS Industry standard
ISIS iOBC ARM9 Linux/FreeRTOS Common in 3U+
Raspberry Pi CM4 (for payload) ARM Cortex-A72 Linux High performance, thermal challenge
STM32-based custom ARM Cortex-M FreeRTOS DIY, good learning platform

Orbit Mechanics (LEO Basics)

Typical LEO: 400–600 km altitude, ~90–95 min period.

Key parameters: - Semi-major axis (a): orbital size - Eccentricity (e): 0 = circular, <0.1 for LEO - Inclination (i): angle to equatorial plane. Sun-synchronous: ~97° - RAAN: Right Ascension of Ascending Node — where orbit crosses equator northbound

Sun-Synchronous Orbit (SSO): - Precesses ~0.9856°/day to match Earth's orbit around Sun - Same local solar time over every pass → consistent lighting for imaging - ~96-100° inclination depending on altitude

Ground Track Coverage: - Single LEO sat: ~10 min pass window, revisit ~14×/day globally - Constellation needed for continuous coverage

Tools: STK (AGI), GMAT (NASA open-source), Orekit (Java), Poliastro (Python)


Licensing & Regulation

  • ITU frequency coordination — must file for frequency allocation before launch. 2–3 year process.
  • FCC (US) or national regulator — need license to operate transmitter.
  • ITAR/EAR — US export controls apply to satellite components and software.
  • Debris mitigation — LEO satellites must deorbit within 5 years (per new FCC rule, IADC 25-year guideline still common).

Open Questions

  • What's the realistic cost breakdown for a 3U CubeSat mission end-to-end (hardware + launch + ops)?
  • How does on-board ML inference (NCNN/TFLite) fare under space radiation — bit flips, latch-up?
  • What's the maturity of inter-satellite links (ISL) for small constellation coordination?
  • Electrospray thrusters — are they reliable enough for CubeSat propulsion in 2026?