Vivek Roy

5G NR-NTN vs DVB-S2X: A PHY and MAC Comparison for Satellite RF and IC Designers

Written by

vivekroyweb

in

Status as of September 2026. Covers 3GPP Rel-17 to Rel-20 NR-NTN and IoT-NTN, and ETSI EN 302 307-1/-2 (DVB-S2/S2X), TS 102 606 (GSE) and EN 301 545-2 (DVB-RCS2).

DVB-S2X is a single-carrier broadcast/broadband forward link built to get close to Shannon through a saturated transponder. 5G NR-NTN is a bidirectional, closed-loop cellular air interface: the UE compensates for the satellite channel (GNSS plus ephemeris for timing and Doppler), so that standard OFDM/HARQ/RRC machinery can serve unmodified handsets. For a VSAT or broadcast forward link, S2X remains the more power- and spectrum-efficient PHY. NR-NTN is the better system wherever mobility, handhelds, uplink multiple access and core-network integration dominate the requirements.

TL;DR

  • PHY: DVB-S2X (APSK/PSK single carrier, roll-off down to 0.05, 64800-bit LDPC+BCH, 90-symbol PLHEADER, ~2.4 % pilots, VL-SNR to about −10 dB Es/N0) typically runs 2–5 dB closer to HPA saturation than NR’s CP-OFDM and carries roughly 3–5 % L1 overhead, against ~20–30 % for NR. NR-NTN answers with DFT-s-OFDM/π/2-BPSK on the uplink, UE-side Doppler/TA pre-compensation (±0.1 ppm residual frequency error), and repetition/HARQ rather than very long codewords.
  • MAC: DVB relies on a gateway scheduler (ACM/VCM on the forward link; MF-TDMA with a Terminal Burst Time Plan, CRDSA/slotted ALOHA and NCR/CMT sync on DVB-RCS2 return) and has no HARQ. NR-NTN keeps dynamic and configured-grant scheduling but stretches every timing relationship (K_offset up to 1023 slots, k_mac, RTT offsets on ra-ResponseWindow, contention resolution and DRX HARQ timers), extends HARQ to 32 processes with per-process feedback disabling, and adds satellite-aware mobility (time/location CHO, Earth-fixed tracking areas).
  • Fit and direction: S2X/RCS2 remains the right choice for GEO/HTS broadband, broadcast, backhaul and trunking, and DVB-NIP makes it native-IP. NR-NTN wins direct-to-device and Ku/Ka VSAT integrated with a 5GC. Rel-18 added Ka-band (n510–n512); Rel-19 (completed December 2025) adds regenerative payloads, RedCap and uplink capacity, with Ku-band in progress; Rel-20 is working on GNSS-resilient NTN and NB-IoT voice over GEO. The two ecosystems are converging at the IP/service layer, not at the waveform.

Key findings

  1. The two standards split the work differently. DVB-S2X puts the complexity in the receiver: blind acquisition of large frequency offsets, VL-SNR burst detection, equalisation, and 64800-bit LDPC decoding. NR-NTN puts it in UE-side prediction: GNSS position plus the SIB19 ephemeris gives a computed timing advance and Doppler pre-correction. That lets the gNB treat NTN UEs almost like terrestrial ones, and it also makes GNSS a single point of failure, which Rel-20 is now addressing.
  2. Waveform efficiency favours S2X on nonlinear transponders. ITU-R S.1878 reports that DVB-S2 single-carrier transmission through a linearised TWTA at its optimum (about 0 dB input back-off) shows essentially no loss for PSK, negligible loss for 16APSK and noticeable loss only for 32APSK. The TR 38.821 satellite EIRP densities used for NR-NTN assume the satellite HPA runs at about 5 dB back-off. ETSI TR 103 297 finds OFDMA the most penalised because of its large PAPR, with SC-FDMA about 1 dB more power-efficient than TDMA in its study.
  3. The operating SNR regimes barely overlap. An NR handheld on an S-band LEO works around 0–10 dB C/N with G/T about −31.6 dB/K and 23 dBm/0 dBi uplink EIRP. A DVB-S2X VSAT or DTH terminal works around 5–20 dB Es/N0, and a professional trunk up to 20+ dB, where 256APSK (about 5.9 b/sym) becomes usable.
  4. Latency handling is the core of NR-NTN’s MAC changes. A GEO transparent RTT of 541.46 ms, against 25.77 ms for a 600 km LEO, drove K_offset (1..1023 slots of 15 kHz), 32 HARQ processes, HARQ feedback disabling, uplink HARQ mode B, and RTT-offset timers. DVB avoids the problem with no HARQ at all (strong FEC at QEF plus upper-layer ARQ/TCP acceleration) and a scheduler that works on superframe timescales.
  5. Overhead: S2X with pilots plus GSE loses roughly 3–5 % to L1/L2 framing, plus the roll-off excess bandwidth (5–35 %). NR loses ~7 % to the CP, 7–14 % to DMRS, 7–21 % to PDCCH, a few percent to SSB/CSI-RS/TRS, and 5–8 % to guard bands. What NR gets for that overhead is per-slot (0.125–1 ms) multi-user scheduling, MIMO and HARQ.

