Why this is not a bandwidth problem
The instinct is to size a lunar network by throughput. That is the wrong first question. The Moon is 384,400 km away on average, light takes 1.28 seconds each way, and no engineering decision changes that. Every architecture choice below follows from latency and from line of sight, not from data rate.
A 2.6 second round trip is short enough for voice and long enough to make remote control of machinery unsafe. A rover driven from Earth travels blind for the length of the loop. At a walking pace of 0.5 m/s that is 1.3 metres of unseen ground on every command, which is roughly the size of the boulder that ends the mission. That single number is why autonomy is a communications requirement rather than a separate subsystem, and it is treated as one in the autonomy document.
The second constraint is that the base sits on a sphere with no ionosphere to bounce off. Anything over the horizon is simply gone. On a smooth sphere an antenna at 1.7 m sees about 2.4 km, and at the Shackleton rim the crater wall and the ridge cut that short in most directions. Once the site grows, the excavation face will be further away than a suit antenna can see.
The three tiers
The network is three separate problems that are usually conflated: talking across the site, talking to orbit, and talking to Earth. Each has a different failure mode and a different fallback.
Surface, inside 5 km
A cellular network on the mast covers the habitat, the pad, the plant and the excavation face. Nokia flew a Bell Labs designed LTE network (Long Term Evolution, the 4G cellular standard) to the surface on IM 2 in March 2025. The lander came to rest on its side, so the system had about 25 minutes of power. In that window it activated its components and sent operational data to Earth, but the rover and hopper it was meant to serve never deployed and no cellular call was made on the Moon. That is partial heritage, and it is still more than any other tier has.
Surface to orbit
Ka band to a relay satellite. NASA's LunaNet interoperability specification, at version 5 since February 2025, and ESA's Moonlight programme, whose Lunar Pathfinder relay is due to launch no earlier than late 2026 with initial services by the end of 2028, are both building this as shared infrastructure rather than mission specific links, which is what makes it worth designing against. The relay carries delay tolerant networking (DTN), covered in its own section below.
Orbit or ground to Earth
Ka band direct to Earth while Earth is above the rim, which at this site is about 57 percent of the time, and relay when it is not. The base never depends on a single path being up.
Why LTE, and not something purpose built
A bespoke radio would be lighter and would fit the environment better. It is still the wrong answer, for three reasons that have nothing to do with radio engineering.
First, terminals. Every suit, rover, camera and sensor needs a radio. Cellular chipsets are made in the hundreds of millions, are radiation tolerant enough in a shielded enclosure, and cost effectively nothing. A bespoke terminal costs six figures and has to be qualified again every time a new device appears.
Second, handover. Cellular networks solved the problem of a moving terminal crossing between cells in the 1980s and have been refining it since. A rover driving from the habitat to the excavation face is exactly that problem.
Third, the upgrade path. 5G is a software and radio head change on the same core. The architecture does not have to be redesigned when the site grows.
Cellular is not designed for vacuum, for ground that averages near 200 K on the sunlit ridge in the polar summer, reaches 220 to 300 K at the summer maximum, and falls to 50 to 70 K in winter, or for regolith. The hardware has to be repackaged into a thermally controlled, dust sealed enclosure, and that repackaging is most of the engineering. What is being reused is the protocol stack and the terminal ecosystem, not the enclosure.
Link budget, surface cell
The vacuum link is easier than the terrestrial one in every respect except thermal. There is no rain fade, no foliage, no multipath from buildings and no atmospheric absorption. What there is instead is an absolutely hard horizon and a noise floor set by the regolith, which on the sunlit ridge runs from near 200 K on a summer average to 300 K at the summer maximum.
| Term | Value | Note |
|---|---|---|
| Transmit powermast radio head | 10 W | 40 dBm, thermally limited not power limited |
| Transmit antenna gainsector | 14 dBi | 3 sectors, 120 degrees each |
| Path loss2 km at 2.1 GHz | 105 dB | free space, no atmospheric term |
| Receive antenna gainsuit | 2 dBi | omnidirectional, body shadowing allowed for |
| System noise temperatureregolith facing | 290 K | hot soil dominates over sky |
| Received power | minus 49 dBm | well above the minus 100 dBm sensitivity floor |
| Margin at 2 km | 51 dB | the link is horizon limited, not power limited |
The margin is enormous, and that is the point. The link does not fail because of distance. It fails because the far end drops below the horizon or behind a berm. Coverage is therefore a geometry problem solved with mast height and relay nodes, not a power problem solved with a bigger amplifier.
Coverage and the horizon
Antenna height sets everything. On a smooth sphere the distance to the horizon is the square root of twice the radius times the antenna height. For the Moon, with a radius of about 1,740 km, that works out to about 1.86 km times the square root of the height in metres, which is why the mast is 8.5 m rather than 3 m. That is the smooth sphere figure. At the rim the real horizon is set by the crater wall and the ridge, and in most directions it is shorter.
Line of sight is not the only obstruction. The blast berm around the pad is 1.5 m tall and stands directly between the mast and anything working on the far apron. Two repeater nodes on the berm crest close that shadow, and they are the same radio head as the mast so there is no second spare to carry.
