The mechanism
Orbital compute needs at least two communication budgets: exchanges inside a distributed cluster and transfers to or from ground users. Optical links require acquisition, pointing, sufficient received power and an available path. Ground links add weather and contact-window constraints. NASA’s TBIRD demonstrated high-rate orbital downlink during particular passes, while Suncatcher’s disclosed high-rate result was a short ground bench path. Peak bit rate cannot be multiplied by a full day unless the link is actually usable for that whole interval. Data placement matters: processing a space-generated image locally may avoid much more traffic than serving a ground-resident training dataset.
- Separate intra-cluster traffic from ground ingress, outputs and operational telemetry.
- Calculate link power and pointing requirements for the actual range and apertures.
- Model contact windows, weather and payload efficiency before integrating delivered bytes.
- Add propagation, routing, queueing and processing to the latency path.
- Check whether the workload’s data generation and consumption rates fit both communication budgets.
Science and assumptions
Contact payload and propagation
Contact payload is bounded by efficiency times bit rate times contact duration divided by eight. One-way propagation time is path length divided by light speed.
Symbols and units
- Q
- Contact payload bound
bytes
- η
- Useful payload efficiency
fraction
- R
- Physical link bit rate
bits/s
- t_contact
- Usable contact duration
s
- 8
- Bits per byte
bits/byte
- T_prop
- One-way propagation time for the selected path
s
- d
- Actual signal path length
m
- c
- Speed of light along the modeled free-space path
m/s
R is bits/s, tcontact seconds, η payload efficiency, Q bytes, d meters and c≈3×10⁸ m/s.
Assumptions Use actual contact availability and path length; pointing, weather, coding and queueing are separate effects.
Engineering example
Synthetic continuous 200 Gbit/s for 300 s has a 7.5 TB raw upper bound. NASA’s 4.8 TB demonstrated pass is a measured delivered quantity and must remain separate from that ideal calculation.
Distinctions and tradeoffs
A five-minute burst demonstration does not establish continuous daily service at its peak rate.
Closer inter-satellite spacing can help link budgets but constrains formation flying; more ground stations can improve contact coverage while adding infrastructure.
Distributed training needs sustained exchanges, while compact inference outputs can tolerate much smaller ground-data budgets.
Specifications and interfaces
- Rates and volumes
- Bits/s, bytes/contact and contact duty cycle
Peak, bidirectional aggregate and sustained payload are distinct.
- Latency path
- Actual slant range and route
Altitude alone is only a special geometric lower-bound input.
Interfaces
- Spacecraft or ground data Input
- Datasets, activations, prompts and telemetry
- Available optical path Output
- Delivered useful bytes with pointing and contact constraints
Check your understanding
What changes when a link is available only one quarter of the time?
- Its peak bit rate must quadruple automatically
- A byte becomes four bits
- Its long-term average delivery bound falls unless other links fill the gaps
Answer and reasoning
Its long-term average delivery bound falls unless other links fill the gaps
Contact duty cycle reduces integrated delivery independently of the instantaneous rate while the link is active.
Evidence and scope
TBIRD was an orbital communication demonstration; Suncatcher’s 800 Gbit/s one-way result was on a ground bench. Neither alone qualifies a fleet-scale compute network.
Close a bounded payload solar, eclipse-energy, radiator and communications scenario; radiation, structural design and lifecycle costs remain unmodeled.