1. Scope, architecture and releases

Item 5G NR-NTN DVB-S2/S2X family
Core specs TS 38.211/212/213/214 (PHY), 38.300 (§16.14 NTN), 38.321 (MAC), 38.331 (RRC), 38.101-5 (NTN UE RF), TR 38.811/38.821 (studies) EN 302 307-1 (S2), EN 302 307-2 (S2X, incl. Annex E superframes and beam hopping), TS 102 606 (GSE), EN 301 545-2 (RCS2 lower layers), TS 101 545-1 (RCS2 system), TS 103 876 (DVB-NIP)
Payload Rel-17/18: transparent (gNB on the ground). Rel-19: regenerative (gNB on board) Transparent or regenerative; payload-agnostic
Orbits GEO, LEO (600/1200 km reference), MEO Any; historically GEO; S2X Annex E and VL-SNR target mobile/HTS
Duplex FDD only in NTN bands Forward: continuous TDM. Return: MF-TDMA (RCS2) or SCPC
Terminal classes Handheld PC3 (23 dBm, 0 dBi), IoT (NB-IoT/eMTC), VSAT/ESIM (Rel-18 Ka), RedCap (Rel-19) DTH receivers, VSAT/RCST, SNG, professional trunking, mobile (aero/maritime)

Release timeline

  • Rel-17 (frozen 2022): first normative NTN release. Transparent payload, FDD FR1 bands n255 (L: UL 1626.5–1660.5 / DL 1525–1559 MHz) and n256 (S: UL 1980–2010 / DL 2170–2200 MHz), GNSS-capable UEs assumed, Earth-fixed tracking areas, and IoT-NTN in parallel.
  • Rel-18 (June 2024): bands n254 (UL 1610–1626.5 / DL 2483.5–2500 MHz) and the Ka bands n510/n511/n512 (UL 27.5–30 GHz, DL 17.3–20.2 GHz) for VSAT/ESIM; a 30 MHz channel bandwidth in FR1-NTN; PUCCH repetition for Msg4 HARQ-ACK and PUSCH DMRS bundling; network-verified UE location; NTN–TN mobility enhancements.
  • Rel-19 (completed December 2025): regenerative payload; DL coverage enhancements with new reference payload parameters (lower EIRP density per beam, beam power sharing); UL capacity via OCC-overlaid repetitions; RedCap in FR1-NTN; intended-service-area signalling for MBS broadcast; IoT-NTN store-and-forward on regenerative payloads, UL capacity and NB-IoT public warning; higher UE power classes and a 3 MHz channel bandwidth; a Ku-band workplan (13.75–14.5 GHz UL, 10.70–12.75 GHz DL).
  • Rel-20 (in progress): GNSS-resilient NR-NTN; NB-IoT voice over GEO, including UE transmit power up to 37 dBm; connected-mode LTE-TN ↔ NR-NTN mobility; the 6G study with NTN as part of the unified design. Stage-3 is targeted for March 2027 and the ASN.1 freeze for June 2027.

2. Waveforms, PAPR and power amplifiers

NR-NTN. The downlink is CP-OFDM. The uplink is CP-OFDM or DFT-s-OFDM (transform precoding), optionally with π/2-BPSK. Rel-17 reused the terrestrial waveforms unchanged. Typical 10−4 CCDF PAPR is about 10–12 dB for CP-OFDM, about 7–8 dB for DFT-s-OFDM QPSK, and about 4–5 dB for π/2-BPSK DFT-s-OFDM (generic engineering values, not 3GPP-specified). A satellite HPA carrying an NR downlink therefore has to sit several dB below saturation; TR 38.821 assumes about 5 dB back-off. A handset uplink is limited by the 23 dBm PC3 cap and MPR/A-MPR, so DFT-s-OFDM with π/2-BPSK is the coverage-limited mode.

DVB-S2/S2X. Single-carrier, root-raised-cosine shaped (roll-off 0.35/0.25/0.20 in S2, plus 0.15/0.10/0.05 in S2X), with QPSK, 8PSK, 8/16/32/64/128/256APSK, and π/2-BPSK in the header and VL-SNR modes. APSK uses few concentric rings, which keeps PAPR low and makes predistortion simple. S2X also defines “-L” MODCODs optimised for quasi-linear, multi-carrier-per-HPA operation with phase noise, gaining up to ~1 dB in some cases. The DVB channel models A (ideal linearised hard limiter) and B (IMUX/OMUX, nonlinear TWTA with AM/AM and AM/PM, ACI) optimise output back-off per MODCOD.

Roll-off in practice. Per the DVB-S2X white paper, going from α = 0.20 to 0.05 under a (1+α) bandwidth rule raises symbol rate by about 14 %, but receiver noise rises about 0.6 dB, so the net spectral-efficiency gain is about 7 % (DTH) and 8–11 % in the professional SNR range. With free symbol-rate optimisation through a real transponder the gain is below 2 %, and with rising uplink ACI 20 % roll-off can even beat 5 % by 0.1–0.2 dB. NR’s equivalent is its guard band: 106 PRBs in 20 MHz at 15 kHz SCS is about 95.4 % occupancy.

Nonlinear predistortion. For S2X, per-ring DPD on the gateway uplink and on-board linearisers are standard practice, and equalisers in enhanced receivers recover 7–9 % spectral efficiency at high symbol rates. For NR through a transparent payload, satellite HPA nonlinearity lands inside the OFDM EVM budget and cannot be undone by UE-side equalisation, so operators back off, use CFR and DPD in the gateway chain, or apply them on board with regenerative payloads. A single-carrier NTN downlink option is being discussed for 6G but is not in any 5G release.

