PaperlayerView exact revision on arXiv

Source-linked AI summary

5G satellite networks for IoT: offloading and backhauling

Beatriz Soret, Israel Leyva-Mayorga, Stefano Cioni, Petar Popovski

arXiv:2011.05202v1cs.NI

TL;DR

This paper examines how LEO constellations could support IoT by offloading congested terrestrial networks or backhauling remote gNBs to the 5G core. It models uplink access and multi-hop satellite paths, evaluating collisions, delay, and Age of Information. Its findings describe trade-offs in access attempts and the potential of satellite paths for these use cases.

  • Problem

    The paper examines how LEO constellations can support IoT traffic through offloading or remote-site backhaul, including access and multi-hop path performance.

    Source: §I · pp. 2–3

  • Method

    The paper models uplink random access and a multi-hop satellite queueing path, analyzing collisions, delay, and Age of Information across offloading and backhauling scenarios.

    Source: §I · p. 3, A. Analysis · p. 17

  • Results

    The authors report that satellite access can alleviate congestion and improve latency in selected offloading cases, while additional RA attempts raise congestion and latency variance.

    Source: B. Performance evaluation · p. 16, §VI · p. 23

  • Takeaways & Limitations

    The access, delay, and AoI analyses provide insights for assessing LEO constellations for IoT offloading and backhaul topology design.

    Source: §VI · pp. 23–24

    The reported latency results quantify the access network; end-to-end latency differences involving the core and satellite or terrestrial paths are left for future work.

    Source: B. Performance evaluation · p. 16

Abstract

from arXiv · show

One of the main drivers of 5G cellular networks is provision of connectivity service for various Internet of Things (IoT) devices. Considering the potential volume of IoT devices at a global scale, the next leap is to integrate Non-Terrestrial Networks (NTN) into 5G terrestrial systems, thereby extend the coverage and complement the terrestrial service. This paper focuses on the use of Low-Earth Orbit (LEO) satellite constellations for two specific purposes: offloading and backhauling. The former allows offloading IoT traffic from a congested terrestrial network, usually in a very dense area. In the latter application, the constellation provides a multi-hop backhaul that connects a remote terrestrial gNB to the 5G core network. After providing an overview of the status of the 3GPP standardization process, we model and analyze the user data performance, specifically in the uplink access and the satellite multi-hop constellation path. The evaluation of the collisions, the delay and the Age of Information, and the comparison of the terrestrial and the satellite access networks provide useful insights to understand the potential of LEO constellations for offloading and backhauling of IoT traffic.

I. INTRODUCTION

LEO constellations could extend 5G coverage and support IoT through satellite offloading in congested areas or multi-hop backhaul for remote gNBs. The paper examines these uses and their access and constellation-path performance.

A. LEO satellite constellations

  • LEO satellites have small ground coverage and move quickly relative to Earth, requiring a densely populated constellation for continuous coverage.Satellites in an orbital plane follow the same trajectory; multi-hop routes can use ISLs, while cross-plane links face time-varying distance and relative-speed challenges.Source: §I · pp. 1–2

I. INTRODUCTION

  • NTN IoT must account for the distinct demands of mMTC and URLLC, while the 1 ms URLLC target is infeasible over satellite and LEO RTT latency is redefined to 50 ms.Reliability and connection density remain unchanged in 5G NR Release 15.Source: §I · p. 2
  • For update-driven IoT, Age of Information captures receiver interest in the freshness of updates rather than per-packet delay.Examples include logistics tracking, vessel data, aircraft broadcasts, and sensor-network status updates.Source: §I · pp. 2–3
  • The paper evaluates two LEO use cases: offloading IoT traffic from congested terrestrial networks and multi-hop backhaul from remote gNBs to the 5G core.Offloading considers access diversity and adapted channel parameters; backhauling examines access delay and ISL routing between distant points.Source: §I · p. 3
  • The paper is organized around 3GPP NTN status, the system model, performance evaluations of offloading and backhauling, and conclusions.The two scenarios examine whether LEO constellations can support IoT traffic alongside terrestrial networks.Source: §I · p. 3

B. 3GPP NTN

3GPP's NTN work covers satellite and airborne platforms and considers service continuity, ubiquity, and scalability. The NTN IoT study focuses on adapting radio procedures and beam management to satellite conditions.

