Messages across a moving network.

What space relays change

Northlink is a proposed delay-tolerant communications network joining community ground stations through a small constellation. This revision adds optical inter-satellite links to the worldwide network simulator, procurement directory and spacecraft model. It preserves the original low-rate ground-radio architecture and makes the enlarged optical variants explicit.

A satellite carrying a complete message can forward it to another satellite that can reach the destination sooner. Light travels at the same vacuum speed whether the carrier is radio or optical; the large improvement comes from avoiding orbital waiting. Laser terminals provide narrow beams and high potential data rates, but acquisition, Earth blockage, pointing and spacecraft resources still matter.

Each row compares the same orbital elements with relays switched off and on. The second row changes the constellation design first; its result is specific to that route and start time. Neither case is a measured service guarantee or a statistical global average.

Decisions introduced in revision 2

Two independently pointed optical terminals are allocated per optical spacecraft. The long-range comparison uses TESAT SCOT80; a smaller SCOT30 development option is also shown. Whole-bundle forwarding, a selectable acquisition delay and Earth-clearance checks determine whether a handoff is useful. Hardware is expanded to include terminal electronics, a faster router, larger arrays, power conversion, battery packaging and radiators.

Status: pre-procurement engineering study. No deployed Northlink satellites, flight-qualified integrated bus, verified optical link budget or committed spacecraft price is established by this paper. The simulator moves synthetic payload counters; it does not transmit files over a real satellite network.

Mission baseline and service boundaries

The original ten-node, approximately 40 kg concept remains the non-optical reference. The current map adds test locations and constellation alternatives; those map pins are planning coordinates, not commissioned stations. A custom town inherits the same radio assumptions unless its settings are changed.

A complete end-to-end route needs an upload opportunity, any selected space handoffs and a download opportunity. With one satellite, inter-satellite routing is impossible. With two or three widely separated satellites, useful pair geometry may be absent. Adding links does not enlarge the radio footprint seen by a ground dish.

Public internet access at a ground gateway is optional to the network architecture. A local user could submit a bundle over a terrestrial mesh and receive it at another independent mesh. The current transfer timer starts at the selected satellite ground terminal and ends at the destination terminal; terrestrial access and backhaul delay are outside that timer.

Live calls and uninterrupted streaming are not supported by the modeled service promise. Short modeled deliveries in a favorable geometry do not establish continuous worldwide service.

Architecture and custody

The data path is: local access network → source ground terminal → first spacecraft → zero or more optical relay spacecraft → destination ground terminal → local access network. The renderer shows possible optical pairs in cyan and highlights the selected hop during playback. A potential link is not a continuously active communications session.

The simulation counts one logical payload. Upload and download can resume over later ground contacts, but the selected carrier on each ground leg does not change. A space hop starts only when the full payload is present. A receiver becomes the next custodian after serialization and propagation complete. A one-second processing allowance precedes the next hop or the downlink.

A practical implementation should retain the sending copy until a verified acknowledgement arrives and apply expiry, deduplication and replay protection. Acknowledgements, retransmissions, duplicate storage and their capacity cost are not explicitly simulated. This paper uses “custodian” to describe the logical carrier; it does not claim that an optional custody-transfer extension is implemented by the browser.

Two optical heads allow a useful relay architecture and later scheduling flexibility. The present simulation only moves one payload through one optical hop at a time, so it does not assume an arbitrary number of simultaneous neighbors. The actual head fields of regard and pair scheduling remain part of the engineering design.

Optical hardware and readiness

The manufacturer channel rates are not the simulator net rates. Northlink deliberately uses lower study caps, but a lower rate does not itself prove an adequate received-power, jitter or coding margin. A paired-terminal optical budget and interface review must validate any operational setting.

SCOT80 is the default feasibility comparison because its published range supports more of the sparse constellation geometry. SCOT30 is explicitly a future development comparison. Neither product listing establishes delivery lead time, final configuration, export eligibility, integration compatibility or a public price.