3. Numerology, frame structure and pilots

Parameter NR-NTN DVB-S2/S2X
Basic unit OFDM symbol; slot = 14 symbols; 10 ms frame, 1 ms subframe 90-symbol SLOT; PLFRAME = PLHEADER (1 slot) + S slots of XFECFRAME
SCS / symbol rate FR1-NTN: 15/30 kHz (60 kHz for some bandwidths); Ka: 60/120 kHz, SSB 120/240 kHz Any symbol rate; common VSAT/DTH 1–70 Mbaud; wideband S2X to ~500 Mbaud
Channel BW n255: 5–20 MHz; n256: 5–20 + 30 MHz; n254: 5–15 MHz; n510–512: 50/100/200/400 MHz Rs(1+α), e.g. 36 MHz transponder → ~30–34 Mbaud
Slot duration 1 ms (15 kHz), 0.5 ms (30 kHz), 0.125 ms (120 kHz) PLFRAME ≈ 8100–33 282 symbols (~1.1 ms at 30 Mbaud for 64800-bit 8PSK)
Header No per-slot header; PDCCH/DCI carries allocation PLHEADER: 26-symbol SOF + 64-symbol PLSCODE (π/2-BPSK)
Pilots DMRS, PTRS, CSI-RS/TRS, SRS Optional 36-symbol pilot block every 16 slots → ≈2.4 %
Superframe None (SFN/hyperframe numbering only) S2X Annex E: 612 540 symbols; SOSF 270 + SFFI 450 symbols; formats 0–5

PLHEADER robustness. The S2 header works reliably to slightly below Es/N0 = −2.5 dB. S2X added a separate 900-symbol VL-SNR header identifying 9 MODCODs plus a dummy frame, keeping the length of a standard PLFRAME so legacy receivers can skip it without losing lock.

Annex E superframes. Formats 0–4 are all 612 540 symbols. The SOSF uses one of 256 orthogonal Walsh-Hadamard sequences (a per-beam signature). Formats 0/1 give about 2.56 % pilot density. Formats 2/3 bundle constant-size, aligned PLFRAMEs across co-channel beams, enabling multi-user detection and gateway precoding; DVB cites 20–100 % capacity gains with full or two-colour reuse. Format 4 targets wideband and VL-SNR use. Format 5 (beam hopping) allows variable superframe length so dwell times line up with the Beam Hopping Time Plan.

NR reference signals. In NTN, DMRS density is the main knob against Doppler residuals, plus Rel-18 PUSCH DMRS bundling for coherent combining across slots. PTRS matters mainly at Ka (120 kHz SCS). CSI reporting works as usual but on GEO the CSI is about 0.54 s stale when used.

4. Modulation, coding and efficiency

NR. QC-LDPC for data (BG1: mother rate 1/3, max 8448-bit information block; BG2: 1/5, 3840 bits; lifting sizes up to Z = 384; IR-HARQ with four redundancy versions). Polar codes for DCI, UCI and PBCH. Modulation from π/2-BPSK (UL) to 256QAM (1024QAM DL in FR1). MCS tables run from QPSK 30/1024 (≈0.059 b/RE) to 256QAM 948/1024 (7.41 b/RE). Link adaptation targets about 10 % initial BLER and relies on HARQ.

DVB-S2/S2X. BCH outer plus LDPC inner, 64800-bit normal and 16200-bit short frames (S2X adds VL-SNR medium frames). S2 has 28 MODCODs, from QPSK 1/4 at about −2.35 dB Es/N0 to 32APSK 9/10 at 16.05 dB. S2X adds finer code rates, 64/128/256APSK and -L variants, shrinking MODCOD spacing from 1–1.5 dB to about 0.4–0.5 dB. 256APSK 3/4 carries 5.90 b/sym at an ideal linear-AWGN Es/N0 of 19.57 dB (roughly 24 dB in the nonlinear channel at about 4 dB OBO). VL-SNR adds nine MODCODs using π/2-BPSK (the two lowest spread ×2) and QPSK 2/9, operating quasi-error-free down to near −10 dB Es/N0.

Distance to Shannon. QPSK 1/2 at 1.0 dB is about 1 dB from capacity; 256APSK 3/4 is about 1.9 dB away because of shaping loss. NR at 10 % BLER with short codewords is about 1–2 dB from the bound per transmission, and HARQ recovers part of that at the cost of RTT — exactly what NTN latency makes expensive.

NR-NTN DVB-S2X
Lowest operating point QPSK 30/1024 + repetition; IoT-NTN adds up to thousands of repetitions π/2-BPSK 1/5 spread ×2, about −10 dB Es/N0
Highest 256QAM 948/1024 (7.41 b/RE), 1024QAM DL (FR1) 256APSK 3/4 (5.90 b/sym)
Codeword ≤8448 info bits per code block 64800 / 32400 / 16200 bits
Error target 10 % BLER + HARQ QEF (no retransmission)
Link adaptation granularity ~1–2 dB MCS steps, per slot 0.4–0.5 dB MODCOD steps, per PLFRAME

5. Synchronisation and low-SNR acquisition

NR. The SSB is PSS and SSS (127 subcarriers each) plus PBCH in 4 symbols × 240 subcarriers. Initial LEO Doppler reaches 24 ppm: ±48 kHz at 2 GHz, ±480 kHz at 20 GHz, ±720 kHz at 30 GHz, so the UE must test several PSS frequency hypotheses on top of crystal error. Doppler rate reaches about −544 Hz/s at 2 GHz (TR 38.811), or 0.27 ppm/s at 600 km. After DL sync the UE computes the uplink Doppler from GNSS plus ephemeris and pre-compensates; TS 38.101-5 requires the UE carrier within ±0.1 ppm of the ideally pre-compensated reference. Timing follows T_TA = (N_TA + N_TA,offset + N_TA,adj common + N_TA,adj UE)·T_c.