B. 3GPP NTN

  • 3GPP's NTN scope includes LEO, MEO, and GEO satellites and airborne platforms such as HAPs, typically operating at 8–50 km.Source: B. 3GPP NTN · p. 4
  • NTN use cases are service continuity for previously connected mobile terminals, service ubiquity where terrestrial coverage is absent or interrupted, and service scalability for multicasting or IoT.Source: B. 3GPP NTN · p. 4
  • The NTN IoT feasibility study considers NB-IoT and eMTC support from space, including adaptations to delay-sensitive random access, HARQ, RRC, beam management, and cell selection or reselection.The first Release 17 work-item baseline described here is adapting eMBB services for satellite networks.Source: B. 3GPP NTN · p. 5
  • Regenerative payloads can act as functional gNBs and use the Xn interface to connect distant gNBs; NTN multi-connectivity may link a UE to one satellite and one terrestrial network or to two satellites.Connectivity combining may apply to uplink, downlink, or both.Source: B. 3GPP NTN · p. 5

C. 5G Core Network and Initial Access Procedure

This section outlines the 5G core and initial access, then describes NB-IoT random access and resource constraints relevant to terrestrial and satellite access. The paper uses this access procedure as a basis for its evaluations.

C. 5G Core Network and Initial Access Procedure

  • The 5G core separates control and user planes; the UPF handles packet processing, while N3 and N6 link it to the gNB and packet data network.The AMF handles connection and mobility management; the SMF handles session management.Source: C. 5G Core Network and Initial Access… · p. 5
  • The evaluations use NB-IoT random access, assuming a similar procedure will be defined for 5G, as devices must complete RA to move from RRC idle to connected mode.Source: C. 5G Core Network and Initial Access… · p. 6
  • Standard RA exchanges a preamble, uplink grant, connection request, and contention resolution; uplink or downlink data transmission follows completion of the procedure.A two-step version under discussion pairs a preamble with PUSCH data transmission and uses contention resolution as an acknowledgment.Source: C. 5G Core Network and Initial Access… · p. 6
  • NB-IoT preambles use orthogonal single-tone hopping sequences, with the available preamble count matching the NPRACH subcarrier count.Preamble format 0 lasts 5.6 ms per repetition; repetitions improve SINR and reduce erasure probability but increase message duration and energy use.Source: C. 5G Core Network and Initial Access… · pp. 6–7
  • NPRACH and NPDCCH scheduling consumes resources otherwise available for data, so implementations must balance resources across NPRACH, NPDCCH, PUSCH, and PDSCH.Dense cells may allocate little to NPRACH and NPDCCH, increasing the interval between random access opportunities.Source: C. 5G Core Network and Initial Access… · pp. 7–8

D. Offload and backhaul architectures

The paper considers a regenerative LEO constellation for offloading congested terrestrial networks and backhauling remote gNB traffic. These architectures address different connectivity needs, and the paper excludes multilink backhauling.

D. Offload and backhaul architectures

  • A regenerative LEO constellation with inter-satellite links is modeled for two functions: terrestrial-network offloading and backhaul to the 5G core.Source: D. Offload and backhaul architectures · p. 8
  • Offloading lets dual-connected devices choose between terrestrial and satellite gNBs when terrestrial networks are congested, with policies accounting for coverage, RTT, and channel differences.Satellite traffic may reach the 5G core directly or traverse other satellites.Source: D. Offload and backhaul architectures · p. 8
  • Satellite backhaul is presented as a feasible option for remote terrestrial access without a cost-effective terrestrial route to the 5G core.The passage notes that dense LEO paths may have latency similar to or lower than terrestrial links over large endpoint distances.Source: D. Offload and backhaul architectures · p. 9
  • Multilink backhauling that combines satellite and terrestrial links is under discussion in 3GPP but is not considered in this paper.Source: D. Offload and backhaul architectures · p. 9

III. SYSTEM MODEL

The system model treats offloading and backhauling as uplink scenarios with IoT updates directed to terrestrial and/or LEO space gNBs. Random-access capacity, collisions, and wireless errors constrain how much update traffic is relayed toward the 5G core.

