PaperlayerView exact revision on arXiv

Source-linked AI summary

End-to-End Simulation of 5G mmWave Networks

Marco Mezzavilla, Menglei Zhang, Michele Polese, Russell Ford, Sourjya Dutta, Sundeep Rangan, Michele Zorzi

arXiv:1705.02882v3cs.NI

TL;DR

mmWave 5G requires cross-layer, end-to-end evaluation because its directional, dynamic channels affect the entire protocol stack. This paper presents and evaluates a modular ns–3 mmWave module with detailed and trace-driven channels, customizable PHY/MAC layers, LTE/EPC integration, and dual connectivity. Representative simulations demonstrate channel, scheduling, latency, mobility, and congestion-control behavior.

  • Problem

    mmWave research needs full-stack simulation because directional, rapidly varying, and blockage-prone channels create challenges across the protocol stack.

  • Method

    The paper presents a modular ns–3 mmWave module combining configurable channel models, custom PHY/MAC layers, LTE/EPC integration, and advanced dual-connectivity support.

  • Results

    Representative simulations report channel-model behavior, scheduling and latency results, LTE-assisted recovery during mmWave outages, and performance of higher-layer protocols.

  • Takeaways & Limitations

    The module provides an open-source, customizable environment for testing 5G mmWave protocols and end-to-end connectivity across configurable network scenarios.

Abstract

from arXiv · show

Due to its potential for multi-gigabit and low latency wireless links, millimeter wave (mmWave) technology is expected to play a central role in 5th generation cellular systems. While there has been considerable progress in understanding the mmWave physical layer, innovations will be required at all layers of the protocol stack, in both the access and the core network. Discrete-event network simulation is essential for end-to-end, cross-layer research and development. This paper provides a tutorial on a recently developed full-stack mmWave module integrated into the widely used open-source ns--3 simulator. The module includes a number of detailed statistical channel models as well as the ability to incorporate real measurements or ray-tracing data. The Physical (PHY) and Medium Access Control (MAC) layers are modular and highly customizable, making it easy to integrate algorithms or compare Orthogonal Frequency Division Multiplexing (OFDM) numerologies, for example. The module is interfaced with the core network of the ns--3 Long Term Evolution (LTE) module for full-stack simulations of end-to-end connectivity, and advanced architectural features, such as dual-connectivity, are also available. To facilitate the understanding of the module, and verify its correct functioning, we provide several examples that show the performance of the custom mmWave stack as well as custom congestion control algorithms designed specifically for efficient utilization of the mmWave channel.

I. INTRODUCTION

mmWave promises high-throughput, low-latency 5G connectivity but introduces propagation, mobility, control, and higher-layer challenges requiring full-stack simulation. The paper presents an open-source ns–3 mmWave module designed for cross-layer and end-to-end evaluation.

  • Motivation: mmWave is attractive for 5G because it offers massive bandwidth, high-gain antenna arrays, and potential multi-gigabit throughput.Available mmWave spectrum can exceed 1 GHz, while directional antennas help offset poor propagation.
  • Contribution: The proposed ns–3 module provides a modular, customizable platform for evaluating LTE-like 5G mmWave protocols across layers and end-to-end connectivity.It follows LTE LENA architecture and exposes interfaces to LTE/EPC protocols.
  • Challenges: Directional transmission requires continual beam tracking and can suffer complete blockage from materials or moving objects.A blocked link may require jointly searching for and selecting another path.
  • Challenges: Traditional synchronization and initial-access procedures are complicated because directional broadcasts may not provide adequate detection range.Cell discovery must account for directional gain and beam alignment.
  • Challenges: Rapid channel dynamics and shadowing create challenges for mobility, handovers, dual connectivity, and transport-layer congestion control.Dual connectivity can maintain an LTE connection while a primary 5G link fails.

Potentials and Challenges of System-level Simulations of mmWave Networks

Accurate mmWave system simulation must capture channel behavior, mobility, deployment, and protocol-stack interactions across the network. The ns–3 module addresses this need through an LTE-based, modular architecture for end-to-end 3GPP-style cellular simulations.

  • Simulation Requirements: End-to-end simulation links channel and PHY effects to behavior across the whole protocol stack.The paper emphasizes detailed modeling of interacting cellular-system elements for accurate results.
  • Simulation Requirements: Realistic channel models must represent LOS/NLOS propagation, beamforming, Doppler, and the accuracy–complexity trade-off.These factors affect link budget, interference, control procedures, and end-to-end performance.
  • Simulation Requirements: Dense deployment and user mobility require frequent access-point updates and realistic movement modeling.Mobility also affects beam-tracking performance.
  • Simulation Gap: Open-source tools previously lacked thorough integration of mmWave channels, cellular protocol stacks, TCP/IP protocols, realistic scenarios, and mobility.The cited 802.11ad simulator does not support cellular and 3GPP-like scenarios.
  • Module Architecture: The module builds on ns–3 LTE/LENA, adds custom PHY and MAC layers, and supports LTE/mmWave dual-stack devices.Its SAP interfaces enable interoperability with LTE RLC and EPC components.