DVB-S2/S2X. Blind acquisition via 26-symbol SOF correlation, PLSCODE decoding and pilot-aided tracking. Receivers typically handle coarse frequency offsets of the order of 20 % of the symbol rate (an implementation figure), which covers LNB drift and GEO Doppler easily. VL-SNR detection at −10 dB needs very narrow loop bandwidths (around 10−4 normalised), which lengthens acquisition and limits tracking of fast Doppler. On LEO a raw S2X receiver therefore needs ephemeris-driven NCO pre-correction, just as NR does — DVB simply never standardised that aiding.

DVB-RCS2 return link. Terminals derive time from the Network Clock Reference on the forward link, log on in dedicated or random logon slots (with support for large timing uncertainty), and receive burst frequency, timing and amplitude corrections via the Correction Message Table — the functional analogue of NR’s TA command plus TPC, but slower and aimed at GEO.

6. Doppler and delay: numbers side by side

Quantity (TR 38.821 / 38.811) GEO LEO 1200 km LEO 600 km
Max RTT, transparent 541.46 ms ≈41 ms 25.77 ms
Max RTT, regenerative 270.73 ms ≈20.5 ms ≈12.9 ms
Max differential delay in cell ≈10.3 ms — ≈3.12 ms
Max Doppler (Earth-fixed UE) 0.93 ppm 21 ppm 24 ppm

NTN-Config (SIB19) carries the ephemeris (state vectors or orbital elements with epoch time), ta-Common (up to about 270.73 ms full scale) and its drift terms, cellSpecificKoffset (1..1023 slots of 15 kHz, i.e. up to about 1.023 s, covering the GEO RTT), kmac (1..512 slots), and ntn-UlSyncValidityDuration (5 s to 900 s, driving timer T430). K_offset stretches every UL-after-DL relationship: DCI→PUSCH, RAR→Msg3, PDSCH→HARQ-ACK and MAC CE activation. DVB-S2X needs none of this on the forward link because nothing in it depends on round-trip timing.

7. Link budgets, terminals and bands

TR 38.821 Set-1 (per beam) GEO S LEO-600 S GEO Ka LEO-600 Ka
Satellite DL EIRP density 59 dBW/MHz 34 dBW/MHz 40 dBW/MHz 4 dBW/MHz
Tx gain / aperture 51 dBi / 22 m 30 dBi / 2 m 58.5 dBi / 5 m 38.5 dBi / 0.5 m
Beam diameter 250 km 50 km 110 km 20 km
Satellite UL G/T 19 dB/K 1.1 dB/K 28 dB/K 13 dB/K
  • Handheld: 23 dBm and 0 dBi, linear polarisation, NF 7 dB, giving DL G/T about −31.6 dB/K, plus a 3 dB polarisation mismatch against a circularly polarised satellite.
  • Ka VSAT: 60 cm equivalent aperture, 39.7 dBi Rx / 43.2 dBi Tx, NF 1.2 dB, giving DL G/T about 15.9 dB/K and UL EIRP 76.2 dBm.
  • The gap: VSAT G/T is about 47.5 dB better than the handheld’s. That single figure explains why S2X-class high-order MODCODs make sense for VSATs and not for phones.
  • DVB regimes: Ku/Ka DTH with 45–90 cm dishes (G/T around 10–18 dB/K) runs at 5–14 dB Es/N0; DSNG at 10–14 dB; trunking at 12–18 dB; VL-SNR mobile at −10 to −2 dB.

RF front-end implications. L/S NR handhelds have no antenna gain, so the system closes the link with satellite aperture (AST’s arrays are about 223 m²). Handset priorities are PA efficiency at 23 dBm with DFT-s-OFDM MPR, a low-NF LNA, and coexistence filtering (n256 overlaps terrestrial n65/n66; n255 overlaps n24/n99). Ka VSAT phased arrays for either standard need a few hundred to a few thousand elements; an NR VSAT must additionally track ±480 kHz (DL) and ±720 kHz (UL) Doppler and TA drift at LEO and support mobility RRM.

8. Multi-beam, beam hopping and MIMO

NR-NTN: typically one satellite beam per cell or per SSB beam; frequency reuse 1, 3 or 4 via carriers or BWPs; polarisation (RHCP/LHCP/linear) can be signalled so it serves as a reuse colour. Handhelds are 1T2R-class, so single-layer links are the rule; multi-satellite schemes remain research or 6G items.

DVB-S2X: beam hopping (Format 5) with a Beam Hopping Time Plan; co-channel interference handled with orthogonal SOSF/pilot codes and seven selectable Gold scrambling codes; precoding and MUD via formats 2/3; channel bonding of up to 2–3 transponders (CCM only) for statistical multiplexing.

9. Phase noise, interference and nonlinearity

  • OFDM turns phase noise into CPE plus ICI, which argues for wider SCS at Ka plus PTRS. Its high PAPR produces in-band EVM and spectral regrowth that a single-tap equaliser cannot remove.
  • High-order APSK is sensitive mainly to low-frequency phase noise on its outer rings; 256APSK needs integrated phase error well under 1° rms after the carrier loop. S2X supplies phase-noise masks (Annex H.8) for design.
  • Interference: NR relies on beam isolation and frequency reuse with PCI-based scrambling. S2X has selectable Gold codes and orthogonal superframe sequences, and PL descrambling spreads a narrower co-channel interferer, reducing its effect.