RF crosslinks are another possible technology, particularly at shorter range, but would need their own antennas, frequencies, interference coordination and link budget. This revision implements the sourced optical candidates. It does not assume access to Starlink terminals or compatibility with a third-party satellite network.

Primary sources: TESAT SCOT80 data sheet, 20 August 2025; SCOT30 data sheet, 3 August 2026; TESAT product family. Exact optical wavelength, channel plan and interoperable pairing must come from the final vendor interface document.

Line of sight, range and acquisition

Earth is an obstruction. The distance from the Earth center to the line segment joining two satellites must exceed the Earth radius plus a 100 km study clearance. Each pair must also remain within the selected terminal range throughout acquisition, serialization and the short propagation allowance.

These ring thresholds apply only to a single uniformly populated orbital plane. Multi-plane constellations can create intermittent crossings and useful relay paths with fewer satellites. Conversely, having ten spacecraft does not guarantee a connected space network at every instant.

Continuous contact windows

For the model’s equal-altitude circular orbits, the normalized dot product of two position vectors is A + B cos(2nt) + D sin(2nt), where n is mean motion. The code derives the coefficients and solves the allowable intervals analytically. Earth rotation cancels in pair distance. Short crosslink opportunities are therefore not skipped on a coarse sampling grid.

Acquisition can be set to 0, 10, 30, 60 or 120 seconds. Thirty seconds is the default sensitivity assumption, not a manufacturer guarantee. Zero represents an already acquired link. A complete bundle must fit in the remaining uninterrupted window after acquisition. The final transmit-range distance divided by 299,792.458 km/s supplies the modeled propagation delay.

The geometry model assumes unobstructed pointing within each usable pair. It does not yet enforce terminal gimbal limits, body/wing occlusion, Sun exclusion, Earth brightness, ephemeris error, vibration or acquisition failure.

Routing, serialization and timing

For a fixed constellation, the planner first computes the best single-carrier route. It then seeds an earliest-arrival search with the time each spacecraft could finish uploading the complete payload. From the earliest available carrier, it tests the next full-bundle optical contact to each other spacecraft and the next achievable complete downlink. The best end time wins.

Keeping the no-relay route as a candidate means enabling space links cannot intentionally worsen a completed route in this model. A route may contain zero optical hops even with optical hardware selected. Each comparison uses the same payload, ground settings, time and orbital elements on both sides.

With a 10 Mbps ground cap, the two ground legs alone take at least 160 seconds for a 100 MB file. A one-gigabit laser hop does not turn the entire end-to-end route into a one-gigabit service. Large files can also miss a ground contact that a tiny text message could use, producing a much larger difference than serialization alone suggests.

Ground transmission is scheduled in five-second intervals and admitted only when start, midpoint and end satisfy elevation and RF headroom checks. The last interval is shortened to the remaining byte count. This is a conservative numerical approximation, not continuous contact integration. Ground-radio acquisition, ground propagation, ACK traffic and contention are outside the current timer.

Payload MB means 1,000,000 bytes. The model conserves uploaded, onboard and delivered logical bytes. A delivered payload is exactly complete; fractional floating-point residue is not displayed as undelivered data.

Measured comparison for 1–10 satellites

The following reproducible run uses the saved global-coverage preset for each satellite count. Source: Sandy Hook, Canada (50.55°, −96.99°). Destination: London (51.507°, −0.128°). Payload: 100 MB. Start: model second 0. Altitude: 550 km; mask: 20°; dish: 0.9 m; weather allowance: 3 dB; adaptive rate cap: 10 Mbps; SCOT80 study: 1,000 Mbps net; acquisition: 30 s.

The layouts at different counts were optimized for global ground access, not for monotonic performance on this one route. Consequently, five satellites can beat six at this particular start time. The paired on/off comparison within each row is the evidence for the relay benefit. With one, two and three satellites in these specific layouts, the fastest modeled delivery retains a single carrier.

