PaperlayerView exact revision on arXiv

Source-linked AI summary

Wireless Access in Ultra-Reliable Low-Latency Communication (URLLC)

Petar Popovski, Čedomir Stefanović, Jimmy J. Nielsen, Elisabeth de Carvalho, Marko Angjelichinoski, Kasper F. Trillingsgaard, Alexandru-Sabin Bana

arXiv:1810.06938v1cs.ITcs.NI

TL;DR

URLLC requires wireless systems to meet stringent reliability and latency constraints, challenging both access protocols and transmission techniques. The paper develops a communication-theoretic framework for these tradeoffs and discusses access protocols, massive MIMO, multi-connectivity, and statistical assessment. It highlights reliability gains from interface diversity and the need to account for CSI acquisition, channel-model mismatch, and synchronization constraints.

  • Problem

    URLLC couples high reliability with low latency, creating challenging wireless-access design requirements for mission-critical communication.

  • Method

    The paper uses a communication-theoretic framework to discuss URLLC tradeoffs, access protocols, massive MIMO, multi-connectivity, and statistical methodology.

  • Results

    Interface diversity using two independent networks is always superior or equal to dual connectivity in outage, with an order-of-magnitude improvement when cellular-link outage is below 10^-2.

  • Takeaways & Limitations

    URLLC design must jointly consider reliability, latency, access procedures, spatial diversity, interface diversity, and statistically sound reliability assessment.

Abstract

from arXiv · show

The future connectivity landscape and, notably, the 5G wireless systems will feature Ultra-Reliable Low Latency Communication (URLLC). The coupling of high reliability and low latency requirements in URLLC use cases makes the wireless access design very challenging, in terms of both the protocol design and of the associated transmission techniques. This paper aims to provide a broad perspective on the fundamental tradeoffs in URLLC as well as the principles used in building access protocols. Two specific technologies are considered in the context of URLLC: massive MIMO and multi-connectivity, also termed interface diversity. The paper also touches upon the important question of the proper statistical methodology for designing and assessing extremely high reliability levels.

I. INTRODUCTION

URLLC combines stringent reliability and latency requirements, making wireless access design challenging across protocols, transmission techniques, and statistical assessment. The paper develops a framework covering these tradeoffs, access protocols, massive MIMO, multi-connectivity, and reliability measurement.

  • URLLC couples ultra-reliability with low latency in 5G, creating challenging and restrictive wireless-access requirements.
  • The paper provides a framework for analyzing and designing ultra-reliable wireless systems, extending prior work with detailed treatment of access protocols and transmission techniques.
  • Its technical scope includes communication-theoretic tradeoffs, medium access control, massive MIMO, multi-connectivity, and statistical requirements for verifying ultra-reliability.
  • 5G URLLC use cases include cable replacement, Industry 4.0 interactions, automotive driving, industrial control, and tactile Internet applications.
  • The paper notes that 3GPP specifications do not fully describe diverse use-case requirements and are insufficient from a statistical viewpoint.
  • The discussion emphasizes latency and reliability while noting that availability, data rates, payload sizes, deployment, and security are treated elsewhere.

B. Relating Latency and Reliability

URLLC reliability is defined jointly with a latency deadline: reliability is the probability of meeting the deadline, while outage is the probability of exceeding it. Packet design then trades bandwidth, rate, energy, reliability, and latency, with metadata and auxiliary procedures consuming resources.

  • Reliability is the probability that packet latency does not exceed a predefined deadline, while outage is the probability that it does.
  • URLLC commonly involves small payloads, motivating short-packet transmission and finite-blocklength information theory.
  • For fixed payload D and latency T, bandwidth B determines available channel uses N = 2BT and the corresponding transmission rate.
  • With fixed transmission energy, channel and interference conditions determine SINR and achievable reliability; alternatively, bandwidth can be selected to meet a target reliability.
  • Short-packet formats devote substantial resources to metadata, synchronization, and packet detection, which cannot be assumed perfectly reliable in URLLC.

D. URLLC Packet Structure

URLLC packet design must account for small payloads, substantial metadata and auxiliary procedures, whose separate failures multiply and reduce overall reliability. Joint encoding, bandwidth and blocklength choices expose key tradeoffs under finite-blocklength constraints.

  • A 32-byte URLLC packet must meet reliability 1−10−5 within T = 1 ms while also carrying signaling information and metadata.
  • Separating auxiliary procedures, metadata and data makes overall success the product of their success probabilities, deteriorating reliability.
  • Jointly encoding 16 bytes of data and 16 bytes of metadata uses fewer channel uses and therefore requires lower bandwidth than separate encoding.
  • Aggregating packets can increase blocklength for broadcast, but cannot combine data chunks transmitted by different nodes.
  • For B ≤ Bc, increased bandwidth lowers the rate per channel use and lengthens the blocklength, improving reliability and finite-blocklength efficiency.
  • For B > Bc, frequency-separated channel uses add SNR-statistical diversity beyond the diversity from more channel uses.

