PaperlayerView exact revision on arXiv

Source-linked AI summary

Ultra-Reliable Low Latency Communication (URLLC) using Interface Diversity

Jimmy J. Nielsen, Rongkuan Liu, Petar Popovski

arXiv:1711.07771v1cs.ITmath.IT

TL;DR

URLLC calls for very high reliability alongside low latency, and achieving these targets with single technologies is difficult. The paper analyzes coded payload distribution across multiple interfaces and finds that interface diversity can reduce latency in practical settings, with optimized weighting sometimes outperforming simple k-out-of-n strategies.

  • Problem

    URLLC may require packet-delivery success rates from 5-nines to 9-nines and latency down to a few milliseconds, while very high reliability is difficult for a single wireless technology.

    Source: §I · p. 1

  • Method

    The paper combines traditional reliability models with technology-specific latency distributions to analyze coded payload allocation across interfaces, including correlated failures.

    Source: §I · pp. 1–2, 2) Splitting · p. 2, §VI · p. 11

  • Results

    Interface diversity can lower latency by up to around 40% in practical systems with large messages and low-rate technologies, and optimized weighting can provide low-latency reliability where simple k-out-of-n strategies cannot.

    Source: §VI · p. 11

  • Takeaways & Limitations

    Optimized weighted splitting can fine-tune the latency-reliability trade-off for a specific scenario, while simple k-out-of-n strategies do not account for interface characteristics.

    Source: A. Independent interfaces · p. 8, §VI · p. 11

    Brute-force optimization becomes computationally constrained by the number of interfaces and the chosen step size for payload weights.

    Source: E. Weighted splitting · p. 5

Abstract

from arXiv · show

An important ingredient of the future 5G systems will be Ultra-Reliable Low-Latency Communication (URLLC). A way to offer URLLC without intervention in the baseband/PHY layer design is to use interface diversity and integrate multiple communication interfaces, each interface based on a different technology. In this work, we propose to use coding to seamlessly distribute coded payload and redundancy data across multiple available communication interfaces. We formulate an optimization problem to find the payload allocation weights that maximize the reliability at specific target latency values. In order to estimate the performance in terms of latency and reliability of such an integrated communication system, we propose an analysis framework that combines traditional reliability models with technology-specific latency probability distributions. Our model is capable to account for failure correlation among interfaces/technologies. By considering different scenarios, we find that optimized strategies can in some cases significantly outperform strategies based on $k$-out-of-$n$ erasure codes, where the latter do not account for the characteristics of the different interfaces. The model has been validated through simulation and is supported by experimental results.

I. INTRODUCTION

The paper proposes using multiple communication interfaces to pursue URLLC’s stringent reliability and latency goals without relying on retransmissions. It develops an analysis framework and coded transmission strategies for trading off latency, reliability, and bandwidth across heterogeneous interfaces.

I. INTRODUCTION

  • Multipath protocols that rely on retransmission do not fit the strict latency requirements considered here, motivating interface diversity without retransmission.The paper identifies flexible latency-reliability trade-offs as an unaddressed feature in the multipath work it reviewed.Source: §I · p. 1
  • The contribution combines interface diversity with coded payload allocation to study reliability and latency gains across multiple interfaces.The paper presents this manuscript as an expanded treatment of the concept.Source: §I · p. 1
  • The framework combines reliability engineering with interface-specific latency-reliability characteristics to evaluate transmission strategies.The analysis also models correlated failures and infrastructure restoration dependencies.Source: §I · pp. 1–2

II. SYSTEM MODEL

  • The system model considers M2M devices communicating with an end host over wired and cellular interfaces with differing latency, loss, and failure correlation.Interface latency-reliability functions are assumed to come from end-to-end network monitoring measurements.Source: §II · p. 2

1) Cloning:

  • Cloning sends a full copy of each message over every available interface, increasing robustness at the cost of N-fold redundancy.The receiver needs only one copy to decode the message.Source: 1) Cloning · p. 2

2) Splitting:

Splitting sends coded message fractions over different interfaces, trading reliability and transmission time through the selected fractions. Its latency benefit depends on transmission time being substantial relative to access delay.

