Timer, Window & RTT Scaling in 5G NTN

Introduction

Timer and RTT scaling is the set of Rel-17 NR-NTN adaptations that keep the protocol stack from mistaking propagation delay for failure. It is not one feature but a pattern applied across every layer — MAC, RLC, PDCP, RRC and up into NAS — together with one new master lever, cellSpecificKoffset (K_offset), broadcast in SIB19.

It matters at every request-response exchange in the connected UE's life: random access, scheduling requests, HARQ, ARQ, connection setup, handover, radio-link-failure detection, and registration. On the ground each of these waits a sub-millisecond round trip; over a satellite each waits tens to hundreds of milliseconds. Any timer that was implicitly sized for the terrestrial case fires early on a perfectly healthy satellite link unless it is widened.

The two mechanisms are deliberately kept distinct. K_offset changes when an uplink transmission is scheduled relative to its grant, so the transmission lands in the future rather than in an already-passed slot. The extended timer ranges change how long the stack waits before declaring loss. Confusing the two is the most common conceptual error in this area, so this page separates them layer by layer and then gathers everything into two reference tables.

Why terrestrial timers break over satellite

A timer is a bet on how long a legitimate response can take. Size it just above the worst-case terrestrial round trip and it fires only on genuine failures. Move the same bet to a link whose round trip is a hundred or a thousand times longer and it fires constantly on successful exchanges — the response was always coming, it just had further to travel.

MAC — K_offset and the access/HARQ timers

The MAC layer (TS 38.321) is where the round-trip problem bites first, because MAC drives random access and HARQ, both of which are tight request-response loops. The linchpin is cellSpecificKoffset (K_offset): it shifts the uplink scheduling timing relationships so that a grant received now points to an uplink slot far enough ahead to be reachable across the propagation delay. Without it, a grant would schedule an uplink transmission at a time that, once the delay is accounted for, has already passed. K_offset underpins both the random-access timing (see random access) and the timing-advance framework (see timing advance). A companion offset, kmac, shifts the downlink action timing for the same reason.

K_offset applies to the uplink timing relationships a UE otherwise derives from downlink control: the K2 gap from a DCI format 0_0/0_1 grant on PDCCH to the scheduled PUSCH, the gap from PDSCH to its HARQ-ACK on PUCCH, the RAR-grant-to-Msg3 timing, and the timing of aperiodic CSI and SRS. It is a cell-level value (1…1023 slots referenced to 15 kHz SCS, so up to ≈1 s), broadcast in SIB19, and it is added on top of the normal terrestrial timing gaps rather than replacing them — so a UE that already knows K2 simply adds K_offset before acting.

On top of that scheduling shift, several MAC timers and windows are given extended ranges. The random-access reply window ra-ResponseWindow and the ra-ContentionResolutionTimer must be long enough for Msg2/Msg4 to return across the link. The scheduling-request prohibit timer sr-ProhibitTimer and the BSR and PHR timers must not fire before a grant can come back. And the DRX HARQ round-trip timers drx-HARQ-RTT-TimerDL and drx-HARQ-RTT-TimerUL, together with drx-RetransmissionTimer, must be stretched so the UE does not start hunting for a retransmission before the real HARQ feedback has had time to arrive (see HARQ in NTN).

RLC and PDCP — protecting reliability over the delay

Above MAC, the ARQ and reordering timers matter more in NTN than on the ground, because when HARQ feedback is disabled or reduced (as it often is over long RTT — see HARQ in NTN), RLC Acknowledged Mode carries more of the reliability load. Its timers (TS 38.322) must be extended so ARQ does not falsely conclude that data is lost while it is merely in transit: t-Reassembly waits for a segmentation gap to fill, t-StatusProhibit paces how often status reports are sent, and t-PollRetransmit decides when an unacknowledged poll should be re-sent. Each must comfortably exceed the NTN round trip, or the transmitter will re-poll and the receiver will NACK data that was always going to arrive.

There is a second-order constraint here too: the RLC AM window itself. Because a poll and its status report take a full RTT to complete, far more PDUs sit unacknowledged in flight over a satellite than over a terrestrial cell, so the sequence-number space and the AM transmit/receive windows must be large enough that the window does not stall before a status report can advance it. Under-sized timers and an under-sized window produce the same visible symptom — throughput collapses as the transmitter waits — for different reasons.

PDCP (TS 38.323) has the same exposure at the packet level. Its discardTimer decides when a buffered SDU is dropped as too old, and t-Reordering decides how long to wait for an out-of-order packet before delivering past it. Both must be sized for the long one-way delay so that packets which are simply late — not lost — are neither discarded prematurely nor declared missing.

RRC and NAS — not declaring failure too early

The most visible failures are the RRC ones, because they abort a whole connection. The RRC timers (TS 38.331) that govern connection setup and radio-link failure are given extended enumerations for NTN so the UE does not give up while a message is still crossing the link. T300 guards the RRCSetupRequest — the wait for a setup response; T301 guards re-establishment; T304 guards handover completion. The radio-link-failure machinery is the subtlest: T310 runs after a run of out-of-sync indications (counted by N310), and is only stopped by enough in-sync indications (N311); T311 then bounds the re-establishment search. Over an NTN RTT, a healthy link can look momentarily out-of-sync simply because feedback is delayed, so T310/T311 (and the counters) are extended to keep a working satellite link from being torn down as a false radio-link failure. T319, guarding the resume request, is likewise widened.

The delay does not stop at the RAN. NAS and higher-layer timers on the core side must also tolerate the extra propagation — procedures like registration and their acknowledgements travel the same long path, so their guard timers need equivalent headroom (see NTN & the 5GC). The unifying picture is one master scheduling offset plus a family of widened windows.

Terrestrial vs NTN round trip against a fixed window

The clearest way to see the problem is to draw a fixed timer window and lay two round trips against it. On the ground the reply returns well inside the window. Over NTN the same reply returns long after an unscaled window would have expired — unless the window is widened and the uplink transmission is pushed out by K_offset so the exchange fits.

What scales, and by how much — a reference

A layer-by-layer view of what NTN stretches and why. The master lever is cellSpecificKoffset (from SIB19); the individual timers gain larger value ranges in the Rel-17 ASN.1.

Summary

NTN breaks terrestrial timers because it changes exactly one physical fact — the round trip — by two to three orders of magnitude, and almost every timer in the stack had that round trip baked into its range. Rel-17 fixes it with two clearly separated tools: cellSpecificKoffset (K_offset), the master uplink scheduling offset broadcast in SIB19 that pushes every uplink timing relationship into a reachable future slot; and a family of extended value ranges on the affected MAC, RLC, PDCP, RRC and NAS timers so a legitimately delayed reply is still accepted.

Walk it by layer and the pattern repeats: MAC widens the access windows (ra-ResponseWindow, ra-ContentionResolutionTimer) and the DRX HARQ timers; RLC widens t-Reassembly/t-StatusProhibit/t-PollRetransmit because AM carries more load when HARQ is turned down; PDCP widens discardTimer/t-Reordering; RRC widens the setup and RLF timers (T300/T301/T304/T310/T311/T319 and the N310/N311 counters); NAS widens its own guard timers end to end. The one-sentence takeaway: K_offset changes when, extended ranges change how long — and NTN needs both, at every layer.