IV. ACCESS NETWORKING

URLLC access networking must coordinate activity detection, synchronization, grants, data, acknowledgements and higher-layer timing under stringent reliability-latency requirements. Static reservation suits deterministic traffic but demands precise synchronism, while four-step access adapts resources to stochastic arrivals at the cost of more error contributions.

  • Access networking resolves user activity, performs auxiliary procedures, decodes metadata and data, and interacts with higher-layer protocols under service-specific traffic and reliability requirements.
  • A. Access Networking for URLLC Services with Deterministic Traffic Arrivals: Deterministic arrivals support periodic pre-configured resource reservations, providing deterministic uplink and downlink timing.
  • A. Access Networking for URLLC Services with Deterministic Traffic Arrivals: Static allocation relies on precise synchronization, whose maintenance is challenging when service jitter requirements are stringent.
  • 1) Four-step access:: A four-step procedure sends a request, receives a grant, transmits data and receives an acknowledgement, with error probabilities ϵR, ϵG, ϵD and ϵA.
  • 1) Four-step access:: Meeting a target overall error probability is harder for stochastic four-step access than deterministic access because more steps contribute errors.
  • 1) Four-step access:: Four-step access can use resources more efficiently when Pr[α = 1] ≪1 because it supports only active devices.

2) Three-step access:

Three-step access skips the request and polls devices directly, reducing signaling but making data reception and frame synchronization more challenging. Grant-free access removes the grant as well, reducing latency while introducing contention and synchronization burdens; marker-based synchronization trades marker length against receiver complexity.

  • 2) Three-step access:: Three-step access skips the request and sets α = 1, making it suitable for polling or accurately predictable packet arrivals.
  • 2) Three-step access:: Without timing-offset estimation, three-step access makes correct data reception more challenging and requires the BS to synchronize directly on the data transmission.
  • 3) Grant-free access:: Grant-free access skips the grant and sends the actual data in the first device transmission, but contention can cause interference and the BS must acquire frame synchronization.
  • 3) Grant-free access:: Under stringent reliability requirements, pure ALOHA becomes infeasible when access load or interference is high.
  • C. Frame Synchronization: Frame synchronization detects the packet start, commonly by correlating a marker across sliding windows and selecting the highest correlation.
  • C. Frame Synchronization: For a 32-byte packet, achieving correct synchronization of 1−10−5 requires a marker longer than 24 bits even at high SNR.
  • C. Frame Synchronization: Increasing the list length l allows a shorter marker Nm at a fixed upper bound on correct synchronization probability.

V. URLLC IN MASSIVE MULTI-ANTENNA SYSTEMS

Massive multi-antenna systems support URLLC through high-SNR, channel-hardening, and spatial-multiplexing properties. The section considers these benefits in a factory scenario with an antenna-rich access point and comparatively small terminal arrays.

  • Massive antenna systems create many spatial degrees of freedom that benefit URLLC.These degrees of freedom support high SNR, quasi-deterministic links, and spatial division multiplexing.
  • Array gain provides high-SNR links, while channel hardening makes links nearly immune to fast fading in rich-scattering environments below 6 GHz.Together, these properties relax the need for strong coding and can reduce retransmissions for short packets.
  • Spatial division multiplexing lets multiple users exchange data simultaneously, potentially reducing multiple-access latency.Multi-antenna processing for user separation can also introduce computational delay.
  • The illustrative factory scenario uses a massive-MIMO access point serving terminals equipped with one or a small number of antennas.The terminals are modeled as workstations receiving single-stream transmissions in the described setup.
  • The channel model represents propagation through clusters of localized paths characterized by departure and arrival directions and independently attenuated path coefficients.Path directions vary on a long-term scale, whereas path coefficients represent small-scale fading and vary with small movements.

C. Covariance-based Design

The covariance-based design uses long-term channel structure to support reliable, low-latency beamforming while reducing dependence on instantaneous CSI. Zero-forcing separates users through covariance-defined signal subspaces before applying terminal-specific transmission design.

  • The design relies on channel structure rather than small-scale fading to promote reliability and low latency.Covariance matrices capture the structural properties used by the beamforming design.
  • Covariance-matrix singular vectors define the transmit and receive subspaces used for beamforming.The transmitter decomposition groups nonzero singular values in a diagonal matrix, while receiver singular vectors describe the terminal-side covariance structure.
  • Zero-forcing removes inter-terminal interference by projecting transmission into the subspace orthogonal to the interfering terminal’s signal subspace.This operation depends only on long-term channel statistics rather than instantaneous CSI.
  • After interference removal, transmission to one terminal can exploit different levels of CSI knowledge at the transmitter.The general precoder combines an interference-orthogonal term, the desired terminal’s signal subspace, and a linear-combination vector.
  • Only the final coherent-combining term generally requires channel knowledge projected onto the desired covariance subspace.The other precoder terms depend only on long-term channel statistics.

