PaperlayerView exact revision on arXiv

Source-linked AI summary

One-Way Delay Model for 5G Low Latency, High Data Rate Communications

Hamid Hassani, Sonia M. Heemstra de Groot, Ignas G. Niemegeers, Kishor C. Joshi, George Exarchakos

arXiv:2610.13761v1cs.NI

TL;DR

Measuring and interpreting OWD to detect queue buildup is difficult, while queueing delay at the BS is challenging to measure directly. The paper models queueing delay as aggregation delay plus channel access time; near saturation, queueing delay exceeds 90% of OWD, and simulations validate the component-based model across loads and scenarios.

  • Problem

    OWD measurement is challenging because of sender–receiver clock synchronization issues, and inferring queue onset from OWD can be inefficient.

    Source: §I · p. 1

  • Method

    The framework approximates BS queueing delay by combining aggregation delay, measurable at the UE, with channel access time estimated from network conditions.

    Source: §I · pp. 1–2, A. Modeling Delay · p. 3, 2) Channel Access Time · p. 4

  • Results

    Near saturation, queueing delay exceeds 90% of OWD in both single- and multi-UE scenarios, and aggregation delay plus channel access time matches queueing delay across tested loads and scenarios.

    Source: §VI · p. 6

  • Takeaways & Limitations

    The simulations support the framework’s potential for practical queueing-delay estimation across network loads and user scenarios.

    Source: §VII · p. 6

Abstract

from arXiv · show

Achieving low delay and high data rate communication in 5G requires preventing persistent queue backlogs at path bottlenecks. A key indicator of queue buildup is an increase in one-way delay (OWD) of packets. However, accurately measuring OWD is challenging due to synchronization issues between the sender and receiver. Moreover, using OWD to infer the onset of queueing can be inefficient. For instance, congestion control mechanisms that rely on latency as a congestion signal may result in cellular network underutilization, see, e.g., [1]. Thus, to develop more efficient rate control approaches, it is crucial not only to estimate OWD accurately but also to perform a more fine-grained analysis that provides detailed insights into the queue status. This paper presents a framework for modeling OWD in 5G edge networks, where the wireless link is typically the path bottleneck. %, by modeling it through measurable delay components. We demonstrate that OWD is dominated by queueing delay (QD), particularly at high data rates. We analyze QD by decomposing it into two main components: aggregation delay and channel access time. Aggregation delay can be accurately measured at the user equipment (UE), and channel access time can be estimated based on network conditions. Our key contribution shows that QD, while challenging to measure directly at the base station (BS), can be closely approximated by combining these components. We validate this model through NS3 simulations under varying network loads and single- and multi-UE scenarios. This framework provides a foundation for developing advanced mechanisms for congestion control and bottleneck detection in 5G networks.

I. Introduction

The paper motivates finer-grained delay analysis for 5G transport because queue buildup matters for low-latency, high-rate applications, yet delay signals are difficult to measure or interpret. It introduces a framework for estimating queueing delay and situates it among data-driven and model-driven approaches.

I. Introduction

  • Next-generation applications demand high data rates, low latency, and reliability, making well-designed transport protocols important alongside physical- and link-layer advances.The cited XR requirements include 100 Gbps to 1 Tbps data rates, latency below 1 ms, and reliability of five to seven nines.Source: §I · p. 1
  • Prior OWD comparisons of congestion-control mechanisms on a commercial LTE network found significant performance variability in data rate and one-way delay.The comparison used the Pantheon platform in AT&T's LTE network [1].Source: §I · p. 1
  • Loss signals act only after queue overflow, while RTT can misrepresent queueing on asymmetric paths and OWD is difficult to measure and interpret.OWD measurement can be affected by sender–receiver clock offset and skew.Source: §I · p. 1
  • The framework models packet delay in 5G edge architectures with the BS RLC buffer as the path bottleneck and targets queueing delay, which is challenging to measure directly at the BS.The proposed analysis decomposes queueing delay into aggregation delay and channel access time, then validates the model in NS3 simulations across scenarios.Source: §I · pp. 1–2

II. Related Work

  • Data-driven delay prediction can require large, high-quality training datasets and substantial computation, while model-driven approaches may have limited accuracy in complex, dynamic systems.The cited work includes ML approaches to OWD or RTT-violation prediction, and models based on queueing theory or stochastic network calculus –.Source: §II · p. 2

III. Preliminaries

The preliminaries describe how RLC packets are assembled for transmission in transport blocks and identify the RLC buffer as the typical wireless-path bottleneck.