Machine-readable orbital elements, endpoint coordinates and all settings are included in Northlink-Crosslink-Benchmarks.json. Values are rounded only for display. They are deterministic simulation examples, not percentiles or service-level commitments.

Global coverage and route optimization

Coverage and fastest delivery are different objectives. The global optimizer searches plane count, plane spacing, phasing and orientation, evaluates equal-area Earth samples over seven days, and minimizes mean waiting while retaining at least 99% of the original layout’s sampled average availability and payload capacity. Optical relays do not change direct ground visibility, so this coverage metric remains a ground-link calculation.

The route optimizer searches bounded layout candidates for the selected endpoints, payload and current model time using the active relay mode. When increasing satellite count, it retains the previous full fleet as a candidate and adds one spacecraft. This prevents the search from losing a previously achieved route purely because the candidate family changed. It is a repeatable bounded search, not proof of a global mathematical maximum.

This table uses the same Canada–UK settings as the previous page, but the orbital layouts have been optimized for that route. An early arrival at one time can sacrifice other routes or later opportunities. A deployed constellation cannot instantly change orbital planes; these are pre-launch design alternatives, not a maneuver command or an operational retuning capability.

Before selecting a real constellation, evaluate many cities, both directions, start times, outages and seasonal conditions; include launch deployment, collision avoidance, energy and traffic constraints.

Regional messages, pictures and video

Regional cases use the ten-satellite global-coverage preset at model second 20,070 with the same ground radio, altitude and SCOT80 study settings. “Text” is a 57-byte example, “picture” is 5 MB, and “video/file” is 100 MB. Content type does not change physics; byte count does.

The London–Accra route shows no benefit in this run because its best carrier already reaches the destination sooner than any useful relay alternative. Accra–São Paulo gains a useful handoff. São Paulo–London shows why a 100 MB transfer can wait much longer than text: the larger payload requires enough usable ground contact, not just the first moment of visibility.

The same search is available for the world city catalog and custom locations. A place name is only a coordinate shortcut. Terrain, nearby obstructions, measured rain statistics, installation height and actual RF equipment must be added before interpreting a town-level result as deployable coverage.

The downloaded benchmark file also includes all of these cases with the SCOT30 development profile. No claim of globally constant latency is made.

Spacecraft accommodation and mass

The original 40 kg system allocation is carried forward as the base subtotal. The increments sum to the stated optical-variant targets. Only the SCOT80 terminal mass is a specific published channel figure; SCOT30 uses a rounded upper allocation and the other increments are engineering allowances. Supplier bills of mass, harness, mounting, launch adapter and final contingency may require further growth.

The interactive spacecraft contains sixteen inspectable assemblies. The optical version shows two end-mounted terminal heads, terminal electronics, a payload router, additional battery/power packaging and enlarged radiators. Selecting the base profile hides the optical-specific assemblies and restores the smaller bus and arrays.

The GLB download depicts the long-range SCOT80 study in metres, with named assemblies and revision metadata. External envelopes guide the scene, but gimbals, internal optics, boards, cells, connectors and heat paths are illustrative. It is not a vendor CAD assembly, a dimensioned manufacturing drawing or a verified interference-free layout.

The larger bus changes inertia, center of mass, drag area, deployment modes and launcher requirements. The simulator has no structural dynamics, collision-clearance or thermal-accommodation solver. Detailed accommodation must resolve optical sight lines, Ku scan directions, plume zones and solar orientation together.

Power, eclipse energy and thermal design

At 550 km, the circular period is approximately 95.5 minutes. A worst-geometry cylindrical-shadow estimate gives about 35.6 minutes of eclipse. A constant 300 W load would consume about 178 Wh in that eclipse. A 500 Wh pack with 80% usable depth of discharge and 90% discharge efficiency yields 360 Wh in this simplified example; aging, cold temperature, contingency and battery current limits still need margins.