III. SYSTEM MODEL

  • A fraction κ of IoT nodes access the terrestrial gNB, while the remaining 1−κ access the space gNB.The model defines separate failure probabilities and relayed rates for the two paths.Source: §III · p. 10
  • Only a fraction of the total update rate is relayed by the terrestrial and space gNBs because access capacity, collisions, and wireless errors limit delivery.The failure probabilities depend on NPRACH and NPDCCH capacity and on update rates toward each gNB.Source: §III · p. 10
  • Random-access capacity depends on the interval between RAOs, the available preambles per RAO, and the response-window length.The access model uses NB-IoT random-access parameters, including a multichannel slotted-ALOHA preamble process.Source: §III · pp. 10–11

A. Analysis

The analysis models random access, collisions, retries, and access delay for terrestrial and satellite paths. It identifies an access-stability condition and accounts for satellite propagation and Doppler assumptions.

A. Analysis

  • Successful access attempts are maximized at x = R, and the system is stable when E[X] ≤ R/e expected attempts per RAO.The maximum throughput corresponds to R slotted-ALOHA channels.Source: A. Analysis · p. 12
  • Preamble collisions occur when multiple UEs select the same preamble in one RAO; the model assumes the gNB detects them and withholds responses.The analysis defines the expected number of collided UEs and collision probability for x contenders.Source: A. Analysis · p. 12
  • The access model allows up to A preamble attempts with a fixed erasure probability; failed UEs back off and retry until success or the attempt limit.The fixed probability represents the assumed strategy of using relatively high power from the first attempt.Source: A. Analysis · pp. 12–13
  • Access delay includes preamble and response timing, processing, possible backoff, and extra queuing during the RA response window and HARQ processes.The minimum delay assumes one attempt, an early RA response, and no Msg3 or Msg4 transmission failures.Source: A. Analysis · p. 13
  • The satellite path adds a 2 ms preamble prefix and repetitions for comparable erasure probability; Doppler compensation is assumed at the space gNB.Designing mechanisms for NB-IoT connectivity under satellite Doppler is outside the paper’s scope.Source: A. Analysis · p. 14

B. Performance evaluation

The evaluation compares terrestrial-only access with split terrestrial-satellite traffic and examines access latency and random-access outcomes. A satellite path can relieve terrestrial congestion, while repeated attempts can increase NPRACH congestion; the reported latency is access-network latency, not end-to-end latency.

B. Performance evaluation

  • Adding a satellite link significantly alleviates terrestrial congestion and improves access latency for both A = 1 and A = 10 attempts.The lowest latency is achieved by the spatial link; the comparison includes terrestrial-only traffic (κ = 1) and an equal split (κ = 0.5).Source: B. Performance evaluation · pp. 14–15
  • With one attempt, successful access probability is upper bounded by 1−ϵ; more attempts reduce transmission-error failures but increase access-channel load.Figure 3 presents the probability mass function of total, collided, and successful uplink access for A = 1 and A = 10, for terrestrial and satellite paths.Source: B. Performance evaluation · p. 14, Figure 3 · p. 16
  • With A = 1, access performance is relatively stable, whereas allowing 10 attempts increases latency variance and is especially detrimental at κ = 1.In that case, added attempts increase NPRACH congestion and success probability reaches around 0.16.Source: B. Performance evaluation · p. 16
  • The reported latency results cover the access network; end-to-end latency also depends on core-network architecture and terrestrial and satellite path performance.Those end-to-end differences are left for future work.Source: B. Performance evaluation · p. 16

A. Analysis

The analysis treats backhaul as terrestrial random access feeding a multi-hop satellite queue, then models delay and Age of Information under specified service, arrival, and error assumptions.