A. Transport Block (TB)

  • In LTE/5G, transport blocks refer to MAC PDUs (PHY SDUs), while an SDU is a higher-layer data entity and a PDU is its corresponding lower-layer entity.The figure illustrates RLC segmentation in 5G.Source: A. Transport Block (TB) · p. 2, Figure 1 · p. 2
  • RLC selects packets, adds headers to form RLC PDUs, and delivers them to the MAC layer in a transport block; the final PDU may contain an SDU segment to fill the block.The transport block size depends on allocated PRBs and the modulation and coding scheme.Source: A. Transport Block (TB) · p. 2

B. RLC Queue

  • Although 5G has queues at several layers, the wireless link is typically slower than the wired domain, placing the path bottleneck at the RLC buffer,.Each UE can have dedicated RLC buffers for its data radio bearers.Source: B. RLC Queue · p. 2

IV. System Architecture

The system architecture places an edge server near end users and treats the BS wireless link as the path bottleneck. Its data-flow model focuses on a single RLC queue while allowing periodic or bursty traffic.

IV. System Architecture

  • The considered edge architecture places the BS as the end-to-end bottleneck because of the rate mismatch between the wired backhaul and air interface, with a dedicated BS queue per UE.The schematic locates the bottleneck at the BS RLC layer [caption].Source: §IV · p. 2, Figure 2 · p. 3

1) System Model:

  • The system model places the edge server near end users and locates the path bottleneck in the wireless link.The framework considers a 5G edge network.Source: 1) System Model · p. 2

2) Data Flow:

  • The data-flow model treats traffic toward a UE as sharing one DRB and RLC queue, and the approach remains applicable to non-identical flows carried by separate DRBs.The model assumes an ideal backhaul, consistent ES-to-BS packet transmission times within a TB, packets fitting within one MTU, and RLC unacknowledged mode without retransmission.Source: 2) Data Flow · pp. 2–3
  • The architecture schematic depicts an edge transport path with a wireless-link bottleneck and an abstract BS resource-allocation view.The figure caption identifies the RLC layer of the BS as the bottleneck location.Source: Figure 2 · p. 3

V. Latency Analysis

The latency analysis focuses on OWD from the edge server to UEs and treats BS RLC queueing as its most significant, often overlooked component. It models queueing delay through aggregation delay and channel access time.

V. Latency Analysis

  • OWD includes processing, queueing, transmission, and alignment delays, but queueing at the BS RLC layer is typically the most significant and often overlooked component,,.The analysis uses “queueing delay” for this source unless otherwise stated.Source: §V · p. 3
  • Queueing delay is the time a packet waits in the BS RLC queue until delivery to the MAC layer, and it depends on access technology, earlier packets, scheduling, other-user occupancy, buffer sizes, and channel quality.With empty RLC buffers and no packet transmitting, delay is nearly zero or at most one scheduling interval; substantial cross-traffic and backlog prolong it.Source: §V · p. 3
  • The MAC scheduler assesses connected-UE buffer occupancy and requests PHY resource allocation at each scheduling interval, which is fixed at 1 ms in LTE and variable in 5G.Queued packets are encapsulated into transport blocks for transmission.Source: §V · p. 3

A. Modeling Delay

  • Each UE has a dedicated BS queue, and the aggregation level is the number of packets grouped in one transport block.For a block containing packets i through i+n, the aggregation level is n+1; this differs from the PHY aggregation level.Source: A. Modeling Delay · p. 3
  • With MAC-layer information, the UE can accurately measure aggregation level because packets in a transport block share MAC timestamps.The paper notes that this can use open-source UEs or software tools running on commercial handsets.Source: A. Modeling Delay · p. 3
  • Queueing delay for packet j in transport block b is modeled as its aggregation delay plus that block’s channel access time.The equation is QD_j = AD_j + T_access,b.Source: A. Modeling Delay · p. 3

1) Aggregation Delay:

Aggregation delay is the time a packet waits in the RLC queue for the last packet in its transport block (TB) to arrive.

1) Aggregation Delay:

  • A packet’s aggregation delay is its RLC-queue wait until the last packet in its TB arrives.The first packet in a TB has the largest aggregation delay; the last packet’s is zero.Source: 1) Aggregation Delay · pp. 3–4
  • The aggregation-delay expression can be rewritten when packets have identical ES-to-BS delay and the physical link has no reordering.Source: 1) Aggregation Delay · p. 4
  • The UE can measure aggregation delay by extracting the ES-added send timestamp from each packet’s payload.Source: 1) Aggregation Delay · p. 4

2) Channel Access Time:

Channel access time spans from the last packet’s arrival at the BS RLC queue to its TB’s departure, and is shared by packets in that TB.

2) Channel Access Time:

  • Channel access time runs from the last packet’s RLC-queue arrival until the corresponding TB departs the queue.Source: 2) Channel Access Time · p. 4
  • All packets in a TB share the same channel access time because they reach the PHY layer simultaneously and depart the RLC queue together.Source: 2) Channel Access Time · p. 4
  • In the assumed edge architecture, OWD is approximated with QD as a major component; significant ES-to-BS latency can be included in Tsys.Tsys covers fixed or bounded delay factors and can be comparable to QD under low network loads.Source: 2) Channel Access Time · p. 4

3) Estimating QD:

The model approximates packet QD by combining aggregation delay and channel access time, and identifies the first packet in a TB as having the greatest QD.

