Satellite Constellations: LEO Routing
High-Speed Laser Links in Orbit
The Altitude Revolution
For decades, satellite internet meant Geostationary (GEO) satellites sitting 35,000 km up. The light-speed round trip alone created ~600ms of latency, making real-time applications and VoIP essentially unusable. LEO satellites sit at 550-1,200 km, bringing round-trip latency down to 20-40ms - competitive with transcontinental fiber links and far better than GEO's irreducible physics penalty.
Walker Delta Constellations
Managing thousands of satellites requires a structured orbital geometry. Starlink uses a Walker Star pattern - planes of satellites whose orbits converge at the poles - giving maximum coverage at high latitudes and creating predictable pole-to-pole orbital planes ideal for ISL routing. Amazon Kuiper uses a Walker Delta pattern, which favors equatorial latency for densely populated tropical regions.
The distinction matters for routing: in a Walker Star, the inter-plane links near the poles carry disproportionately high traffic as all cross-polar routes converge. Traffic engineering must distribute load dynamically to prevent polar orbital segments from becoming bottlenecks.
Laser Links: The Orbital Backbone
Early LEO satellites were 'bent pipes' - they had to see both the user and a ground station simultaneously to work, severely limiting coverage over oceans and remote areas. Inter-Satellite Laser Links (ISL) changed everything.
Satellites can now beam data to each other in a vacuum at the speed of light. Because light travels approximately 30% faster in a vacuum than in fiber-optic glass (due to the glass refractive index of ~1.5), an orbital path from New York to London can actually be faster than the subsea cable - even accounting for the extra vertical distance.
Ground Segment Synchronization
The handover is the critical moment. As a satellite disappears over the horizon (roughly every 5-7 minutes), the user terminal must seamlessly evaluate 40+ visible candidates and electronically 'steer' its phased-array beam to the next incoming satellite without dropping a single packet. This requires:
- Phased-Array Antennas: Electronically steerable arrays that can retarget in microseconds, eliminating the mechanical pointing delays of older dish systems.
- Predictive Handover: Because orbital paths are deterministic and published, the terminal can predict and pre-authenticate with the next satellite before the current one goes below the horizon, maintaining session continuity.
- Multi-Satellite Diversity: High-availability terminals simultaneously track two satellites, switching instantly with no service interruption - analogous to Wi-Fi 6's multi-link operation (MLO) but in orbit.
Conclusion
LEO constellations represent a fundamental restructuring of the global internet backbone. By moving routing intelligence into the vacuum of space and connecting orbital nodes with laser ISLs, operators are building a parallel internet backbone that is immune to fiber cuts, cable ship sabotage, and terrestrial geography. We are not merely extending the internet to rural areas; we are creating a lower-latency alternative to the terrestrial fiber grid for all traffic.
Multi-Layer Routing Optimization Across the Orbital Stack
The routing architecture in a modern LEO satellite constellation like Starlink is not a single protocol but a carefully layered stack of routing decisions operating at different timescales and abstraction levels. At the lowest layer, the physical routing of laser beams is a purely geometric problem: given the known positions of all satellites in the constellation (published in two-line element sets with meter-level accuracy), the onboard computer must calculate which neighbor is the optimal target for each of its four laser terminals. This calculation, performed every 100 milliseconds, considers the line-of-sight visibility (ensuring no obstruction by the Earth's limb), the angular rate of change (must be within the gimbal tracking limit of 0.1° per second), and the link distance (which affects propagation delay and free-space path loss). The result is a dynamically reconfigured physical topology that changes as satellites move through their orbital planes.
