4.1 KiB
| title | tags | summary |
|---|---|---|
| Topology, handover and routing | isl, handover, intersatellite, routing, disruption | When there is more than one way across the network, the question of which way, and what it costs when the choice changes. |
Topology, handover and routing
A constellation network is not a set of point-to-point links to ground. It is a graph whose nodes move, and every problem of graph routing has to be re-asked every time the graph changes shape.
Definition
An inter-satellite link is a radio link between two satellites, giving laser or radio crosslink connectivity between them. Handover is the transfer of an active session from one satellite to another. Routing is the choice of path across the graph. Disruption tolerance is the mechanism that keeps a session alive through a handover.
The distinction that matters operationally is between a link that is planned and a link that exists opportunistically, because the first can be scheduled and the second cannot be.
Mathematics
Visibility of a crosslink
Two satellites can see each other when the elevation of one above the other's horizon is above a mask, which is the same geometry as the ground case with the observer at orbit altitude. The consequence is that crosslinks are intermittent: a pair of satellites in adjacent planes are in contact for a fraction of their relative motion, not continuously, and the contact interval is shorter for lower orbits.
Why handover costs
The cost of a handover is not the signalling; it is the buffering. A session in flight holds state, and that state must be transferred or reconstructed on the far side. The cost rises with the data rate on the session and with how much of the protocol's state is not reconstructible from the data itself.
This is why a system that can tolerate disruption pays for it in the protocol rather than in the network, and why the disruption-tolerant protocols are built around the assumption that a handover is normal.
Graph shape over time
Because nodes move, the adjacency matrix is a function of time, and a route optimal at one instant is not optimal at the next. Two approaches follow:
- Scheduled routing, which uses a precomputed time-varying topology and plans against it. Predictable, and limited to links that exist on schedule.
- Opportunistic routing, which uses whatever is present. Reaches further, and can fail at a moment when nothing can be done.
Capacity per node
A satellite is a network node with a fixed set of radio terminals. Its throughput is therefore shared between every session it carries, and the capacity available to one session falls as the constellation gets busier at the same location.
Unit conventions
- Crosslink data rate in bits per second per link; aggregate in the same.
- Handover interruption time in milliseconds.
- Link elevation in degrees; the horizon geometry uses the same quantities as the ground case.
Applicability
Routing decisions are constrained by the physics long before they are constrained by the protocol. A route that needs a link which is only present for 30 seconds per orbit is not a route; it is a hope.
The hardest case is not the dense shell, where a route exists, but the sparse case, where sometimes no path crosses a given region for minutes. Designs handle it by accepting the disruption and reconstructing state rather than by searching for a route that does not exist.
Capacity per node explains a behaviour that looks anomalous from outside: a service gets slower in a city during busy hours even though coverage is unchanged, because the same satellites are serving more sessions.
Standards
- Inter-satellite link characterisation and the laser physical layer: the CCSDS Inter-Satellite Laser Communications specifications
- Session continuity and disruption tolerance: the recommendations for handover and for the service data, in the CCSDS Space Link Protocol family
See also
- access-and-availability for the ground access a handover moves between
- constellations for the geometry that decides adjacency
- interference for the physical layer under the topology