PaperlayerView exact revision on arXiv

Source-linked AI summary

Resource Slicing for Time-Sensitive Data Traffic in Non-Terrestrial Networks with FSO ISLs

Abid Afridi, Alexis A. Dowhuszko, Risto Wichman, Jyri Hämäläinen

arXiv:2610.31802v1eess.SP

TL;DR

Queueing across multi-hop LEO satellite routes makes upper-tail E2E deadlines difficult to guarantee. The paper coordinates per-flow time-frequency reservations across the route, aligning consecutive links for pipelined forwarding. At λ = 0.33, the proposed scheme requires about 35 ms, with delay reductions of approximately 88%, 96%, and 97% relative to FIFO, FSF, and Pure-TDMA, respectively.

  • Problem

    Queueing delay accumulates across LEO satellite hops, while time-sensitive services require E2E latency and reliability guarantees that average delay alone cannot establish.

    Source: §I · p. 1

  • Method

    The framework dimensions flow-specific RB reservations from traffic arrivals, E2E delay budget, and reliability target, and temporally aligns reservations across successive links.

    Source: §I · p. 2, C. Reservation Model · p. 3, §III · p. 4

  • Results

    At λ = 0.33, the proposed scheme requires approximately 35 ms, reducing delay by approximately 88%, 96%, and 97% relative to FIFO, FSF, and Pure-TDMA, respectively.

    Source: §IV · p. 5

  • Takeaways & Limitations

    Simulations show substantially better performance than Pure-TDMA, FIFO, and FSF under high traffic loads, while remaining competitive under light load.

    Source: §V · p. 6

Abstract

from arXiv · show

In multi-hop Low Earth Orbit (LEO) satellite networks, time-sensitive packets experience queueing across multiple satellite hops, causing per-hop waiting times that accumulate and make the maximum end-to-end (E2E) delay difficult to guarantee. This paper presents a novel multi-hop slicing framework that reserves time-frequency resources for each time-sensitive flow of data originating at each satellite, aiming to maintain the reservation of communication resources along every inter-satellite link (ISL) on its route to the ground station (GS). The number of time-frequency resource blocks (RBs) reserved per-hop is determined from the data packet arrival statistics, E2E delay budget, and target level of reliability. Reservations on consecutive links are temporally aligned to enable pipelined forwarding of the time-sensitive data flows aggregated from terrestrial user terminals on the different LEO satellites. We evaluate the proposed framework against conventional scheduling and resource-allocation baselines, including pure time-division multiple access (Pure-TDMA), first-in-first-out (FIFO), and farthest-satellite-first (FSF), over the same multi-hop LEO satellite path. Under light traffic conditions, the baseline schemes provide comparable delay performance; however, their high-percentile delays increase sharply as data traffic load grows, whereas the proposed framework exhibits substantially slower delay growth. At the high arrival rate and high reliability target considered, the proposed scheme requires about 31 ms, a factor of 2.5-6.8 lower than the 76, 180, and 210 ms required by FIFO, FSF, and Pure-TDMA.

I. INTRODUCTION

Multi-hop LEO forwarding queues make upper-tail E2E delay central to latency and reliability guarantees. Prior approaches reduce average delay or allocate resources hop by hop, while this paper coordinates flow-specific reservations across the full route.

I. INTRODUCTION

  • Queueing at intermediate ISLs accumulates across hops and can dominate the latency budget, especially under heavy traffic.LEO satellites cannot all maintain continuous ground-station visibility, requiring ISLs for forwarding.Source: §I · p. 1
  • Upper-tail E2E latency determines whether packets meet maximum delay constraints, making average or median latency insufficient for time-sensitive services.The introduction links stringent latency and reliability to mission- and safety-critical applications.Source: §I · p. 1
  • Prior routing and traffic-control methods optimize aggregate delay or throughput, but the cited approaches do not provide per-flow probabilistic E2E latency guarantees –.The work in and targets aggregate E2E delay expectations, while lacks a probabilistic E2E guarantee.Source: §I · pp. 1–2
  • Existing slicing work allocates resources at individual links or network-resource levels rather than pre-allocating flow-specific resources across successive satellite hops.The cited single-access-hop reservation approach for Industry 4.0 does not extend across intermediate hops.Source: §I · p. 2
  • The framework reserves flow-specific time-frequency RBs across every forwarding-path link, dimensioned by arrival rate, E2E delay budget, and reliability target, with consecutive hops aligned for pipelining.It covers FSO ISLs and the RF feeder link, unlike independent hop-by-hop allocation.Source: §I · p. 2
  • The paper is organized into a system model and problem formulation, a proposed slicing scheme, simulation results, and a conclusion.The sections are II through V, respectively.Source: §I · p. 2

II. SYSTEM MODEL AND PROBLEM FORMULATION

The model follows traffic aggregated at each LEO satellite over a route of FSO inter-satellite links and a final RF feeder link to the ground station. It specifies synchronized time-frequency resources and independent Poisson packet arrivals.