Delay tolerant networking
TCP, the transmission control protocol the internet runs on, assumes that a missing acknowledgement means congestion. Over a link with a 2.6 second round trip and scheduled outages measured in hours, that assumption produces a protocol that spends most of its time backing off. The standard answer, and the one LunaNet is built on, is the Bundle Protocol: store and forward, with custody transfer at each hop.
In practice this means the base never waits for Earth. Science data, telemetry and video are written to a store at the mast and forwarded when a path exists. A file transfer that spans a relay outage resumes rather than restarting. The crew see a mailbox, not a connection.
| Traffic | Priority | Path | Behaviour on outage |
|---|---|---|---|
| Crew voice and videoreal time | 1 | Relay, else direct | Degrades to voice, then to text |
| Caution and warningsafety | 1 | Both, duplicated | Never queued, always duplicated |
| Command and controloperations | 2 | Relay preferred | Queued, executed on arrival |
| Science and imagerybulk | 4 | Whichever is up | Queued at the mast, forwarded later |
| Housekeeping telemetrybulk | 4 | Whichever is up | Summarised, then forwarded |
What the outage schedule actually looks like
A single relay in a low lunar orbit is above the horizon for a few tens of minutes per pass. A frozen elliptical orbit of the kind LunaNet and Moonlight are planning gives much longer coverage from the apoapsis end, at the cost of range. Neither gives continuity from one spacecraft, which is why both programmes are constellations.
For planning purposes the base assumes a relay gap of up to four hours, and a direct to Earth path only while Earth is above the rim. At this site that is not permanent. The Moon is tidally locked, so from most of the near side Earth holds one patch of sky for good. From the pole it sits on the horizon, and libration carries it above and below the rim through each month. A published study of candidate south pole sites puts Earth visibility at the Shackleton rim at 57 percent of the time.
A site nearer the equator on the near side sees Earth all the time and can point a fixed dish once at installation. The Shackleton rim does not get that. Earth rises and sets with the libration cycle, and when it is up it is only a few degrees above the horizon, so the Ka band dish needs a gimbal and a clear line along Earth's bearing, and about 43 percent of the time it has nothing to point at. That is why the relay is not optional here, and why store and forward is the normal mode rather than the fallback.
Mass, power and what it costs to land
Failure modes
| Failure | Effect | Time to crew impact | Response |
|---|---|---|---|
| Mast radio head failssurface network down | No site comms | Immediate | Second sector head carries the load at reduced coverage. Crew fall back to suit to suit direct. |
| Relay constellation gapno orbital path | Bulk traffic queues | Hours | DTN stores at the mast. Direct to Earth carries safety and voice. |
| Ka band dish misalignsgimbal or pointing fault | Relay only | Hours | Manual repointing on an extravehicular activity (EVA). The relay carries traffic meanwhile, and Earth is below the rim for much of the month anyway. |
| Suit radio failsone crew member silent | Safety critical | Minutes | Never fewer than two crew outside. The buddy relays and the EVA is terminated. |
| Dust on the dish feedgain falls | Slow degradation | Weeks | Electrodynamic clearing of the kind NASA demonstrated on Blue Ghost in March 2025. The 51 dB margin above belongs to the surface cell, not to the Ka band link, so the dish is the one antenna that has to stay clean. |
No single failure may leave a crew member outside without a path to the habitat. That is why the EVA rule is two people minimum, and why the suit radio is the only component on the base carried in quadruple redundancy.
What is not yet known
When LunaNet or Moonlight will have coverage worth pointing at is not known. Lunar Pathfinder is due no earlier than late 2026 and Moonlight's initial services by the end of 2028, so the relay gap is planned at a worst case of four hours and the first increments run without a relay. Earth visibility from the exact mast position has not been computed. The 57 percent figure is for one point on the Shackleton rim in a published site study, and a few hundred metres of position or a few metres of mast height can change it either way. The horizon around the site has not been surveyed from the ground, so two berm repeaters is an estimate. The repackaging of cellular hardware for vacuum, the polar thermal range and regolith is most of the engineering, and it has run once on the Moon, for about 25 minutes, on IM 2. The cabling mass is the item most likely to be underestimated.
Increment plan
Increment 1, land with the mast
The mast goes up before the habitat, because construction telemetry and rover control need it first. One sector of LTE and a direct to Earth link is the minimum viable network, accepting that Earth is below the rim about 43 percent of the time until a relay is on station.
Increment 2, three sectors and DTN
Full site coverage and the store and forward router. This is the point the base stops needing a continuously staffed ground link.
Increment 3, berm repeaters and relay
Shadow filling around the pad, and the Ka band relay terminal once LunaNet or Moonlight has coverage worth pointing at.
Increment 4, 5G radio heads
Same core, same mast, new radio heads. Buys the throughput for real time video from multiple construction machines at once, which is what the autonomy work needs.
Sources and further reading
Every figure in this document traces to one of these. Agency documents and peer reviewed work first, reporting only where it is the primary record.