With about 59.9 minutes of sunlight, 600 W and an assumed 85% collection/conversion factor yield roughly 509 Wh per orbit, versus about 478 Wh consumed at a continuous 300 W load. That is a narrow illustrative energy margin before propulsion and other losses. A single active optical channel reduces this example load, but two-head duty, pre-acquisition and heaters need a time-resolved schedule.

The linked NanoPower P80 is a 300 W comparison and cannot be treated as a verified 600 W optical EPS. The procurement directory explicitly requires a redesigned system or a qualified parallel architecture. Propulsion may compete with communications for power and must be duty-managed.

At 300 K with emissivity 0.85, an ideal deep-space radiator emits roughly 390 W/m². Rejecting 300 W would need about 0.77 m² before Earth/Sun heating, view factors and margins. The long-range model allocates roughly 1 m² gross radiator area, including the enlarged bus radiator. This is a sizing illustration, not thermal closure.

Current route planning does not inhibit a link for low battery, thermal limit or propulsion priority. The mass, power and thermal budgets remain open engineering gates.

Pointing, navigation and optical operations

A narrow optical beam demands a terminal pointing, acquisition and tracking chain. Bus attitude control supplies coarse orientation and a stable platform; a terminal supplies the fine optical functions appropriate to its design. Existing reaction-wheel and sensor candidates are not evidence that Northlink already meets an optical jitter requirement.

Opposite-end heads in the 3D study provide a starting arrangement, not proof of all-sky access. Their allowed fields of regard must be intersected with the pair geometry. In a sparse constellation, rejecting a single head direction can remove the only useful path, so this is a priority extension before procurement.

The acquisition selector is intended for sensitivity testing. A scheduled, already locked pair can use zero additional acquisition in the model, while a newly selected partner can use a longer allowance. The implementation charges the selected allowance for each new space hop and only accepts a window long enough to complete the bundle.

The receiver-side propagation addition uses instantaneous range at the end of transmission. Relativistic clock handling, point-ahead angles and receiver motion during light time belong in the terminal/navigation implementation. They are not resolved by this network-level timing model.

Ground radio and antenna design

Free-space loss is 92.45 + 20 log10(f in GHz) + 20 log10(range in km). The ground gain uses 60% of the ideal dish aperture; the spacecraft gain uses the stated aperture area and 45% efficiency. Received power is compared against kT and the net bit rate, then the assumed requirement is subtracted to obtain headroom.

These are selected design assumptions. There is no measured modem profile, validated coding threshold or antenna scan pattern behind the adaptive rate table. A rate is allowed only when the selected profile has nonnegative headroom and the terminal is above its mask. The ground-network test uses the best available spacecraft at each sample and never multiplies one dish’s rate by constellation size.

A practical ground terminal needs a tracking dish or a compatible electronically steered aperture, Ku feed/duplexer, low-noise receiver, uplink amplifier, modem, timing/orbit data, controller, weather protection and protected power. A Starlink consumer terminal is not a drop-in Northlink terminal. Current hardware candidates require a complete interface and pointing design.

These frequencies are calculation inputs, not operating permission or a country-specific allocation decision. Licensing and coordination must be established for the selected service and locations before transmitting.

Ground mesh and independent gateways

The satellite ground terminal and local terrestrial mesh solve different distances. Satellite Ku frequencies are retained for the orbit path. Directional terrestrial backhaul uses a separate 5.8 GHz study profile, while local access uses appropriate independent channels and equipment. “Optimal” depends on local interference, path clearance and the applicable authorized limits.

The EIRP input is not a verified legal limit. Terrain, building obstructions, frequency availability, DFS behavior, measured spectrum occupancy and equipment certification are not imported automatically from the map. A planning pass must be followed by a field survey and current local requirements.