A. Network and Traffic Model

  • Each satellite aggregates time-sensitive traffic from terrestrial terminals in its coverage area and forwards it along a path to the ground station.The path comprises satellites indexed by k and destination ground station g.Source: A. Network and Traffic Model · p. 2
  • Fig. 1 identifies f as the source satellite of a flow and k as the link's satellite index during forwarding toward the ground station.The depicted path delivers aggregated traffic from satellite K to the ground station via an RF feeder link.Source: Figure 1 · p. 2
  • The physical path has K links: ℓk is the FSO ISL from satellite k to k + 1, while ℓK is the RF feeder link to the ground station.A flow originating at satellite k traverses Lk = {ℓk, …, ℓK}, or K − k + 1 links.Source: A. Network and Traffic Model · p. 2
  • Each ISL and feeder-link frame is divided into equal-duration synchronized slots, with NRB frequency-domain RBs per slot.Data units have size D bits and, by assumption, fit in one time-frequency RB.Source: A. Network and Traffic Model · p. 2
  • Traffic generated by terrestrial terminals at each satellite is modeled as an independent Poisson arrival process with mean rate λk.The model treats the arrival processes at different satellites as independent.Source: A. Network and Traffic Model · p. 2
  • Packets forwarded during slot i become available at the next satellite at the beginning of slot i + 1.The traffic load grows along the route because each satellite adds local traffic to packets relayed from upstream satellites.Source: A. Network and Traffic Model · p. 2

B. Delay Model

The delay model separates queueing delay from fixed per-link delay and imposes a probabilistic end-to-end deadline requirement. It assumes sufficient buffering to neglect packet loss from queue overflow.

B. Delay Model

  • The model assumes enough buffer capacity that packet loss caused by queue overflow can be neglected.This is a modeling assumption, not a result about finite-buffer operation.Source: B. Delay Model · p. 2
  • In the illustrated three-satellite scenario, each satellite's traffic follows a Poisson arrival process, and the path has two FSO ISLs followed by one RF feeder link.The links are labeled ℓ1, ℓ2, and ℓ3.Source: Figure 2 · p. 3
  • Each packet's delay comprises queueing delay before transmission begins and fixed delay from processing, transmission, and propagation along the route.The fixed component depends on route geometry.Source: B. Delay Model · p. 3
  • Processing delay is onboard packet processing time, transmission delay is D/B for a D-bit packet over a link of rate B, and propagation delay is Lℓ/c.The three components are assumed to be fixed per hop.Source: B. Delay Model · p. 3
  • The latency-reliability requirement specifies a maximum E2E delay τmax and tolerated deadline-violation probability ϵ, giving required reliability 1 − ϵ.Reliability is interpreted as the probability that a packet reaches the destination within its deadline.Source: B. Delay Model · p. 3

C. Reservation Model

Reservations are evaluated over synchronized slot windows and must meet a per-flow minimum service budget on each link. The actual allocation spreads reserved RBs over the time-frequency grid subject to per-slot capacity.

C. Reservation Model

  • Each link uses a Wk-slot consecutive reservation window within a TH-slot synchronized allocation horizon, and each flow's reserved RB count is evaluated within that window.The window start is restricted so the complete window remains inside the allocation horizon.Source: C. Reservation Model · p. 3
  • For every window, the actual reservation for flow f on link ℓk must meet its minimum RB service budget, which is selected to satisfy the target E2E latency reliability.The allocation is distributed over a two-dimensional time-frequency grid while respecting per-slot RB capacity.Source: C. Reservation Model · p. 3

D. Problem Formulation

The formulation minimizes reserved RBs while satisfying end-to-end latency and reliability requirements, per-flow reservation windows, and per-link capacity limits.

D. Problem Formulation

  • The objective minimizes total RBs over the reservation window while meeting end-to-end latency and reliability requirements.The formulation is for time-sensitive data traffic.Source: D. Problem Formulation · p. 3
  • Constraint (9a) represents the end-to-end latency and reliability requirements.Source: D. Problem Formulation · p. 3
  • Constraint (9b) requires each flow to receive its allocated RBs on link ℓk within every Wk-slot reservation window.Source: D. Problem Formulation · pp. 3–4
  • Fig. 3 arranges allocation strategies by row and forwarding links ℓ1, ℓ2, and ℓ3 by column.Source: Figure 3 · p. 4
  • Constraint (9c) limits each physical link’s total per-slot reservation to the available NRB.Source: D. Problem Formulation · p. 4

III. PROPOSED 2D SLICING

The proposed 2D slicing assigns flow-specific reservations along each flow’s route and aligns consecutive-link service opportunities. Its reservation process uses traffic and path inputs, while the evaluation compares it with shared-resource and equal-share baselines.

