NTN Architecture — Transparent vs Regenerative Payload

Introduction

A Non-Terrestrial Network (NTN) is 5G NR carried over a satellite or high-altitude platform instead of a ground tower. 3GPP introduced NTN as a work item in Rel-17 so that the same NR air interface (NR-Uu) a phone already speaks can reach it from orbit — extending coverage to oceans, deserts, aircraft and disaster zones where no terrestrial cell will ever be built.

The architecture question comes first because everything downstream depends on it. Where the gNB physically sits — on the ground behind the gateway, or on the satellite itself — decides what the feeder link carries, how long the radio round-trip is, whether satellites can route between themselves, and how much the 5G core has to change. Get the two payload models straight and the delay, mobility and timing pages that follow all fall into place.

This is a structural page: it names the physical elements (UE, satellite, gateway, gNB, 5GC), the two interfaces that can ride the feeder link (NG and F1), and the beam-to-cell relationship. The physics of altitude — delay, Doppler, footprint motion — lives in ntn-orbits, and the bands in ntn-spectrum.

Why NTN needs its own architecture

Terrestrial NR was designed around a base station a few kilometres away: a round-trip measured in microseconds, a cell that never moves, and a core that assumes both. A satellite breaks all three assumptions at once — it is hundreds to tens of thousands of kilometres away, it may be sweeping across the sky at 7.5 km/s, and the "cell" it paints can slide across the ground. The architecture's job is to insert that satellite into the NR path without forcing a redesign of the whole system.

The elegant trick 3GPP chose is to keep the interfaces the same and only move where they terminate. In every option the UE still runs NR-Uu and the core still speaks NG; what changes is whether the satellite is transparent to those interfaces or terminates them on board. That single design choice is what the rest of this page unpacks.

The building blocks

Before arguing about where the gNB lives, get the physical pieces straight. An NTN system inserts a satellite and a ground station into the path between the UE and the gNB, splitting the radio journey into two hops with very different roles.

The two hops matter because they are physically and often spectrally different. The service link is the low-band, low-EIRP link to a small terminal; the feeder link is a high-capacity trunk to a fixed gateway with a large dish, frequently in a different band entirely (Ka-band feeder against an L/S-band service link). The band split is detailed in ntn-spectrum. What rides across the feeder link — a raw radio waveform, or a network interface — is the crux of the two payload models below.

Transparent payload — the bent pipe (Rel-17 baseline)

The simplest satellite is a mirror. A transparent payload does nothing to the bits: it applies an RF filter, converts frequency between the service-link band and the feeder-link band, amplifies, and re-radiates. There is no demodulation, no decoding, no protocol awareness on board. This is why it is called a "bent pipe" — the signal goes up and comes back down, just bent toward the ground station.

Because the payload only translates frequency and amplifies, noise and interference picked up on the uplink service link are amplified and passed straight down the feeder link — there is no on-board cleanup. The gNB on the ground therefore sees the accumulated impairment of both hops. In exchange, the satellite is light, cheap and long-lived, and the operator can upgrade the entire "base station" by touching only the ground equipment. This is why direct-to-cell constellations launching today are overwhelmingly transparent.

Regenerative payload — a gNB in orbit (Rel-18)

Put processing on the satellite and the picture changes. A regenerative payload demodulates, decodes and re-encodes the signal on board — it actually recovers the bits before sending them on. That means the gNB (or a gNB-DU) lives on the satellite.

The split-gNB option is worth dwelling on because it is the natural middle ground. The gNB-DU — the lower layers (RLC, MAC, PHY) that must react to the radio in real time — goes on board, close to the antenna, so the tight HARQ and scheduling loops do not have to cross the feeder link. The gNB-CU — the slower RRC/PDCP layers — stays on the ground where it is easy to scale and update, joined to the DU by F1 tunnelled over the feeder link. This keeps the heavy, upgradeable brains on the ground while still terminating the radio in space.

Beams, cells and footprints

One satellite does not make one cell. It radiates many beams, and a beam maps to one or more NR cells. Firing dozens of narrow beams from a single platform lets the operator reuse frequency and polarisation across the coverage area — the same spectrum can serve many beams as long as neighbouring beams differ in frequency or polarisation, exactly like a terrestrial frequency-reuse pattern but drawn on the ground from orbit.

Where those footprints land, and how they behave over time, comes in three flavours:

A single LEO beam typically lights a patch tens to around a thousand kilometres across — vastly larger than a terrestrial cell — so a "cell" in NTN can span a whole region and hold thousands of users. The footprint choice above drives how often a UE must re-select or hand over, which is the subject of ntn-mobility and depends on the orbit family in ntn-orbits.

What the core sees, and the two payloads side by side

Here is the elegant part of the transparent design: from the 5G core's point of view, a transparent satellite plus its gateway just look like a gNB — one with an unusually large, time-varying delay. The core does not need to know a satellite is in the path. That is why 5GC changes for NTN are minimal; the heavy lifting is in the RAN and the UE. The core-side impacts that do exist — mostly around mobility, location and regulatory identity — are covered in ntn-5gc.

Summary

NTN drops a satellite into the NR path while keeping the interfaces intact: the UE always runs NR-Uu and the core always runs NG. The one decision that colours everything else is where the gNB sits. In a transparent payload (Rel-17 baseline) the satellite is a bent pipe — filter, frequency-convert, amplify — and the gNB stays on the ground behind the gateway, so the feeder link carries the raw NR-Uu waveform and the UE's round-trip includes both hops. In a regenerative payload (Rel-18) the satellite decodes and re-encodes on board, so a full gNB (feeder carries NG) or a gNB-DU (feeder carries F1, CU on ground) lives in space, the noise is cleaned each hop, the radio round-trip is the service link only, and ISL routing becomes possible.

Around that, a satellite paints many beams that map to cells whose footprints may be earth-fixed, quasi-earth-fixed or earth-moving — the source of NTN's mobility churn. And because a transparent satellite masquerades as a large-delay gNB, the 5GC barely changes; the real engineering lands in the RAN and the UE. Read this page structurally, then let ntn-orbits turn the altitude into hard delay and Doppler numbers.