One gateway can accept local bundles, store them during an orbital outage, and forward them over a scheduled space route. Multiple independent ground stations reduce reliance on a single operator, but inter-gateway traffic policy, receipts and scheduling still need agreement. No user should have to expose message contents to a gateway to obtain service.

Ground power design should include station controller, modem, amplifier duty, dish motors, heaters or cooling, local radios, network equipment and battery autonomy. Foundation loads, cable runs, lightning protection, earthing, physical security and maintenance dominate some remote deployments and are not captured by a radio-only equipment subtotal.

Orbit tradeoffs, deployment and environment

The default 550 km orbit is a design-study starting point. Lower altitude reduces slant range and can improve RF headroom, but generally narrows the footprint and increases drag sensitivity. Higher altitude widens geometric reach while increasing free-space loss and changing disposal and radiation considerations. Inclination fixes the latitude reached by the subsatellite point; a footprint can extend beyond it when the radio and elevation constraints allow.

All displayed satellites use the same circular altitude and 97.6° inclination. Their plane orientations and phases vary. The equations omit J2 precession, eccentricity, injection dispersions, drag evolution, maneuvers, conjunction avoidance and seasonal Sun geometry. An inclination near a sun-synchronous value does not establish an actual sun-synchronous mission without the corresponding orbit design.

The propulsion model is an ideal sizing tool, not a selected thruster system. The existing ENPULSION Micro comparison sheet lists a 0.5–1 mN class system with 64–120 W and 28 V; suitability depends on thrust duty, total impulse, propellant, lifetime, plume and minimum operating point. Do not assume it can support a larger spacecraft at every chosen altitude.

Before a launch commitment, propagate the selected constellation with higher-fidelity dynamics, verify conjunction and disposal requirements, size injection and phasing maneuvers, obtain the launcher interface and close the power/thermal impact of propulsion. This paper does not issue a compliance or flight-readiness determination.

Software, security and network governance

Delay-tolerant networking suits a service where contacts are scheduled and interruptions are normal. Bundle Protocol version 7 is a relevant protocol basis; BPSec provides bundle security building blocks. A production implementation needs an explicit convergence layer over both Ku and optical interfaces, interoperable addressing, clock handling, expiry and tested forwarding behavior.

Encryption does not hide every metadata feature. Timing, size, routing and gateway presence can still reveal patterns. Node distribution alone does not create anonymity or remove legal obligations. A governance design should state what each operator can observe and which operational logs are retained.

The current browser simulator uses synthetic payload sizes and a single transfer. It does not implement live Bundle Protocol, encryption, wallet transfers, real spacecraft commands, multi-user traffic scheduling or guaranteed delivery. Its playback queue is a visualization of computed events. The website source separates geometry/radio math, optical contact routing, optimization, hardware data and interface rendering so the assumptions remain inspectable.

Future routing should account for energy, limited terminal occupancy, queue priority, failures and uncertainty. Replanning should preserve verified receipts and avoid duplicate delivery. Small messages may reasonably receive a different queue policy from large video files, but that policy must be visible and tested rather than hidden inside the latency claims.

Procurement, prices and programme scope

The directory contains 56 procurement entries. Each links to an official product, data sheet, supplier catalog or engineering enquiry. SCOT80 and SCOT30 are alternative terminal choices. The optical router replaces the base payload-router allocation, and the new optical EPS replaces the base EPS allocation; those alternatives must not be added as complete duplicate systems.

The four priced local-mesh item types total US$1,506 for the listed quantities. The two development SDRs add US$4,774, producing a US$6,280 subtotal across the five priced rows. This is neither a complete ground station nor a spacecraft or programme price. The other 51 entries require supplier quotations or configuration work. Published unit prices are dated and exclude taxes, shipping and installation.

No exact current SCOT80 or SCOT30 configuration price was verified publicly. Optical cost must be requested for two terminals per spacecraft, the selected data/range configuration, engineering support, flight acceptance, spares and delivery schedule. Router FPGA and memory design, EPS redesign, array/battery sizing, ADCS, thermal/structural work and qualification are additional scope.