10. MAC and upper layers

10.1 Resource allocation and scheduling

NR-NTN DVB-S2X forward / DVB-RCS2 return
Granularity PRB × slot (0.125–1 ms), per UE, per HARQ process Forward: BBFRAME/PLFRAME per MODCOD (ACM) or per stream (VCM). Return: MF-TDMA slot per superframe (tens of ms)
Grant signalling DCI on PDCCH; configured grant type 1/2; SPS Forward: implicit (receiver filters by ISI/label). Return: TBTP2 broadcast on the forward link
Capacity request SR (PUCCH) + BSR MAC CE CRA, RBDC, VBDC, FCA capacity categories
Delay impact K_offset on all DL→UL timings; GEO SR→grant→PUSCH ≈ 2 RTT ≈ 1.1 s unless configured grant GEO VBDC request→allocation ≈ 0.6–0.7 s; CRA/FCA and RA hide it

10.2 HARQ vs FEC-only

  • NR-NTN (Rel-17): up to 32 HARQ processes in DL and UL; per-process downlink HARQ feedback disabling (reliability then falls back to RLC ARQ and a lower BLER target); uplink HARQ mode A/B per process; DRX HARQ RTT timers extended by the UE-gNB RTT. NB-IoT NTN has only 2 HARQ processes, so HARQ disabling (Rel-18) is essential there. Section 13 below walks through the mechanics in detail.
  • DVB: no HARQ at all. S2X targets QEF with 64800-bit LDPC and relies on ACM margin, TCP acceleration/PEP and application FEC. On GEO that is rational: a HARQ retransmission costs about 0.55 s, a delay DVB systems would rather spend as margin.

10.3 Random access

  • NR-NTN: 4-step and 2-step RACH with full TA and Doppler pre-compensation applied to PRACH. ra-ResponseWindow and msgB-ResponseWindow start after the UE-gNB RTT estimate; ra-ContentionResolutionTimer starts after Msg3 plus the RTT. Msg3 repetition and Msg4 PUCCH repetition (Rel-18) close the link for handsets.
  • DVB-RCS2: slotted ALOHA for logon/control slots (throughput peaks at 0.36 packets/slot); CRDSA transmits multiple burst replicas and cancels interference, raising throughput several-fold at the same packet loss rate, with back-pressure control in the FAT.

10.4 Timers, link adaptation and feedback

Rel-17 extended value ranges across the NR stack (sr-ProhibitTimer, BSR retransmission, RLC t-Reassembly and t-StatusProhibit, PDCP discardTimer and t-Reordering, the RLF family, NAS timers for GEO). LEO geometry is predictable, allowing elevation-based open-loop link adaptation; at GEO the CQI is about 0.54 s old. In DVB-S2X ACM, terminals report forward-link SNR over the return channel and the gateway picks a MODCOD with hysteresis. With a GEO loop delay of 0.5–0.7 s and Ka rain fades evolving at 0.1–1 dB/s, ACM margin is typically 0.5–1 dB — which is why S2X’s fine MODCOD spacing matters.

10.5 Mobility, paging and tracking areas

NR-NTN has Earth-fixed tracking areas (no TAU storms as beams sweep), conditional handover with time and location triggers, location/time-based measurement triggering with neighbour-satellite ephemeris in SIB19, and feeder-link switchover for transparent LEO. Rel-18 improved NTN–TN mobility, Rel-19 adds mobility for regenerative payloads, and Rel-20 adds connected-mode LTE-TN↔NR-NTN. DVB’s model is largely static: a terminal is provisioned to a gateway and beam, and beam or satellite handover (maritime/aero) is vendor-specific, with no standardised equivalent of paging, tracking areas or CHO.

10.6 Encapsulation, framing and QoS

NR: SDAP (QoS flow→DRB), PDCP (ciphering/integrity, ROHC), RLC (UM/AM, segmentation) and MAC (multiplexing, MAC CEs) add roughly 6–10 bytes per IP packet before ROHC. 5QI QoS runs end to end with the 5GC, and NR(LEO)/NR(MEO)/NR(GEO) RAT types tell the core the access is satellite.

DVB: an 80-bit BBHEADER followed by a data field of up to ~7 kB. Input is MPEG-TS (IP via MPE/ULE, about 10 % overhead) or Generic Stream with GSE (minimum 2-byte header, labels, fragmentation, CRC-32 on the last fragment only), cutting overhead to about 2–3 % for 5–15 % throughput gain. Each ISI stream can carry a different QoS or MODCOD class. DVB-NIP (TS 103 876) replaces TS entirely with IP over GSE, with multicast via FLUTE/ROUTE and DVB-I service discovery.

10.7 Overhead as a share of capacity (indicative)

Component NR-NTN DL (20 MHz, 15 kHz, SISO) DVB-S2X forward (normal frames, pilots on)
Spectral occupancy 95.4 % (106 PRB) 1/(1+α): 95 % (α=0.05) … 74 % (α=0.35)
Cyclic prefix / header ≈6.7 % CP PLHEADER: 0.3–1.1 %
Pilots / RS DMRS 7–14 %, CSI-RS/TRS 1–3 %, SSB ~1–4 % 2.4 % (plus superframe ≈0.1–2.6 %)
Control PDCCH 7–21 % None in-band (tables, typically under 1–2 %)
L2 ≈6–10 B per IP packet (~1–2 % at 600 B) BBHEADER ≈0.1–0.5 %, GSE 2–3 %
Total (indicative) ≈25–35 % of the raw OFDM grid ≈3–6 % plus roll-off excess

These totals are engineering estimates from the frame structures, not figures from a standard.