E. Beamforming Methods

The evaluated beamforming structures are ordered by how much instantaneous CSI they use, ranging from interference-free and fully coherent designs to non-coherent and averaged strategies. Their receiver processing correspondingly estimates either aggregate channels or selected channel projections.

  • The transceiver structures are classified by decreasing exploitation of instantaneous CSI at the transmitter.The comparison includes an interference-free upper bound and several covariance- or partial-CSI-based strategies.
  • The All SV–Coh strategy uses all effective singular vectors with coherent combining and requires instantaneous CSI.The receiver estimates the aggregate channel matrix before applying zero-forcing and matched filtering.
  • Strongest SV–Inst selects the strongest singular vector using partial instantaneous CSI at the transmitter.The receiver is matched to the corresponding dominant receive singular vector, while the transmitter estimates the channel projection and selects the strongest mode.
  • All SV–NCoh uses all singular vectors but transmits non-coherently across them.This structure differs from coherent combining while retaining the full singular-vector set.
  • For methods using only the dominant receive singular vector, the receiver estimates the corresponding aggregate-channel projection rather than the full channel matrix.This reduces the channel information needed for receiver processing compared with full-singular-vector methods.

F. Numerical Evaluations

The numerical evaluation compares SINR and packet-error behavior for two users under spatial or time multiplexing and different CSI strategies. Results show that greater transmitter-side channel knowledge usually improves performance, but receive diversity creates an exception.

  • F. Numerical Evaluations: Fig. 7 compares post-processing SINR with the number of terminal antennas in a two-user scenario at ρ = 0 dB.The access point uses M = 100 antennas.
  • F. Numerical Evaluations: Full-CSI methods show a SINR gap over methods based on second-order statistics or partial CSI, which show little differentiation.This comparison concerns the terminal-antenna sweep in Fig. 7.
  • F. Numerical Evaluations: Performance generally improves as the transmitter exploits more channel information.This is the stated general tendency across the evaluated transceiver structures.
  • F. Numerical Evaluations: For N = 4 terminal antennas with receive diversity, All SV–NCoh performs best in the reported Fig. 8 case.The passage attributes this result to receive coherent processing extracting diversity despite non-coherent transmission.
  • F. Numerical Evaluations: Fig. 8 evaluates packet error rate versus transmission slot for single-antenna users, with spatial multiplexing shown by solid lines and time multiplexing by dashed lines.The normalized bandwidth makes channel uses directly reflect delay; the payload contains 100 BPSK bits.
  • F. Numerical Evaluations: Space multiplexing is not always favorable; its benefit depends on the level of CSI exploited at the transmitter.This is reported as a conclusion from the SINR and PER comparisons.
  • VI. MULTI-CONNECTIVITY AND INTERFACE DIVERSITY: Multiple radio interfaces provide an additional diversity dimension for meeting URLLC latency-reliability requirements.The paper calls this multi-connectivity and also discusses interface diversity.
  • VI. MULTI-CONNECTIVITY AND INTERFACE DIVERSITY: Interface diversity avoids dependence on a single access-network point of failure by using different wireless technologies or physically independent mobile operators.It requires endpoint support for duplicating packets and handling multiple received copies.

A. Reliability Model and Numerical Illustration

The paper models multi-link reliability through series/parallel architectures and evaluates end-to-end outage under varying cellular-link conditions. It also frames ultra-reliability assessment around statistical channel models and limited transmitter-side channel knowledge.

  • Reliability model: Independent interfaces are modeled with a series/parallel reliability analogy for single-link, dual-connectivity, and interface-diversity architectures.The independence assumption can be supported by using mobile networks that do not share physical infrastructure.
  • Numerical illustration: IFD with two independent networks is always superior or equal in outage to DC, with roughly an order-of-magnitude advantage when cellular-link outage is below 10^-2.An inferior but independent Wi-Fi connection can also outperform DC below this outage level.
  • Numerical illustration: When the mobile-network core is more reliable, the outage difference between DC and IFD becomes almost negligible, while both improve substantially over a single link for outages between 10^-3 and 10^-2.The comparison depends on the assumed core reliability.
  • Statistical perspective: URLLC reliability figures require a statistical model of the deployment environment, whose properties can be estimated using statistical machine learning.The wireless context is not known a priori.
  • Statistical perspective: With limited channel knowledge, transmitting at a chosen rate over a Rayleigh channel produces an outage probability tied to the estimated channel conditions.The received power follows an exponential distribution with scale parameter θ = 2σ^2 under the Rayleigh assumption.