A credible programme estimate should separate non-recurring engineering, recurring flight units, ground pilots, spectrum/coordination work, launch and integration, test facilities, operations, insurance where applicable and contingency. Earlier concept allowances are not treated as supplier-backed prices in this revision. No purchase, launch reservation or fundraising receipt is represented by this document.

The support page only publishes project-owner-supplied receiving addresses. No donation address has been invented. A future public contribution process should specify the receiving network, purpose, custody and reporting policy.

Verification, risks and next decisions

These are software checks, not hardware qualification. The largest unresolved factors are optical pointing access and jitter, received-power margin, flight router throughput, EPS capacity, thermal closure, mass growth, weather and obstructions at real stations, concurrent traffic and reliable acquisition under navigation uncertainty.

Practical next steps

First, request paired-terminal configuration and interface data from TESAT, including head/electronics allocation, pointing requirements, operational keep-out zones, channel plan and a two-unit quote. In parallel, close a CAD accommodation study and a time-resolved power/thermal model for the 80 kg case. Use the 65 kg case only with its stated development schedule.

Second, build a ground-only bundle-forwarding bench that reproduces scheduled contact outages and the modeled file sizes. Add terminal interface emulators, bandwidth caps, corrupted data, clock error and reboot cases. Test optical acquisition assumptions using supplier-supported equipment and procedures before incorporating their measured distributions into the simulator.

Third, validate the Ku modem/antenna chain and local mesh on authorized test arrangements, with measured pointing, interference and weather allowances. Then evaluate many start times and locations, congestion and failures. Release a revised feasibility decision only after mass, energy, thermal, optical and radio interfaces close.

Equations, reproducibility and sources

Reproduce with the committed model modules and verify-crosslinks.cjs. The benchmark JSON includes complete layout elements and settings. The coverage table uses saved coverage presets; the route table uses the bounded route search and preceding-fleet seed. Five-second ground admission and contact-edge tolerances can move an arrival slightly relative to a higher-fidelity scheduler.

Revision 2 replaces the current white paper and reader while preserving the complete original version as an archive. The revision synchronizes worldwide locations, 18 reference nodes, constellation optimization, three spacecraft profiles, optical routing, 56 procurement entries and the downloadable 16-assembly spacecraft.

Reference links follow on the next page. Manufacturer specifications and availability statements are attributed to the linked dated sheets; all Northlink net rates, budgets, geometry margins and simulation results are explicitly design assumptions or calculations.

Reference register

TESAT SCOT80 data sheet (20 August 2025) ↗

TESAT SCOT30 data sheet (3 August 2026) ↗

TESAT optical product family ↗

STAR-Dundee SpaceFibre IP cores ↗

NASA: what is CLICK? ↗

NASA: delay/disruption tolerant networking ↗

NASA Small Spacecraft Technology: communications ↗

IETF RFC 9171: Bundle Protocol version 7 ↗

IETF RFC 9172: Bundle Protocol Security ↗

GomSpace NanoPower P80 ↗

ENPULSION Micro comparison data sheet (17 March 2026) ↗

ITU-R P.618 propagation recommendation family ↗

Northlink Network Atlas and current artifact downloads ↗

Optical sources reviewed 28 September 2026. Existing hardware links and the five public price rows retain their 25 September 2026 check date. A source link does not certify compatibility, qualification, current inventory or authorization to operate. The following appendix carries the complete procurement register, including per-entry links and price basis.

Procurement register · 01–04

Procurement register · 05–08

Procurement register · 09–12

Procurement register · 13–16

Procurement register · 17–20

Procurement register · 21–24

Procurement register · 25–28

Procurement register · 29–32

Procurement register · 33–36

Procurement register · 37–40

Procurement register · 41–44

Procurement register · 45–48

Procurement register · 49–52

Procurement register · 53–56