Above this physical layer sits the link-state routing protocol, which operates on a timescale of seconds rather than milliseconds. Each satellite broadcasts its link-state advertisements (LSAs) containing the current cost metrics for each of its active laser links—typically calculated as a weighted combination of propagation delay (the dominant factor, ranging from 3–20 ms depending on link distance), link utilization (as a percentage of the 100 Gbps capacity), and bit error rate (BER) on each link. The LSAs are propagated through the constellation using a controlled flooding mechanism that limits the time-to-live (TTL) to a maximum of 20 hops, ensuring that routing information does not propagate across the entire 10,000+ node constellation—which would create an unsustainable O(N²) update overhead—but instead remains localized to a region of roughly 200 satellites surrounding the origin. This region-based routing, analogous to OSPF's area concept but adapted to the orbital geometry, allows the constellation to scale to tens of thousands of nodes without collapsing under the weight of its own routing updates.
The third layer of routing optimization is the inter-plane coordination that happens at the ground control level. While the onboard routing protocol handles local dynamics (link failures, congestion, Doppler tracking), the ground-based network operations center (NOC) performs global traffic engineering on a timescale of minutes to hours. Using a digital twin of the entire constellation that incorporates the known orbital positions for the next 72 hours (which are deterministic and can be calculated with centimeter-level accuracy), the NOC computes the optimal end-to-end paths for the major traffic flows—for example, from the Pacific region aggregation point to the European ground station. These global path policies are then pushed to the affected satellites as policy-based routing entries that override the local link-state metrics. This hierarchical routing architecture—local reactive routing at the millisecond scale, regional link-state routing at the second scale, and global policy routing at the minute scale—is what enables the Starlink constellation to offer latency that is competitive with terrestrial fiber despite the enormous complexity of operating thousands of moving routers in space.
A particularly challenging aspect of LEO satellite routing is the handling of inter-plane handovers. As satellites in different orbital planes rotate in their respective orbits, the relative geometry between them changes continuously. A link that was optimal for routing traffic from Europe to Asia might become suboptimal 30 minutes later as the two satellites move to opposite sides of their orbital intersection. The routing protocol must anticipate these geometric changes and pre-compute the handover sequence. The technique used is known as "look-ahead routing": each satellite maintains a cache of routing tables computed for 5-minute intervals over the next 2 hours. As the actual time approaches, the satellite loads the pre-computed table and initiates the necessary laser link switches to align with the new topology. This predictive routing is computationally intensive—each 5-minute routing table for a constellation of 10,000 satellites requires solving a shortest-path problem over a graph with 10,000 nodes and 40,000 edges—but it is feasible because the orbital positions are deterministic and the computation can be distributed across the thousands of satellites, with each node computing only its local portion of the global routing solution.
The final element of the routing optimization stack is the integration with application-layer requirements via an explicit quality-of-service (QoS) framework. Not all traffic through the constellation is equally sensitive to latency variation. Real-time voice and video calls require consistent sub-50 ms latency with minimal jitter, while bulk data transfers (such as satellite firmware updates or weather data downlinks) can tolerate latency of several hundred milliseconds as long as throughput is high. The Starlink routing protocol assigns each traffic flow to one of four QoS classes: real-time (expedited forwarding), interactive (assured forwarding), bulk data (best effort), and control-plane (highest priority). The per-link cost metric used in the routing calculation is then adjusted based on the QoS class, so that real-time traffic is preferentially routed over the shortest available laser paths (even if those paths are more congested), while bulk data traffic is pushed to longer but less utilized paths. This QoS-aware routing ensures that the constellation can simultaneously serve latency-sensitive voice traffic and throughput-intensive video streaming without compromising either class of service, and it represents one of the key architectural innovations that distinguishes second-generation LEO constellations from their predecessors.
Regulatory and Sovereignty Constraints on Orbital Routing
While the engineering challenges of LEO satellite routing are formidable, the regulatory constraints may ultimately be even more restrictive in shaping how the orbital mesh operates. The fundamental principle of satellite communications—that a satellite can communicate with any ground station within its field of view—conflicts directly with the principle of national sovereignty, which dictates that communications traffic originating in one country cannot be intercepted or terminated in another country without regulatory approval. When a Starlink satellite passes over China, for example, the 1,160 km radius of its field of view encompasses not only Chinese territory but also parts of India, Russia, and the South China Sea. The routing protocol must therefore maintain a geofencing database that maps IP address ranges to permissible ground station jurisdictions and ensures that traffic is not downlinked in a country where it does not have landing rights.
