01
On-orbit inference
Run filtering, detection, classification, compression, and autonomy close to the sensor.
Satellite InferenceOrbital Computing InfrastructureWhitepaper A program of RFID INC
Compute where
space data begins.
Orbital computing infrastructure for spacecraft, constellations, and autonomous systems, scaling from hosted flight tests to megawatt-class networks.
500-600 kmLEO baseline
1 kWfirst owned node
10 kWsolar at launch
DATA ORIGIN
USEFUL RESULT
FILTER
INFER
PRIORITIZE
SIMULATION 01 Processing moves closer to the sensor
Earth observationSynthetic aperture radarHyperspectral sensingCivil RF analyticsWeather and scienceAutonomous spacecraftEarth observationSynthetic aperture radarHyperspectral sensingCivil RF analyticsWeather and scienceAutonomous spacecraft
01 / THE THESIS
Do not downlink a problem
you can solve in orbit.
Space sensors create more data than constrained links can always move. Satellite Inference turns raw streams into prioritized, verifiable results before transmission to Earth.
01SenseRaw orbital data
HIGH VOLUME
02InferFilter and classify
REDUCED
03DeliverUseful result
02
Downlink reduction
Transmit prioritized results and evidence instead of every raw byte generated on orbit.
03
Power-thermal scheduling
Schedule useful workloads against sunlight, battery state, radiator capacity, and contact windows.
04
Radiation-aware runtime
Detect faults, checkpoint state, recover safely, and create a traceable flight evidence record.
02 / ORBIT LAB
Place the node.
See the physics.
Compare a 550 km reference LEO with geostationary orbit. Geometry, propagation delay, orbital period, and idealized maximum eclipse are calculated in the browser from published physical constants.
ORBIT SELECTIONProgram baseline
MODEL RUNNING LOCALLY
LEO / CALCULATED
Reference LEO
Fast access to space-originated data, lower link loss, and a credible early deorbit path.
Orbital period95.65 min
One-way space leg1.83 ms
Ground-to-node response floor3.67 ms
Idealized maximum eclipse35.61 min
Sunlit fraction at beta 062.8%
Why LEO first
The first market processes data already generated on spacecraft. LEO closes the near-term flight, link, disposal, and customer-learning loop without requiring GEO insertion.
Circular orbit, spherical Earth, beta angle 0, cylindrical shadow, zenith slant range. Real links and eclipses depend on geometry, atmosphere, inclination, season, pointing, and ground location. GEO eclipse is seasonal.
Open equations, constants, and sources
ORBITAL PERIODT = 2pi sqrt(r^3 / mu)
Two-body circular-orbit screen using WGS-84 Earth radius and gravitational parameter.
PROPAGATION FLOORt = rho / c
Vacuum delay only. It excludes compute time, routing, queues, gateway, and protocol overhead.
MAXIMUM ECLIPSEte = T asin(RE / r) / pi
Beta 0 cylindrical-shadow estimate. GEO eclipse occurs seasonally near the equinoxes.
CONSTANTS AND CROSS-CHECKSNOAA WGS-84NASA JPL astrodynamic parametersNIST speed of lightNOAA GEO eclipse operations
03 / GATED SCALE
Flight evidence before
industrial scale.
Every step is a separate mission, budget, and review gate. The number in each Node name always means continuous compute power. Solar generation is stated separately.
MISSION 0SI-HP
Hosted Pathfinder
Hosted compute flight test
SolarProvided by hostBatteryProvided by hostRadiatorHost interface dependentArchitectureHosted payload
00
MISSION 0
Hosted Pathfinder
Compute, recovery, memory behavior, workload packaging, and flight telemetry in the real environment.
0.2-1.0 kW allocationProvided by host solar
01
MISSION 1
Node 1 kW
Deployable power, eclipse continuity, thermal rejection, owned-bus autonomy, customer workload, and disposal.
1 kW continuous10 kW BOL solar
02
MISSION 2
Node 10 kW
Commercial utilization, high-rate payload data flow, operational scheduling, and repeatable service economics.
10 kW continuous30.0 kW BOL solar
03
SCALE STAGE
Node 100 kW
Large deployable dynamics, modular integration, industrial workload demand, and launch architecture at scale.
100 kW continuous288 kW BOL solar
04
NETWORK STAGE
Grid 1 MW
A ten-module reference cluster, not a claim of one spacecraft or one launch. Architecture follows measured node economics.
1 MW aggregate2.88 MW aggregate BOL solar
04 / THE STACK
Hardware, runtime,
and flight evidence.
The product is not an accelerator placed in a box. Useful orbital compute is a coupled power, thermal, data, reliability, communications, and operations system.
FLIGHT NODE
Integrated orbital compute
Compute module, storage, PMAD, battery, deployable solar, thermal transport, communications, ADCS, propulsion, and disposal.
SensorRuntimeComputeResult
SOFTWARE
Radiation-aware runtime
Signed workloads, monitoring, checkpoint, rollback, fault isolation, and reproducible telemetry.
CONTROL
Power-thermal scheduler
Workload admission based on energy, temperature, communications, pointing, and mission priority.
COMMERCIAL OUTPUT
Evidence, not abstract FLOPS
Customer value is measured through reduced downlink, time to decision, verified model quality, available processing windows, and a traceable result package.
05 / PUBLIC BASELINE
Read the engineering case.
Version 0.1 is a transparent, pre-SRR working baseline. It exposes assumptions, separates known facts from model outputs, and states which supplier and customer evidence must replace early estimates.
WHITEPAPER / ENGLISH / PDF
Orbital Computing Infrastructure
Program thesis, market entry, proposed mission ladder, parametric model, evidence gates, and public risk register.
Version 0.1
MISSION SUMMARY / ENGLISH / PDFNode 1 kW Public Mission Definition
Mission purpose, success criteria, preliminary architecture, public requirement categories, risks, and review gates.
Pre-SRR baseline
What v0.1 is not
Not flight-release data, not manufacturing CAD, not a launch reservation, not a supplier quotation, and not an offer to sell securities.
SATELLITE INFERENCE
The next data center
may begin as a payload.
We are looking for spacecraft operators, sensor owners, compute and thermal partners, and U.S. mission integrators.
Start a technical conversation