Routing and protocol translation for low Earth orbit.

This note is written for engineers deciding whether the approach is sound. It assumes the reader knows what an optical terminal is and does not gloss its acronyms. The commercial argument is made elsewhere. What follows is the mechanism, the published work it rests on, and the order in which it would have to be proven.

Every figure and every substantive claim in this document carries one of two labels.

SOURCED

The claim comes from published literature or a government report, and carries a citation. It says nothing about Elevsoft’s own work.

DESIGN TARGET

Intent. A number or a behaviour the design is aimed at, not measured.

The card sits between the spacecraft optical terminals and the host bus. Ingress framing is parsed into packets, each packet is matched against a forwarding table already resident on the card, and the result is framed for whichever bus the host actually speaks. The forwarding table is an input rather than a product: it is computed on the ground for a given epoch and uplinked, so the lookup stage never waits on a protocol to converge. Fig. 1 shows the arrangement.

LEO topology is deterministic. Contact windows between satellites follow from ephemeris, so which neighbours a spacecraft can see, and when, is predictable well ahead of the pass. This is a property of the orbits, not of any particular design.[LEO networking survey]

The design generates forwarding tables on the ground per epoch, uplinks them ahead of time, and swaps them at scheduled link make and break events. Nothing converges in orbit; the routing decision has already been made by the time the packet arrives.

The alternative is running a terrestrial link-state protocol across the constellation. One published emulation of OSPF over a 48-satellite LEO constellation counts 235 LSU and LSAck packets for a single topology change, spent as control traffic on the inter-satellite links that are the scarcest resource on the spacecraft. Fig. 2 sets that figure beside what this design does instead.[OSPF convergence overhead]

Ingress is optical and mixed: SDA OCT profiles, vendor-proprietary OCT waveforms, CCSDS framing, Ethernet. Egress is whatever the host expects: SpaceWire and SpaceFibre on older buses, Ethernet on newer ones, CCSDS where the ground segment demands it. The spread is a fact of the terminal and avionics market as it stands, not an assumption of this design.[Terminal vendor landscape]

Translation happens in the datapath rather than in a host processor. At line rate there is no room for a store-and-forward round trip through software, or for a per-packet cost that grows with the number of distinct protocols in the constellation.

Fig. 3 is the ingress-to-egress coverage the design targets.

The design implements the SDA Optical Communications Terminal standard, v3.2.0 and v4.0.0, and the NEBULA network standard that sits above it, alongside CCSDS optical for civil interoperability.[SDA NEBULA v3.05]

In 2025 GAO recommended that SDA prove its laser links work in orbit in each batch of satellites before committing to the next, and noted that SDA had yet to demonstrate a link between two terminal vendors in space. A merchant part built to the published standard is aimed at that gap.[GAO-25-106838]

The commercial shape follows from the technical one: a card, plus the same core licensable as an IP block for integrators who would rather build it into avionics they already make.

Optical downlinks from small LEO satellites are specified at up to gigabits per second, and DLR notes that compact LEO satellites often cannot afford such a link on mass and power grounds.[DLR optical downlinks for small LEO]

Link-state routing and protocol translation both run at packet rate. In software on a general-purpose processor their cost scales with traffic, and we do not believe a small satellite’s power envelope has headroom for that at gigabit rates. That is the argument for putting the datapath in reconfigurable silicon and leaving the processor to configuration. It rests on the published rate figures above and on our reading of the power constraint, not on a measurement of ours.

A radiation-tolerant COTS-plus card for bench work and early flight demonstration, and a radiation-hardened flight unit for production programmes. The two sit at very different price and qualification points, and the split exists so that one piece of logic can reach both.

Both builds run the same routing and translation core, and that core is licensable on its own.

Packet-level simulation first, then hardware in the loop, then a ground interoperability test across terminals from more than one vendor, then a hosted payload for on-orbit heritage. Each step gates the one after it. Fig. 5 lists the sequence and the state of every step; work starts at its first line.