A Lunar Relay Is a Geometry Problem Before It Is a Satellite

The far side of the Moon is the only radio-quiet neighbourhood within reach, and that quiet is not a feature you can buy — it is the shadow of 3,474 km of rock. Every mission that wants to work there needs something on the other side of the shadow to carry its bits home, and the hard part is not the spacecraft. It is the orbit, the contact schedule, and the protocol that keeps a stored packet alive until somebody asks for it.

People talk about lunar relays the way they talk about Wi-Fi extenders: put one somewhere in the middle and the dead zone goes away. That framing hides the three things that actually decide whether a relay architecture works. A relay has to be visible to the ground station, visible to the surface asset, and awake with enough power to serve both at once. On the Moon, those three conditions are frequently not true simultaneously, and no amount of link budget fixes a geometry that is simply blocked.

Radio Silence Is Geometry, Not Interference

Radio at S, X and Ka band is line-of-sight propagation. There is no useful diffraction around a body with a 1,737 km radius, and no atmosphere to bounce off. When the Moon is between a lander and Earth, the link is not degraded — it is absent, and it stays absent for roughly half of each lunar day, which is about fourteen Earth days at any given far-side site. That shielding is exactly why the far side is scientifically valuable: it is the one nearby place screened from the ionosphere, the terrestrial noise floor and every transmitter on Earth.

The distance adds a second, subtler constraint. Mean Earth-Moon distance is about 384,400 km, which puts one-way light time at roughly 1.28 seconds. That is fine for commanding a rover. It is not fine for the transport protocols the rest of the internet runs on: a TCP round trip is 2.56 seconds before any processing, and the congestion window is built on the assumption that a lost segment gets reported back quickly. Add a link that vanishes for hours when the asset drives behind a crater rim and the assumptions collapse.

The Orbit Chooses the Product

The default answer is a relay in a halo orbit around the Earth-Moon L2 point, roughly 60,000 km beyond the far limb. It is a good answer for a specific reason: from there a spacecraft sees Earth continuously and most of the far-side hemisphere continuously, so a single relay avoids most mid-pass handovers. But a halo orbit is not stationary relative to the surface. The relay's view of a given site still sweeps, and low-elevation sites near the limb drop out of view regularly. A single relay in one halo orbit cannot serve an entire hemisphere at once, which is why coverage is quoted as a percentage of a site's day rather than as "the far side."

Two other families compete. High-eccentricity orbits with long dwell over the south pole are cheaper to reach, but they give long passes separated by long gaps, so you need a small constellation to stitch coverage — and a constellation needs crosslinks, a second and harder radio problem. Surface towers solve the last kilometre cheaply and do nothing about the Earth link: the horizon problem is the same at 30 metres as at ground level.

Every orbit choice drags a power budget behind it. A halo relay spends part of each month in the Moon's shadow, so its arrays are sized for the longest eclipse plus the transmit duty cycle of its busiest hour, not its average. And halo orbits are unstable: propellant for station-keeping becomes the real limit on mission life.

Store-and-Forward Beats a Better Modem

The protocol layer is where lunar networking stops looking like terrestrial networking. The Delay-Tolerant Networking architecture in RFC 4838 replaces the end-to-end conversation with custody: a bundle is handed to a node, that node takes responsibility for it, and it survives in storage across an outage. The Licklider Transmission Protocol in RFC 5326 handles the piece most likely to break — retransmission over a link whose round trip is measured in seconds or minutes and whose availability is scheduled rather than assumed.

That shift changes what the relay is. It is no longer a bent pipe; it is a storage device with an intermittent uplink, and its memory is a consumable exactly the way propellant is. If a rover dumps a day of imagery into a relay that has four other customers and a two-hour Earth contact, somebody's data waits, and the scheduling policy that decides who waits is a product decision disguised as a queue.

What "We'll Just Use a Satellite" Skips

Beyond a transponder, a relay has to carry:

• Power sized for eclipse, not for average draw, plus a battery that survives thousands of cycles.
• Thermal control for a bus that alternates between full sun and deep shadow on a two-week rhythm.
• Pointing good enough for optical crosslinks, which need the terminal to hold a beam narrower than the Moon's apparent size while the bus moves.
• Spectrum coordination for every band it touches, plus the ground stations to close the link.
• Scheduling and billing, because the relay is shared and somebody has to arbitrate the passes.

There is an upside worth naming: a relay carrying a disciplined clock and a known ephemeris can broadcast ranging and timing alongside data, turning the comms layer into a navigation layer. Ranging accuracy is a function of clock stability and orbit determination, though, so the navigation service is only as good as the least-funded part of the system.

What I Think

Single-relay lunar architectures are going to be oversubscribed within a year of going operational, and the first operators will find that out the hard way. The honest metric is not peak downlink rate; it is bits per site per lunar day at an availability you can plan a campaign around, and by that measure a relay that is dark for a fifth of the month and queuing four customers is a bottleneck with a launch date.

My read is that the agencies are right to buy relay capacity as a service rather than build bespoke spacecraft, not because commercial providers are cheaper, but because service contracts put the operations burden where it can be priced. The failure mode I expect is not a link budget that does not close. It is a schedule that nobody owns, a data queue that grows faster than the Earth contacts can drain it, and a far-side science community that discovers its instrument is the lowest-priority tenant on a satellite it does not control.