III. PROPOSED 2D SLICING

  • Flow-specific 2D slices are pre-allocated along each flow’s forwarding path to provide isolation while meeting its end-to-end targets.Packets wait for the next reserved opportunity when no reserved RB is available; unused reserved RBs under light traffic are the resource overhead for maintaining end-to-end performance.Source: §III · p. 4
  • Unlike FIFO and FSF, the method assigns reserved RBs to individual time-sensitive flows rather than placing all flows in a common resource pool.On shared links, it adds a small reservation margin for flows originating farther from the GS to reduce or eliminate their queueing delay.Source: §III · p. 4
  • Reservations on consecutive links are aligned with propagation and processing delays so packets reach the next queue near their reserved service opportunity.This coordination enables pipelined forwarding.Source: §III · p. 4
  • The algorithm initializes reservation windows and minimum reservations from λ, |Lk|, τmax, and ϵ, then reserves RBs subject to constraints (9b)–(9c).Source: §III · p. 4
  • The algorithm generates local arrivals, computes per-flow arrivals and service, updates queues, and repeats until all flows satisfy (6).Source: §III · p. 4
  • The baseline comparison assumes equal arrival rates λk = λ, making the aggregate load on link ℓk equal to kλ.Source: §III · p. 4
  • Pure-TDMA divides each link’s available RBs equally among its flows, giving each flow about Ck/k RBs.Source: §III · p. 4

P TDMA

Pure-TDMA’s fixed per-flow shares can accumulate queueing delay for flows traversing more hops, while unused shares cannot be reassigned to other flows.

P TDMA

  • Fixed shares cause flows originating farther from the GS to accumulate queueing delay across the forwarding path.Source: P TDMA · p. 4
  • RBs left unused by one flow cannot be used by another, so reserved capacity may be wasted under light traffic.Source: P TDMA · p. 4
  • FIFO places flows in one shared queue and serves them in arrival order, so overload caused by one flow affects every flow on the link.Source: P TDMA · p. 4

P FIFO

As traffic accumulates along the path, the feeder link carries the greatest aggregate load and saturates first; its shared queue then grows rapidly under high load.

P FIFO

  • The feeder link ℓK carries the largest aggregate amount of data as offered load grows along the path, making it the first link to saturate.Source: P FIFO · p. 4
  • Under high load, rapid growth of the feeder link’s shared queue causes the long pooled tail delay.Source: P FIFO · p. 4
  • FSF serves flows in strict priority order, placing Sat-1 first and the newly entering flow last in the considered path.Source: P FIFO · p. 4

P FSF

The section describes a Poisson-CDF expression for k = 1 and compares required E2E delay across reliability targets, traffic loads, and scheduling schemes. Under strict priority, the lowest-priority flow faces aggregate traffic and can be starved when the link is busy.

P FSF

  • For k = 1, out,k(Wk) is written as one minus the Poisson CDF evaluated at Ck(Wk) with mean λWk.Source: P FSF · p. 4
  • Fig. 4 compares required E2E delay across reliability targets for Pure-TDMA, FIFO, FSF, and 2D slicing at three offered loads.Source: Figure 4 · p. 5
  • Strict priority leaves the lowest-priority flow facing aggregate traffic, so it is starved whenever the link is busy.In the three-satellite path, that flow is Sat-3.Source: P FSF · p. 5

IV. SIMULATION RESULTS

The simulations compare E2E delay across loads, arrival rates, and source satellites. At high load, 2D slicing has the lowest reported delay and limits feeder-link waiting, while its tail delays remain similar across the three source paths.

IV. SIMULATION RESULTS

  • The simulation models a 550-km LEO path with three satellites and one ground station, using 10^4 realizations and 3 × 10^3 warm-up slots per configuration.Source: §IV · p. 5
  • At high load, 2D slicing needs about 31 ms at 99.9999%, versus 76 ms FIFO, 180 ms FSF, and 210 ms Pure-TDMA.At medium load, it needs about 20 ms, versus 24 ms for FIFO and FSF and 49 ms for Pure-TDMA at 99.9999%.Source: §IV · p. 5
  • At λ = 0.33, 35 ms corresponds to delay reductions of about 88%, 96%, and 97% versus FIFO, FSF, and Pure-TDMA.Source: §IV · p. 5
  • Proposed reservations limit feeder-link waiting to about 3, 7, and 15 ms, keeping total delays within 19–22 ms; the feeder-link queue is the path's dominant bottleneck.Source: §IV · pp. 5–6, Figure 6 · p. 6
  • The proposed scheme confines the three source flows' tail delays to about 25–33 ms and reduces Sat-3's tail by about 52%, 83%, and 84% versus FIFO, FSF, and Pure-TDMA.Source: §IV · p. 6, Figure 7 · p. 6

V. CONCLUSION AND FUTURE WORK

The paper concludes that coordinating time-frequency reservations across a multi-hop LEO path limits queueing delay and supports reliable E2E latency. Its simulations show better performance at high loads, while future work will extend the framework to additional network conditions and adaptive allocation.

V. CONCLUSION AND FUTURE WORK

  • The paper presents an E2E resource-slicing framework for time-sensitive traffic in multi-hop LEO satellite networks.Source: §V · p. 6
  • Coordinated time-frequency reservations across the forwarding path limit queueing delay and provide reliable E2E latency performance.Simulations show substantial gains over Pure-TDMA, FIFO, and FSF at high loads, with competitive performance at light load.Source: §V · p. 6
  • Future work will consider time-varying topology, link-dependent capacities, mixed traffic classes, and DRL-based adaptive multi-hop resource allocation.Source: §V · p. 6
Loading 2610.31802v1…