2) Splitting:

  • Splitting sends only fractions of the message on each interface, and can reduce latency when transfer time dominates access latency.Since splitting reduces transfer time but not access latency, small packets with access delay much greater than transfer time show no noticeable gain.Source: 2) Splitting · p. 2
  • Coded fragments can be generated with rateless or Reed Solomon codes; decoding succeeds with very high probability after receiving approximately 100(1 + ε)% of the original message size.The passage gives ε = 0.05 as a typical value and denotes the threshold γ_d = 1.05.Source: 2) Splitting · p. 2
  • In k-out-of-N splitting, the receiver needs at least k of n equally sized coded fragments, trading greater reliability for longer transmission time as redundancy increases.Smaller redundancy offers shorter transmission times but lower error protection.Source: 2) Splitting · p. 3, Figure 2 · p. 3

B. Achievable Latency Reduction

  • For two interfaces, the analysis considers latency reduction across different transmission-time combinations and access-latency settings.The analysis distinguishes equal access latencies from a case where interface 2 has twice interface 1’s access latency.Source: B. Achievable Latency Reduction · p. 3, Figure 3 · p. 3
  • Close to 50% latency reduction is possible when two interfaces have equal access latencies and equal transmission times 6–10 times larger than access time.The cited scenarios include GPRS, EDGE, UMTS, and HSDPA with payloads of 3000 bytes or more; unequal access latencies narrow the range of transmission times that yield latency reduction.Source: B. Achievable Latency Reduction · p. 3, Figure 3 · p. 3
  • Splitting reduces latency while sacrificing reliability, so the framework quantifies their interplay using latency probability distributions.The section introduces this trade-off as the basis for the subsequent analysis.Source: B. Achievable Latency Reduction · p. 3

C. Latency-reliability Function

The latency-reliability function captures the probability that an interface delivers a packet of a given size by a deadline. For independent interfaces, the framework evaluates system reliability by combining successful reception outcomes that allow decoding.

C. Latency-reliability Function

  • For interface i, F_i(x, B) is the probability of transmitting B bytes within latency x, and γ_i is its assigned payload fraction.The latency-reliability functions are assumed static; dynamically changing transmission strategies are left for future work.Source: C. Latency-reliability Function · p. 4

III. RELIABILITY OF TRANSMISSIONS OVER INDEPENDENT

  • The independent-interface methodology applies when interfaces do not share common error causes.Source: §III · p. 4

A. Evaluating reliability for weight assignment

  • System reliability is evaluated by summing the probabilities of successful reception outcomes that let the receiver decode the original message.The outcome enumeration uses a 2^N × N matrix whose entries indicate successful or failed reception on each interface.Source: A. Evaluating reliability for weight… · p. 4

B. Cloning

For independent interfaces, packet cloning sends a full copy over each interface, while k-out-of-N splitting assigns an equal payload fraction to each. The latter is optimal only for identical interfaces.

B. Cloning

  • For independent interfaces, cloning combines the interfaces’ latency-reliability functions using the traditional parallel-systems method.The method is presented as an alternative to the framework described earlier.Source: B. Cloning · p. 4
  • In cloning, each interface receives the full payload, so γ_i = 1 for every interface.Source: B. Cloning · p. 4
  • The k-out-of-N splitting strategy is optimal only for identical interfaces and is best suited to interfaces with comparable properties.Its latency-reliability function is evaluated with γ_i = 1/k for all interfaces.Source: C. k-out-of-N splitting · p. 4

D. Weighted splitting between two interfaces

The two-interface analysis formulates an analytic subproblem for choosing how to split coded payload between interfaces A and B to minimize expected latency. It models each interface’s latency as Gaussian and the completion time as the later fragment arrival.

D. Weighted splitting between two interfaces

  • The two-interface subproblem seeks the coded-payload split between A and B that minimizes latency.It is an analytical subproblem of the general weighted-splitting optimization problem.Source: D. Weighted splitting between two… · p. 4
  • With two Gaussian interface latencies, completion occurs when the last fragment arrives, so expected latency is E[max(X_A, X_B)].The model assumes σ_A and σ_B are constant and independent of μ_A and μ_B.Source: D. Weighted splitting between two… · p. 4
  • The analysis approximates the expected maximum of two normal random variables, then differentiates L with respect to γ to find a minimum.The stated optimality condition is dL/dγ = 0.Source: D. Weighted splitting between two… · pp. 4–5

E. Weighted splitting

Weighted splitting selects payload-allocation weights to optimize reliability across chosen latency targets, which can be assigned different importance. The numerical evaluation uses a tractable brute-force search because reliability changes nonlinearly with the weights.