The mechanism for implementing this geofencing in the routing layer is known as "localization-aware forwarding." Each satellite carries a read-only database of geopolitical boundaries, encoded as polygon regions in the World Geodetic System (WGS-84) coordinate frame. When a packet enters the orbital mesh from a ground station in the United States, the satellite appends a geolocation tag to the packet header indicating the regulatory domain of origin. As the packet propagates through the constellation, each intermediate satellite checks whether the current subsatellite point (the point on Earth directly below the satellite) falls within a restricted jurisdiction for that traffic class. If it does, the packet must be routed to a laser link that keeps it in space until the satellite moves past the restricted region, or it must be handed off to a satellite in a different orbital plane whose subsatellite point is over permissible territory. This regulatory routing constraint effectively creates "no-fly zones" in the orbital mesh—regions of space where packets of certain regulatory classes cannot be downlinked.
The complexity of regulatory routing is compounded by the fact that national regulations are not static. During political crises, countries may restrict or ban satellite communications over their territory for foreign-operated constellations. In 2022, for example, several countries restricted Starlink operations during the early stages of regional conflicts. The routing protocol must support a "fast-revocation" mechanism that can propagate regulatory changes across the entire constellation within minutes. The mechanism used is a regulatory update broadcast via the satellite control channel (typically operating at a lower frequency that penetrates atmospheric attenuation more reliably than the Ka-band user links). When the ground NOC issues a regulatory update, it is embedded in the telemetry, tracking, and command (TT&C) stream and relayed through the constellation using a prioritized flooding algorithm that bypasses the normal routing protocol's TTL limits. Within 60 seconds of a regulation change, every satellite in the constellation has updated its geofencing database and adjusted its routing tables accordingly.
Spectrum allocation is another regulatory dimension that directly affects routing decisions. The laser links operating at 1550 nm are not subject to the same spectrum licensing requirements as radio-frequency links, which gives Starlink significant operational flexibility. However, the ground-to-satellite links (the user links and the feeder links to ground stations) operate in licensed Ka-band (27–40 GHz) and Ku-band (12–18 GHz) spectrum, and the specific frequency allocations vary by country. A satellite passing over Europe may have authorization to use a different frequency band than when it passes over North America. The routing protocol must track the frequency allocation for the current geographic position of each satellite and ensure that ground links are established only using the authorized spectrum. If a satellite is approaching a geographic boundary where its current frequency allocation is no longer valid, the routing protocol must pre-emptively hand over the active ground connections to another satellite that is in a position where the authorized spectrum is available, or it must switch the ground link to a different frequency channel—a process that requires physical-layer reconfiguration and can introduce a transient outage of 50–100 milliseconds.
The most wide-reaching regulatory constraint on orbital routing is likely to be data sovereignty and privacy regulations such as the European Union's General Data Protection Regulation (GDPR) and similar laws in other jurisdictions. These regulations impose strict requirements on where personal data can be stored and processed, and they create complex routing constraints for a constellation that carries traffic for users around the world. When a European user's data packet enters the Starlink mesh, the regulatory routing layer must ensure that the packet is not downlinked at a ground station located outside the European Economic Area unless the destination server is also outside that jurisdiction and the traffic is classified as a permissible data transfer. This data sovereignty routing constraint has been a major driver of Starlink's investment in ground station infrastructure within each regulatory zone—the company now operates ground stations in over 40 countries, with a particular density in Europe, to ensure that traffic can be downlinked within the originating regulatory jurisdiction without requiring the orbital mesh to carry packets through restricted airspace. The routing protocol maintains a mapping of IP prefixes to regulatory zones and ensures that the egress ground station selection respects these data sovereignty requirements, even if the physically shortest path would exit in a different jurisdiction.