10.8 Latency and throughput

GEO transparent one-way UE→gNB is about 270 ms plus processing; LEO-600 is about 13 ms one-way at worst, and regenerative payloads roughly halve that. S2X in a 36 MHz transponder delivers about 60–200 Mbit/s depending on MODCOD, and wideband carriers reach Gbit/s. NR-NTN handheld evaluations and commercial systems show single-digit to tens of Mbit/s per beam. AST SpaceMobile advertises peaks above 150 Mbps per coverage cell (vendor design targets), while a crowdsourced study estimates Starlink Direct-to-Cell at about 3.1 Mbit/s downlink per beam.

11. IoT-NTN in brief

NB-IoT and eMTC over NTN (Rel-17) use the same broadcast ephemeris and common-TA concept with GNSS-based pre-compensation. NB-IoT’s 2 HARQ processes and repetition-heavy operation make HARQ disabling and RLC ARQ central. Rel-18 added HARQ disabling, improved GNSS-measurement gaps and discontinuous-coverage support; Rel-19 adds store-and-forward on regenerative payloads; Rel-20 targets NB-IoT voice over GEO. DVB has no equivalent.

12. Where each excels, and convergence

Use case Better fit Why
DTH/broadcast, multicast CDN feed, DVB-NIP DVB-S2X Near-saturation single carrier, QEF, massive multicast efficiency, deployed ecosystem
GEO/HTS broadband VSAT DVB-S2X + RCS2 (or S2X-based return) Proven ACM, beam hopping and precoding hooks, ~3–5 % overhead
5G backhaul / trunking DVB-S2X today; NR-NTN Ka VSAT emerging 256APSK at high SNR vs native 5GC integration
Direct-to-device smartphones NR-NTN (and LTE-based D2D) Unmodified handsets, standard chipsets, mobility, core integration
Massive IoT via satellite IoT-NTN PSM/eDRX, repetition, store-and-forward
Aero/maritime mobility Both NR has standard mobility; S2X has the better forward link

Convergence is happening at the IP and service layer: DVB-NIP makes S2X a native-IP multicast bearer compatible with 5G media delivery, DVB-I spans DVB, 5G Broadcast and unicast, and S2X is widely used for 5G backhaul. On the 3GPP side, Rel-19 adds broadcast service-area signalling and a Ku-band workplan, putting 3GPP directly into DVB’s traditional spectrum. 6G studies target a unified TN/NTN radio, GNSS-independent operation, and possibly a low-PAPR downlink option. Expect a multi-orbit mix: S2X for GEO broadband and broadcast, NR-NTN for D2D and IoT, integrated at the IP level.

13. HARQ transmission and the baseband chain to the DAC/ADC

HARQ is the single biggest structural difference between the two air interfaces, and it is easiest to understand by following a transport block through the whole baseband chain. This section explains how NR HARQ works, what NTN does to it, and then walks both transmit chains from input bits to the DAC (and the receive chains back from the ADC).

13.1 How NR HARQ works

  • Parallel stop-and-wait processes. Each HARQ process sends one transport block and waits for its ACK/NACK; many processes run in parallel so the link stays busy during the wait. The scheduling DCI carries the HARQ process number (4 bits for 16 processes, 5 bits when 32 are configured in Rel-17), a new-data indicator (NDI, toggled for a new TB, unchanged for a retransmission) and a 2-bit redundancy version (RV).
  • Incremental redundancy via a circular buffer. After LDPC encoding, the systematic and parity bits of each code block are written into a circular buffer. Each transmission reads E bits out of it, starting at one of four positions set by the RV. RV0 and RV3 start near the systematic bits and are close to self-decodable; RV2 and RV1 mostly add fresh parity. A common retransmission order is 0 → 2 → 3 → 1. Limited-buffer rate matching (LBRM) can cap the buffer to bound UE soft-buffer memory.
  • Soft combining at the receiver. The receiver de-rate-matches each transmission’s LLRs back into their buffer positions and adds them to what it stored from earlier attempts, then runs LDPC decoding on the combined buffer. The TB and code-block CRCs decide ACK or NACK.
  • Feedback. Downlink ACK/NACK goes back on PUCCH (or multiplexed on PUSCH) in slot n + K1, packed into a HARQ-ACK codebook: Type 1 (semi-static), Type 2 (dynamic, ordered by DAI) or Type 3 (one-shot report of all processes). With code-block-group (CBG) transmission, the TB is split into 2, 4, 6 or 8 CBGs and only the failed groups are resent.
  • Uplink. NR uplink HARQ is asynchronous and adaptive: there is no PHICH, and the gNB triggers a retransmission simply by scheduling the same process number with NDI not toggled.

13.2 What NTN does to HARQ

To keep one UE continuously scheduled, the number of HARQ processes must cover the round trip: NHARQ ≥ (RTT + processing time) / slot duration. With 1 ms slots (15 kHz SCS):

Scenario (transparent) Max RTT Processes needed (approx.) Effect with 32 processes
Terrestrial <1 ms ~8–16 Plenty
LEO 600 km 25.77 ms ~27–30 Just fits
LEO 1200 km ≈41 ms ~42+ Stalls without feedback disabling
GEO 541.46 ms ~542 With feedback on, one UE gets at most ~32/541 ≈ 6 % of slots

At 30 kHz SCS the slot halves and the required process counts double. The Rel-17/18 answers are:

  • HARQ feedback disabling per downlink process. A disabled process can be reused immediately. Reliability then comes from a more conservative MCS (lower BLER target), blind repetition (pdsch-AggregationFactor n2/n4/n8, PUSCH repetition) and RLC ARQ above the MAC.
  • Uplink HARQ mode B. The UE does not have to wait out the RTT before a mode-B process can carry new data.
  • Stretched timing. PDSCH-to-HARQ-ACK timing becomes K1 + K_offset, and the DRX HARQ RTT timers add the UE-gNB RTT for feedback-enabled processes.
  • Silicon cost. Going from 16 to 32 processes can double the worst-case HARQ soft-buffer storage in the UE modem unless LBRM or the UE’s declared capability limits it. That is a direct SRAM and area cost for the receiver IC.
  • IoT-NTN. NB-IoT has only 2 HARQ processes, so feedback disabling (Rel-18) matters even more there.