E. Weighted splitting

  • The weighted-splitting problem chooses payload weights γ to optimize reliability at specified latency targets and importance weights.The targets and weights define which latency values are optimized and their relative importance.Source: E. Weighted splitting · p. 5
  • Brute-force search spans (1/δγ)^N combinations, making tractability depend on the number of interfaces N and the step size δγ.The numerical results use a sufficiently coarse step size to keep the search computationally tractable.Source: E. Weighted splitting · p. 5
  • Increasing an interface’s γ can lower its reliability, while combinations whose γ-values sum to γ_d can produce jumps in overall reliability.Fig. 4 illustrates equal increases in γ for scenario C at a 0.4 s latency requirement.Source: E. Weighted splitting · p. 5, Figure 4 · p. 5

IV. TRANSMISSIONS OVER INTERFACES WITH FAILURE

The framework combines CTMC failure-state probabilities with each state’s latency-reliability behavior to evaluate interface-diversity strategies, including correlated failures. In the three-interface case study, cloning combines available interfaces by the earliest-arriving packet.

IV. TRANSMISSIONS OVER INTERFACES WITH FAILURE

  • The CTMC weights each state’s latency-reliability function by its steady-state probability to obtain the combined system function.The model is intended for systems where common error causes correlate interface failures.Source: §IV · pp. 5–6

A. Case study: Correlated failures in three-interface system

  • The case study combines Wi-Fi to fiber with two cellular links and models independent link failures plus shared base-station failures that disable both cellular links.The model includes four high-level failure types and is presented as sufficient for evaluating strategies and exploring different failure-correlation probabilities.Source: A. Case study: Correlated failures in… · p. 6
  • Interface latency-reliability functions are normalized by availability, while the same CTMC failure model is used across strategies because equipment and transmission failures are assumed strategy-independent.The state-specific functions also depend on the transmission strategy and its payload-allocation weights.Source: A. Case study: Correlated failures in… · p. 6

B. Packet cloning on three interfaces

  • With cloning, the system latency CDF is the probability that at least one available interface delivers by x, computed as 1 − Π(1 − F_j).The interface functions are combined state by state to obtain H_s(x, B).Source: B. Packet cloning on three interfaces · p. 6

C. 2-of-3 packet splitting on three interfaces

For 2-of-3 splitting, the receiver must obtain at least two coded fragments, so the state-specific latency-reliability functions differ from cloning. The weighted strategy instead sends a full Wi-Fi copy and splits another copy across the cellular interfaces.

C. 2-of-3 packet splitting on three interfaces

  • The state-specific latency-reliability functions for the 2-of-3 strategy are assembled in Table II.The strategy sends coded packets corresponding to 1/2 · γ_d of the source packet on each interface.Source: C. 2-of-3 packet splitting on three… · p. 6
  • For 2-of-3 splitting, state 1 succeeds when all three fragments or any two arrive by x; its CDF sums the four corresponding cases.For identical interfaces, the state-1 expression reduces to the conventional 2-out-of-3 reliability formula; in states 2–4, the second fragment is last to arrive.Source: C. 2-of-3 packet splitting on three… · pp. 6–7

D. Weighted packet splitting on three interfaces

  • The weighted case sends a full message copy over Wi-Fi and splits another copy between the cellular interfaces to reduce latency.The paper applies the optimal γ-values derived for the two-interface case; its latency-reliability vector uses the last-arriving packet’s CDF.Source: D. Weighted packet splitting on three… · p. 7

V. NUMERICAL RESULTS

Across the independent-interface scenarios, weighted splitting achieves higher reliability or shorter latency than simpler k-out-of-N strategies at selected targets. The numerical evaluation uses assumed latency distributions and interface specifications informed in part by RTT measurements.

V. NUMERICAL RESULTS

  • The numerical scenarios use normally distributed transmission latencies, with σ = µ/10 ms, and select latency targets to demonstrate potential optimization gains.The assumed latency distribution affects specific results, while the authors state that the methods and general tendencies do not change.Source: §V · p. 7

A. Independent interfaces

  • For scenario A, the numerical optimization corresponds visually well with the analytical two-interface solution; smaller packet sizes shift the latency-reliability curves left.The analytical solution determines the optimal splitting threshold γ.Source: A. Independent interfaces · p. 7
  • At x = 0.7 s, Weighted improves reliability from 0.95 for 1-out-of-4 Cloning to 0.997, and reduces latency from 1.05 s to 0.7 s at R = 0.997.Weighted has the highest reliability at the 0.7 s target.Source: Figure 8 · p. 8, A. Independent interfaces · p. 8
  • For another comparison, Weighted provides low-latency reliability similar to 1-out-of-5 below 0.2 s and reliability like 2-out-of-5 around 0.4 s; more redundancy raises reliability above 0.4 s.The higher-redundancy result transmits more redundancy data.Source: A. Independent interfaces · p. 8
  • 0.9999 reliability at x = 0.5 s is one decade better than any k-out-of-N strategy, which reaches only 0.999.The scenario’s results show a more mixed data allocation.Source: A. Independent interfaces · p. 8

B. Interfaces with failure correlation

The correlated-failure model accounts for simultaneous cellular outages, which can substantially change strategy rankings. Cloning leads at high latency but uses three times the bandwidth of a single-interface transmission, while weighted allocation performs best at low latency and in the correlated-failure efficiency comparison.