A. Na¨ıve rate selection under channel uncertainty

The naïve approach estimates average channel power from training measurements and uses that estimate for rate selection. Because the resulting rate is random, it cannot guarantee the target outage probability for every measurement realization.

  • Naïve rate selection: With perfect channel knowledge, the transmitter can determine the maximum rate guaranteeing outage probability ϵ, called the ϵ-outage capacity.The rate is defined as the supremum over rates whose outage probability is at most ϵ.
  • Naïve rate selection: Without knowledge of average channel power θ, the transmitter estimates θ by averaging n independent power measurements and plugs the estimate into the rate formula.The resulting transmission rate R(x1, . . . , xn) becomes a random variable.
  • Naïve rate selection: The naïve estimator cannot guarantee that the outage probability under the selected random rate remains at or below ϵ.The sequence of selected rates induces a distribution over outage probabilities.
  • Naïve rate selection: Limited channel knowledge permits only probabilistic reliability guarantees, so the paper seeks the largest rate-selection function satisfying a prescribed statistical constraint.The favorable rate function depends on the constraint, channel knowledge, and channel statistics.
  • Naïve rate selection: The heuristic rate-selection function chooses ϵ_n for each sample size n to maximize rate under a reliability constraint, with ϵ_n = ϵ recovering the naïve solution.The paper considers two formulations of the statistical reliability constraint.

2) Probably Correct Reliability (PCR):

PCR imposes a stricter constraint by controlling the probability that conditional outage exceeds ϵ. This strengthens reliability guarantees but substantially reduces average throughput and slows convergence to ϵ-outage capacity.

  • Probably Correct Reliability (PCR): PCR limits the probability that the outage probability conditioned on the channel measurements exceeds ϵ, making it more conservative for static environments.The estimate may determine rates for multiple future transmission cycles when training is infrequent.
  • Probably Correct Reliability (PCR): Under the Rayleigh model, the PCR rate choice does not depend on θ and is valid for any ϵ when it respects the maximum tolerance ξ.This follows from selecting ε_n to satisfy the PCR constraint.
  • Evaluation: Limited channel knowledge reduces average throughput under both approaches, although the rate converges to the full-information value as n grows.The evaluation compares average achievable throughput with optimal throughput when the channel distribution is known.
  • Evaluation: PCR causes a larger throughput reduction and significantly slower convergence than the less restrictive average-reliability constraint.The stricter meta-probability guarantee accounts for this tradeoff.
  • Alternative channel models: Parametric rate selection is consistent only when the true channel distribution matches the assumed model; even slight mismatch can violate both reliability constraints.Non-parametric alternatives offer consistency under mild restrictions but require extensive training.

VIII. RELATED WORK

Related work positions URLLC around demanding reliability-latency requirements, diversity-enabled access, multi-antenna CSI challenges, and uncertainty in extreme channel tails. The paper synthesizes these themes with access principles, massive MIMO, multi-connectivity, and statistical methodology.

  • Related work: URLLC literature spans automotive, Industry 4.0, intelligent transportation, emergency relief, tele-surgery, industrial automation, and live audio production use cases.These applications impose varied latency, reliability, and isochronous-communication requirements.
  • Related work: At reliability levels around 10^-5–10^-9, the extreme tail of the channel distribution matters, while model uncertainty can affect the overlying communication protocol.Prior work studies tail approximations and uncertainty in cooperative communication protocols.
  • Related work: URLLC access-networking studies address numerology, robust transmission, connection management, grant-free access, replicas, and multi-user detection, while spectral efficiency is reduced under latency or reliability constraints.The cited access literature includes both grant-based and grant-free schemes.
  • Related work: Multiple antennas provide SNR, diversity, and spatial multiplexing benefits, but instantaneous CSI acquisition remains a severe URLLC limitation under mobility and short coherence times.Feedback latency is especially critical for CSIT in FDD, while reciprocity only reduces the issue in TDD.
  • Related work: Massive MIMO can support non-coherent energy detection by aggregating antenna energies, making detection more robust to localized user mobility than coherent detection.A constellation can combine coherent detection at low mobility with energy detection at high mobility.
  • Related work: Independent parallel transmission paths provide redundancy; 3GPP NR Release 15 specifies packet duplication using two paths from different base stations.The cited motivation is to increase reliability and lower latency.
  • Discussion and outlook: The paper concludes by discussing wireless-access principles, massive MIMO, interface diversity, and statistical methodology within the reliability-latency tradeoff.Relaxing latency beyond 10 or 50 ms can open designs with better coexistence or energy efficiency.
Loading 1810.06938v1…