13.3 The NR transmit chain, from bits to DAC

ShareTechnote’s 5G/NR PDSCH page has an excellent diagram of the 12-step PDSCH transport process, with the 3GPP clause for each step. As it points out, steps 1–6 (TS 38.212) work on bits and steps 7–12 (TS 38.211) work on symbols and resources, so knowing which half a problem is in narrows it down quickly. The table below follows the same steps and extends the chain past resource mapping to the DAC, marking where HARQ and NTN come in.

# Step Spec What it does HARQ / NTN relevance
1 TB CRC attachment 38.212 §7.2.1 24-bit CRC if TB size A > 3824, else 16-bit TB CRC decides the final ACK/NACK
2 LDPC base graph selection 38.212 §7.2.2 BG2 if A ≤ 292, or A ≤ 3824 and R ≤ 0.67, or R ≤ 0.25; otherwise BG1 Fixed per TB, so identical on every retransmission
3 Code block segmentation + CB CRC 38.212 §7.2.3 (§5.2.2) Split into code blocks of at most 8448 (BG1) or 3840 (BG2) bits, 24-bit CRC each Code blocks are grouped into CBGs for partial retransmission
4 LDPC encoding 38.212 §7.2.4 QC-LDPC, lifting size up to Z = 384 Produces the mother codeword the circular buffer holds
5 Rate matching 38.212 §7.2.5 Circular buffer, bit selection from the RV start point, bit interleaving This is where HARQ lives: the RV picks which bits go out
6 Code block concatenation 38.212 §7.2.6 Rate-matched blocks joined into one codeword —
7 Scrambling 38.211 §7.3.1.1 Gold-sequence XOR, cinit from RNTI and scrambling ID —
8 Modulation 38.211 §7.3.1.2 QPSK to 256QAM (1024QAM in FR1 DL) MCS may change between retransmissions
9 Layer mapping 38.211 §7.3.1.3 Up to 4 layers per codeword, 8 in total Single-layer is typical for NTN handhelds
10 Antenna port mapping 38.211 §7.3.1.4 Layers to DMRS antenna ports (precoding transparent to the UE) —
11 Mapping to VRBs 38.211 §7.3.1.5 Fill REs, skipping DMRS, CSI-RS, PTRS and reserved REs Extra DMRS against residual Doppler reduces data REs
12 VRB → PRB mapping 38.211 §7.3.1.6 Non-interleaved or interleaved (bundle size 2 or 4) —
13 OFDM baseband signal generation 38.211 §5.3.1 IFFT and cyclic prefix insertion per numerology —
14 Digital front end and DAC Implementation (up-conversion in §5.4) Windowing/filtering for the spectral mask, crest-factor reduction, digital up-conversion, DPD, DAC NTN uplink: GNSS/ephemeris Doppler pre-compensation (NCO shift) and timing-advance adjustment are applied here

Receive side, ADC to ACK. ADC → digital down-conversion, decimation and AGC → (NTN) frequency and timing correction using ephemeris aiding → CP removal and FFT → DMRS channel estimation and PTRS phase tracking → equalisation / MIMO detection → soft demapping to LLRs → descrambling → de-rate-matching and HARQ soft combining with the stored buffer for that process → LDPC decoding → code-block and TB CRC checks → ACK/NACK.

13.4 The DVB-S2/S2X equivalent

DVB-S2 has a direct counterpart: Figure 1, “Functional block diagram of the DVB-S2 System”, in ETSI EN 302 307(-1), which draws the transmitter from input streams through to the RF signal. DVB-S2X (EN 302 307-2) reuses this chain and adds blocks to it. On the receiver side, the DVB-S2 implementation guidelines (DVB A171-1 / ETSI TR 102 376-1) include a matching receiver functional block diagram, and the ESA RC64 DVB-S2 SDR modem paper reproduces both transmitter and receiver chains.

Stage Spec (EN 302 307-1) Blocks Notes
Mode adaptation §5.1 Input interface, input stream synchronisation (optional), null-packet deletion (ACM with TS input), CRC-8 per user packet, merger/slicer for multiple streams, 80-bit BBHEADER insertion Application-dependent; GSE input carries IP
Stream adaptation §5.2 Padding to fill the BBFRAME to Kbch, BB scrambling Frame size is fixed by the MODCOD, so padding (not rate matching) absorbs the mismatch
FEC encoding §5.3 BCH outer code, LDPC inner code (64800 or 16200 bits), bit interleaving for 8PSK and higher orders One fixed codeword per frame; no RVs, no retransmission
Bit mapping §5.4 QPSK, 8PSK, 16/32APSK (S2X up to 256APSK, plus π/2-BPSK VL-SNR) Ring ratios per code rate
PL framing §5.5 Dummy PLFRAME insertion, PLHEADER (SOF + PLSCODE), optional pilot insertion, PL scrambling The PLHEADER makes every frame self-describing
BB shaping and quadrature modulation §5.6 Root-raised-cosine filter (α = 0.35/0.25/0.20; S2X adds 0.15/0.10/0.05), then to the DAC and RF No IFFT; DPD per APSK ring at the gateway
S2X additions EN 302 307-2 VL-SNR header, superframing (Annex E), beam hopping, channel bonding Superframes wrap the PLFRAME stream