B. Interfaces with failure correlation

  • The correlated-failure case study allows C1 and C2 to fail simultaneously from a common base-station failure, unlike the independent results.Failure and restoration rates were parameterized using technology MTTR and availability data, then performance was checked with Matlab-based simulation.Source: B. Interfaces with failure correlation · p. 9
  • Cloning achieves the highest high-latency reliability, but correlated failures reduce its reliability by more than one decade versus the independent model.The two cellular interfaces depend on the same base station being operational.Source: B. Interfaces with failure correlation · p. 9, Figure 11 · p. 9
  • At low latency (x ≤0.5 s), weighted allocation performs best; correlated-failure results for it differ minimally from independent results.The weighted strategy is only slightly worse than independent-model cloning at higher latency.Source: B. Interfaces with failure correlation · pp. 9–10
  • The 2-out-of-3 strategy uses half the bandwidth of cloning, but correlated failures make it the worst strategy except from 0.35–0.42 s.In the independent case it reduces latency up to almost 0.9 s at R = 0.996; with correlated failures it lacks reliability before 0.2 s.Source: B. Interfaces with failure correlation · p. 10
  • Under correlated failures, weighted allocation is the best efficiency choice across the latency range, unlike the independent-model comparison.Efficiency is measured as achieved reliability, in number of nines, relative to coded data transmitted; Fig. 12 presents scenario E.Source: B. Interfaces with failure correlation · p. 10, Figure 12 · p. 10

C. Experimental validation

The authors compare curve-based performance calculations with playback simulations of measured interface traces, and report that the results practically coincide. Their conclusions report latency reductions of up to around 40% in practical systems and identify cases where optimized weighting succeeds where simple k-out-of-n strategies do not.

C. Experimental validation

  • The experimental validation uses 128-byte UDP packets sent every 100 ms during a work day at Aalborg University campus.The method's validation does not explicitly cover all scenarios A–E.Source: C. Experimental validation · p. 10
  • The case study uses a base-station failure probability Pe = 0.99, which the authors note may be high relative to a real-life system.They state that the analysis is intended to show how such factors can be modeled.Source: C. Experimental validation · p. 10
  • For measured LTE, HSPA, and Wi-Fi traces, calculated and playback-simulation performance curves practically coincide, and 1-out-of-3 outperforms each individual interface.The comparison is shown in Fig. 13; the weighted scheme is omitted because the small payload provides no gain from splitting.Source: C. Experimental validation · p. 10, Figure 13 · p. 10

VI. CONCLUSIONS AND OUTLOOK

  • The concluding framework combines traditional reliability models with technology-specific latency distributions and covers independent paths and common error causes.The authors report experimental validation of the performance-combination method and practical gains in a three-interface k-out-of-n scenario.Source: §VI · p. 11
  • Up to around 40% latency reduction is reported for large messages over low-rate technologies when transmission time is substantial relative to access delay.The conclusions also report that optimized weighting can deliver low-latency reliability in some cases where simple k-out-of-n cannot.Source: §VI · p. 11

APPENDIX

The appendix estimates failure and restoration rates for the dependent cellular links by combining known availabilities with assumed restoration rates. It uses state probabilities and balance equations to obtain rates consistent with the model constraints.

APPENDIX

  • The cellular subsystem CTMC represents C1 and C2 with correlated failures, and its inputs include individual link and base-station availabilities.Restoration rates are assumed because they are not known for this case study.Source: APPENDIX · p. 11, Figure 14 · p. 11
  • The state probabilities πi are determined from the known availabilities AC1, AC2, and ABS using a linear equation system.The corresponding state-probability relations are specified for the states shown in Fig. 14.Source: APPENDIX · p. 11, Figure 14 · p. 11
  • Balance equations combine the state probabilities with assumed mean restoration rates to solve for the failure rates.The resulting λC1, λC2, and λBS satisfy the system's state-probability, restoration-rate, and balance-relation constraints.Source: APPENDIX · p. 11
Loading 1711.07771v1…