3) Estimating QD:

  • The model approximates each packet’s QD from its aggregation delay and the channel access time of its TB.Source: 3) Estimating QD · p. 4
  • The first packet in a TB experiences the largest queueing delay among its aggregated packets.Source: 3) Estimating QD · p. 4
  • Using QD instead of RTT provides a more accurate bottleneck-congestion indication and may support congestion control that reacts to RLC-layer queue onset.Source: 3) Estimating QD · p. 4
  • Aggregation delay and channel access time depend on packet arrival rate, MAC scheduling, cross-traffic load, and connected UEs’ channel quality.Source: 3) Estimating QD · p. 4

B. Analyzing Taccess,b

Channel access time is analyzed relative to when the last packet of a TB reaches the RLC queue and subsequent MAC scheduling decisions.

B. Analyzing Taccess,b

  • Channel access time remains zero until the last packet of its TB reaches the RLC queue.Source: B. Analyzing Taccess,b · p. 5
  • The MAC scheduling interval is the duration of one slot.Source: B. Analyzing Taccess,b · p. 5
  • When enough PRBs are allocated at the next scheduling decision, the TB can be transmitted in that decision’s scheduling period.Source: B. Analyzing Taccess,b · p. 5
  • If access time stays within this range across consecutive TBs, the queue drains at each scheduling decision without persistent backlog.Source: B. Analyzing Taccess,b · p. 5
  • If a TB is not granted transmission, it waits for later scheduling periods, for example when earlier-arriving queued packets must be sent first.The sufficient-allocation scenario is described as optimal operation that avoids persistent queue buildup.Source: B. Analyzing Taccess,b · p. 5

C. MAC Layer Retransmission

The paper extends its queueing-delay estimate to account for MAC-layer retransmissions, whose effective delay can be measured at the UE and added to the estimate.

C. MAC Layer Retransmission

  • The initial QD approximation applies only when the UE decodes the transport block (TB) on its first transmission attempt.If decoding fails, the UE sends a NACK and the BS may retransmit the TB.Source: C. MAC Layer Retransmission · p. 5
  • The new data indicator (NDI) distinguishes new transmissions from retransmissions, while the redundancy version sequence can identify retransmission order.A toggled NDI marks a new transmission; an unchanged NDI marks a retransmission.Source: C. MAC Layer Retransmission · p. 5
  • The interval from the first transmission to the final retransmission contributes to QD and can be measured at the UE, then added to the estimate.The TB remains in the BS MAC retransmission buffer during this interval; UE-side measurement uses a single clock reference.Source: C. MAC Layer Retransmission · p. 5
  • A TB that remains undecodable after three retransmissions is discarded.This bounds the retransmission process described in the section.Source: C. MAC Layer Retransmission · p. 5

VI. Model Evaluation with NS3

NS3 simulations evaluate network performance and the QD model under different loads and UE scenarios. The component sum AD + Taccess tracks QD, while the share of delay attributable to queueing and its components changes with load.

VI. Model Evaluation with NS3

  • The NS3 evaluation uses 5G-LENA simulations of non-standalone 5G NR with UDP traffic and a per-UE BS queueing policy.The setup uses a 40 MHz line-of-sight channel, the default round-robin TDMA scheduler, and a fixed MCS of 28.Source: §VI · p. 5
  • Near maximum PRB utilization, AD/QD peaks at 80% for single-UE and 70% for multi-UE scenarios; before queueing, Taccess/QD falls to 20% and 30%, respectively.As congestion begins, Taccess becomes the dominant QD component, exceeding 95%.Source: Figure 5 · p. 6, §VI · p. 6
  • Near saturation, QD exceeds 90% of OWD in both scenarios, making queueing the dominant delay component.The QD share is higher with cross-traffic than in the single-UE scenario, attributed to scheduling latency increasing Taccess.Source: §VI · p. 6
  • Across low- and high-load single- and multi-UE scenarios, AD + Taccess matches the QD curve.The validation covers a single UE and two UEs, with UDP traffic at one UE and TCP Cubic cross-traffic at the other.Source: Figure 6 · p. 6, §VI · p. 6
  • Packets in one TB share an RLC departure time, producing similar sawtooth patterns in OWD and QD and an approximately 2.5 ms gap between them.The reported gap is primarily attributed to constant or bounded system latency sources such as processing time and TTI.Source: §VI · p. 6

VII. Summary and Conclusions

The paper approximates BS queueing delay in 5G edge networks by combining aggregation delay and channel access time, validated in NS3 across network loads and user scenarios. Future work includes experimental evaluation and investigating transport-layer rate allocation.

  • At high network loads, BS RLC queueing delay significantly influences OWD; NS3 simulations validate the approximation across loads and user scenarios.Source: §VII · p. 6
  • QD is approximated as aggregation delay plus channel access time, with aggregation delay measured at the UE and channel access time estimated from network conditions.Source: §VII · p. 6
  • Future work will experimentally evaluate accuracy on open-source 5G platforms and investigate applying the framework to end-to-end transport-layer rate allocation.Source: §VII · pp. 6–7
Loading 2610.13761v1…