Receive side, ADC to stream. ADC → decimation and AGC → coarse frequency recovery → matched filtering and symbol timing recovery → SOF/PLHEADER detection and frame synchronisation → PL descrambling → pilot-aided fine frequency and phase recovery → optional equalisation → soft demapping to LLRs → deinterleaving → iterative LDPC decoding → BCH decoding → BB descrambling → BBHEADER parsing and mode adaptation removal → TS or GSE output.

13.5 The two chains side by side

Aspect NR PDSCH DVB-S2/S2X
Retransmission HARQ with incremental redundancy (four RVs), soft combining None; QEF target with one codeword per frame
Fitting data to resources Rate matching from a circular buffer to any allocation Fixed FECFRAME per MODCOD; padding fills the BBFRAME
Error detection TB CRC + per-code-block CRC BCH outer code, CRC-8 per user packet and on the BBHEADER
Scrambling UE-specific (RNTI-based) Gold sequence BB scrambler plus PL scrambling with selectable Gold codes
Format signalling DCI on PDCCH, decoded before the data PLHEADER on every frame (self-describing)
Waveform generation IFFT + cyclic prefix RRC pulse shaping of a single carrier
Receiver memory Soft buffer per HARQ process held across the RTT (up to 32 processes in NTN) Single-frame LDPC buffer; nothing held across frames

For receiver design this is the practical difference. An NR-NTN modem must decode control before data and keep HARQ state and soft buffers across hundreds of milliseconds. A DVB-S2X demodulator can lock blindly onto any frame and throw the soft data away as soon as the codeword is decoded.

Hardware and system design recommendations

  1. ADC and dynamic range. Wideband S2X receivers (≥500 Mbaud, 256APSK) need about 7–8 ENOB at 1–2 GS/s; keep the jitter-limited SNR, −20·log10(2π·f_in·σ_j), at least 6–10 dB above the Es/N0 target. NR-NTN handsets need back-off headroom for OFDM crest factor (10–12 dB) and blockers from overlapping terrestrial bands, which pushes AGC range and filtering rather than sample rate.
  2. PA linearity. Design NR handset PAs for DFT-s-OFDM/π/2-BPSK efficiency at 23 dBm and plan for higher power classes. Ka NR VSAT array SSPAs will run 5–8 dB back-off for CP-OFDM, so prefer DFT-s-OFDM on the uplink and consider array-level DPD. For S2X gateways, use per-ring APSK predistortion and linearised TWTAs, and -L MODCODs for multi-carrier HPAs.
  3. Phase noise. For 256APSK, keep integrated phase error well below 1° rms post-loop, specify LO masks against Annex H.8, and run pilots on. For Ka NR at 120 kHz SCS with PTRS, reuse terrestrial FR2 synthesiser designs but control LO noise around 10–100 kHz offset.
  4. Doppler compensation. NR-NTN terminals need GNSS (or a Rel-20 GNSS-resilient alternative), an ephemeris propagator, and a fine-resolution TX NCO and timing adjuster that holds ±0.1 ppm and follows up to ~0.27 ppm/s at 600 km. S2X receivers for LEO should add ephemeris-aided NCO pre-rotation, because narrow VL-SNR loops cannot follow LEO Doppler rates unaided.
  5. Phased-array beam steering. LEO terminals need continuous electronic steering plus satellite handover within tens of ms; align NR array handover with CHO/RACH-less procedures and SIB19 neighbour ephemeris. Use true-time-delay or subarray TTD for 400 MHz Ka carriers to avoid beam squint at wide scan angles.
  6. System choice. For greenfield GEO/HTS broadband or broadcast, stay with DVB-S2X (Annex E formats for beam hopping or precoding) plus DVB-RCS2. For D2D, IoT, or services that must appear as 5G access in an MNO core, build on NR-NTN or IoT-NTN, and track Rel-19 Ku-band, regenerative payloads and Rel-20 GNSS resilience before freezing silicon. A dual-mode modem (S2X forward demodulator plus NR-NTN baseband) is a credible architecture for multi-orbit VSATs.

Caveats

  • Rel-19 ASN.1 and performance corrections continue; Rel-20 NTN items are still being specified (Stage-3 March 2027, ASN.1 freeze June 2027).
  • RRC value ranges (cellSpecificKoffset, kmac, ta-Common and drift, ntn-UlSyncValidityDuration) were taken from secondary reproductions of TS 38.331, and sources disagree on the ta-CommonDrift range; check the spec version you target. The NTN timing-error table in TS 38.133 §7.1C could not be verified.
  • PAPR values, overhead percentages, receiver acquisition ranges, ACM margins and phase-noise targets are typical engineering figures, not normative. HARQ process counts in section 13.2 are first-order estimates that ignore gNB/UE processing time beyond a few slots.
  • AST SpaceMobile throughput figures are vendor design targets; Starlink D2C per-beam capacity is an external measurement estimate.

Sources

Comments

Leave a Reply

More posts

Discover more from Vivek Roy

Subscribe now to keep reading and get access to the full archive.

Continue reading