Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LatencyScope: A System-Level Mathematical Framework for 5G RAN Latency

Arman Maghsoudnia, Aoyu Gong, Raphael Cannatà, Dan Mihai Dumitriu, Haitham Hassanieh

TL;DR

LatencyScope tackles the challenge of predicting 5G RAN latency under diverse configurations by combining detailed formal models of latency sources with a stochastic framework and a scalable configuration optimizer. It outputs latency distributions and can efficiently search billions of configurations to meet URLLC targets, validated on open-source testbeds with real and synthetic traffic, showing superior accuracy over prior analytic models and 5G simulators. The optimizer reveals non-monotonic effects of TDD patterns and the potential benefits of grant-free access under certain conditions, enabling operators to identify bottlenecks and configure networks to meet stringent latency-reliability requirements. Overall, LatencyScope provides a practical tool for offline URLLC planning and system design in 5G RANs by delivering fast, distribution-aware latency analysis and actionable configuration insights.

Abstract

This paper presents LatencyScope, a mathematical framework for accurately computing one-way latency (for uplink and downlink) in the 5G RAN across diverse system configurations. LatencyScope models latency sources at every layer of the Radio Access Network (RAN), pinpointing system-level bottlenecks--such as radio interfaces, scheduling policies, and hardware/software constraints--while capturing their intricate dependencies and their stochastic nature. LatencyScope also includes a configuration optimizer that uses its mathematical models to search through hundreds of billions of configurations and find settings that meet latency-reliability targets under user constraints. We validate LatencyScope on two open-sourced 5G RAN testbeds (srsRAN and OAI), demonstrating that it can closely match empirical latency distributions and significantly outperform prior analytical models and widely used simulators (MATLAB 5G Toolbox, 5G-LENA). It can also find system configurations that meet Ultra-Reliable Low-Latency Communications (URLLC) targets and enable network operators to efficiently identify the best setup for their systems.

LatencyScope: A System-Level Mathematical Framework for 5G RAN Latency

TL;DR

LatencyScope tackles the challenge of predicting 5G RAN latency under diverse configurations by combining detailed formal models of latency sources with a stochastic framework and a scalable configuration optimizer. It outputs latency distributions and can efficiently search billions of configurations to meet URLLC targets, validated on open-source testbeds with real and synthetic traffic, showing superior accuracy over prior analytic models and 5G simulators. The optimizer reveals non-monotonic effects of TDD patterns and the potential benefits of grant-free access under certain conditions, enabling operators to identify bottlenecks and configure networks to meet stringent latency-reliability requirements. Overall, LatencyScope provides a practical tool for offline URLLC planning and system design in 5G RANs by delivering fast, distribution-aware latency analysis and actionable configuration insights.

Abstract

This paper presents LatencyScope, a mathematical framework for accurately computing one-way latency (for uplink and downlink) in the 5G RAN across diverse system configurations. LatencyScope models latency sources at every layer of the Radio Access Network (RAN), pinpointing system-level bottlenecks--such as radio interfaces, scheduling policies, and hardware/software constraints--while capturing their intricate dependencies and their stochastic nature. LatencyScope also includes a configuration optimizer that uses its mathematical models to search through hundreds of billions of configurations and find settings that meet latency-reliability targets under user constraints. We validate LatencyScope on two open-sourced 5G RAN testbeds (srsRAN and OAI), demonstrating that it can closely match empirical latency distributions and significantly outperform prior analytical models and widely used simulators (MATLAB 5G Toolbox, 5G-LENA). It can also find system configurations that meet Ultra-Reliable Low-Latency Communications (URLLC) targets and enable network operators to efficiently identify the best setup for their systems.

Paper Structure

This paper contains 25 sections, 1 theorem, 50 equations, 18 figures, 2 tables, 8 algorithms.

Table of Contents

  1. Introduction
  2. Background
  3. Latency Sources
  4. LatencyScope
  5. Formal Models
  6. Stochastic Framework
  7. Configuration Optimizer
  8. Implementation and Setup
  9. Evaluation
  10. Related Work
  11. Limitations & Future Work
  12. Proofs
  13. Proof of lemma \ref{['lemma1']}
  14. Additional Results
  15. Model Evaluation
  16. ...and 10 more sections

Key Result

lemma 1

Let $n = GCD(SR_P, T)$, $\varphi(\cdot)$ be Euler's totient function, and define the set $D = \left\{\left\lceil \frac{d+1 - SR_O}{n} \right\rceil, \ldots, \left\lfloor \frac{T-1 - SR_O}{n} \right\rfloor \right\}$$D = \emptyset \text{ if } \left\lceil \frac{d+1 - SR_O}{n} \right\rceil \geq \left\lf

Figures (18)

  • Figure 1: Path of a Ping Request through the 5G Stack.
  • Figure 2: Overview of the system-level latency for the journey of a packet. A TDD Common Configuration with the DDDU pattern is used.
  • Figure 3: The 5G testbed setup.
  • Figure 4: Latency distributions for various scenarios. We generate synthetic traffic using different inter-arrival distributions and packet sizes: 1) Constant – Packets of 64 with a fixed inter-arrival time of 101m. 2) Gaussian – Packets of 64 with a Gaussian inter-arrival distribution (mean: 105m, standard deviation: 0.05). 3) Large Packets – Packets of 40000 with a fixed inter-arrival time of 201m. 4) Real Applications - Zoom video call, DOTA 2 multi-player gaming data. (a)--(f) and (h) use numerology 1, while (g) uses numerology 0.
  • Figure 5: Comparison of latency bounds between LatencyScope and Real-World, where scenarios (a)--(h) are the same as in \ref{['fig:latency-distribution-2']} and scenarios (i)--(viii) are shown in \ref{['sec:model_evaluation']}
  • ...and 13 more figures

Theorems & Definitions (1)

  • lemma 1