A. Analysis

  • Backhaul analysis separates UE-to-gNB random access from the multi-hop satellite segment; RA uses the Section IV analysis with κ=1 and an unshared gNB.This subsection therefore analyzes latency and AoI in the space segment.Source: A. Analysis · pp. 16–17
  • The space segment is a serial FCFS queueing network of buffered satellites, from the gNB-connected satellite 1 to satellite N and its feeder link to the 5GC.Each satellite buffer is modeled with infinite capacity; simulations found limited buffers have little impact unless the total buffer has only a few positions.Source: A. Analysis · p. 17, Figure 5 · p. 18
  • Random access determines the effective arrival rate λ at satellite 1; later servers represent ISLs, node N represents the feeder link, and ISLs have packet-erasure probabilities εs_n.A packet sent over an ISL is received by the next node with probability 1−εs_n.Source: A. Analysis · p. 17
  • With exponential service, FCFS, and Poisson arrivals, satellite service and waiting times are Hypoexponential; equal departure rates yield an Erlang total system time.The analysis also considers equal service rates at every node.Source: A. Analysis · p. 18
  • Because RA makes first-satellite arrivals non-Poisson, A=1 permits a Poisson approximation over a wide range of arrival rates, while large A invalidates that approximation.The analysis states that RA significantly contributes to total latency and AoI for A=10.Source: A. Analysis · pp. 18–19
  • AoI follows a sawtooth pattern, resetting when updates arrive; a lost packet increases age, represented by additional area in the error-aware analysis.The derivation accounts for extra trapezoid areas associated with errors.Source: A. Analysis · pp. 19–20
  • The total system time sums waiting and service across nodes, while waiting-time analysis considers whether the previous packet remains in the network when a new packet arrives.The general waiting-time expression applies when the new packet finds the last queue nonempty.Source: A. Analysis · p. 20
  • Figure 7 compares average system time and AoI across system load for an error-free backhaul, RA configurations A=1 and A=10, and satellite counts N={1, 2, 4, 6}.The no-RA case is modeled with Poisson arrivals at the first satellite.Source: Figure 7 · p. 21, A. Analysis · p. 21

B. Performance evaluations

The backhaul evaluation finds that random-access attempts affect delay and Age of Information differently across load, while satellite-link errors can reduce delay and high-load AoI.

B. Performance evaluations

  • The error-free comparison varies the number of space-segment satellites across 1, 2, 4, and 6 and compares random access with 1 or 10 attempts.It plots average system time and AoI against the average rate at the first satellite, with λ = ρ and μ_n = 1.Source: B. Performance evaluations · p. 21
  • Without random access or with one attempt, average delay rises with load; ten-attempt access yields much larger, U-shaped delay dominated by collisions at low load.Direct transmission without random access is the latency and AoI lower bound; one-attempt random access has minimal impact on total delay.Source: B. Performance evaluations · p. 22
  • Average AoI is U-shaped in all cases; ten-attempt access separates at low load but becomes negligible at high load, where queues grow near congestion.At high load, source updates arrive too frequently for system capacity, producing long queueing times.Source: B. Performance evaluations · p. 23
  • In error-prone backhaul, packet losses improve delay more as load rises and improve AoI only at high load; error-free links yield the lowest AoI at low load.Dropped packets are excluded from the delay and AoI calculations.Source: Figure 8 · p. 22, B. Performance evaluations · p. 23

VI. CONCLUSIONS

The paper evaluates LEO offloading and multi-hop backhauling for 5G IoT, including random-access trade-offs and backhaul delay and AoI. Its results inform offloading choices and backhaul topology, while full end-to-end evaluation and architectural extensions remain future work.

VI. CONCLUSIONS

  • Satellite access can alleviate congested terrestrial networks, and additional random-access attempts reduce transmission-error failures while increasing access-channel load.A single attempt is reasonable for some AoI applications that send frequent updates and prioritize freshness.Source: §VI · p. 23
  • Backhaul delay and AoI results, with and without errors, provide insights for designing the topology and choosing the number of hops.Source: §VI · p. 24
  • Future work includes multi-connectivity, multi-link backhaul, satellite mobility and control-plane issues, and terrestrial 5GC latency and throughput evaluation.The authors also identify implementation and experimental work as a future direction.Source: §VI · p. 24
Loading 2011.05202v1…