V. CHANNEL AND MIMO MODELING

The module offers channel models spanning detailed 3GPP statistics, trace-driven realism, and lower-complexity statistical modeling. Its 3GPP implementation includes mobility, blockage, pathloss, fading, and spatial-consistency features.

  • Channel Models: The available channel models trade computational complexity, flexibility, and accuracy.They include 3GPP statistical, measurement or ray-tracing trace, and MATLAB-trace statistical models.
  • 3GPP Statistical Channel Model: The 3GPP model supports 6–100 GHz operation, mobility, spatial consistency, random blockage, and outdoor-to-indoor communications.Optional features can be enabled for different deployment scenarios.
  • 3GPP Statistical Channel Model: Pathloss modeling represents statistical LOS/NLOS conditions, outdoor-to-indoor penetration loss, buildings, obstacles, and optional shadowing.Different classes determine LOS conditions statistically or from node and obstacle geometry.
  • 3GPP Statistical Channel Model: Small-scale fading uses a detailed 3D statistical spatial channel representation with clusters and multiple rays.Each cluster combines M = 20 rays with differing arrival and departure angles.
  • 3GPP Statistical Channel Model: Spatial consistency and blockage extend the channel model for mobility and attenuation caused by self-blocking or external objects.Blockage attenuation and correlation depend on orientation, angles, mobility, blocker speed, and scenario.

2) Ray-tracing or Measurement Trace Model:

Trace-driven and statistical channel options support different realism and computational-cost requirements, while interference modeling connects channel gains to SINR and dense-network behavior. The examples demonstrate configurable beamforming, fading, and interference effects.

  • Ray-tracing or Measurement Trace Model: Ray-tracing models use measurement or software-generated traces containing paths, propagation loss, delays, and arrival and departure angles.Because traces describe a specific route, the simulation scenario must be chosen beforehand.
  • NYU Statistical Model: The NYU statistical model uses pre-generated spatial signatures and beamforming vectors to reduce ns–3 computational overhead.Channel matrices are periodically and independently updated to represent large-scale fading.
  • NYU Statistical Model: Semi-empirical modeling overlays statistical channels with measured hand, human, and metal-plate blockage events.This represents soft transitions between LOS and NLOS conditions.
  • Beamforming: Beam search selects the highest-power beam from a discrete codebook, while covariance-based methods use estimated spatial correlation matrices.Both approaches determine transmit and receive beamforming vectors.
  • Interference: Dense topologies can remain interference-limited, and SDMA or multi-user MIMO requires explicit intra-cell interference computation.The module computes desired and interfering beamforming gains when evaluating SINR.

D. Error Model

The error model abstracts multi-carrier channel quality into effective SINR and uses link-level performance curves to estimate Transport Block errors. It accounts for multiple Code Blocks when determining whether a Transport Block can be decoded.

  • The model derives one effective SINR value from mutual-information-based multi-carrier compression metrics.
  • Link-level LTE PHY curves approximate Block Error Rate as a function of code-block size, MCS, and effective SINR.
  • The receiver uses the estimated error probability to decide whether each Transport Block is decoded successfully.
  • Because Transport Blocks can contain multiple Code Blocks, the model combines their errors to obtain the Transport Block error rate.

E. Examples

The examples demonstrate how channel conditions, beamforming updates, blockage, and configurable frame structures affect mmWave simulation behavior. They also show how the PHY supports flexible TDD allocation and variable-length slots.

  • Examples: Beamforming method and blockage strongly affect SINR in the rural scenario, with observed reductions of 20 dB under specific conditions.The fixed-horizontal-plane beam search loses 20 dB when it cannot align with the LOS cluster, while covariance-based beamforming loses 20 dB when that cluster is blocked.
  • Examples: Ray-tracing SINR changes abruptly at LOS/NLOS transitions, remaining relatively stable during LOS and becoming more variable during NLOS.
  • Examples: Building blockage causes rapid capacity loss lasting seconds, whereas human blockage degrades the channel more gradually for a shorter interval.
  • Frame Structure: The module implements configurable TDD frames and subframes with flexible control and data placement for variable-TTI MAC scheduling.
  • Frame Structure: The illustrated design uses a 1 ms frame divided into ten 100 µs subframes, each representing 24 OFDM symbols.

B. PHY Transmission and Reception

The PHY and MAC implementation coordinates configurable mmWave transmission, reception, scheduling, retransmission, and adaptive coding. The examples evaluate these mechanisms under fading and AWGN conditions while exposing an explicit control-channel modeling limitation.

  • PHY Transmission and Reception: The PHY classes handle DL and UL control and data transmission, with slot timing configured dynamically by the MAC.
  • PHY Transmission and Reception: Data-slot transmission updates transmit and receive beamforming vectors, while control slots use an ideal control-channel assumption without beamforming updates.
  • PHY Transmission and Reception: The PHY computes subband SINR, generates CQI feedback, and invokes the error model to determine packet loss, including HARQ-related processing.
  • MAC and AMC: The MAC coordinates scheduling and retransmission, exchanges buffer reports with RLC, and uses AMC to map CQI to MCS and compute Transport Block size.
  • MAC and AMC: The AMC example evaluates a single-user uplink under AWGN and time-varying multipath fading generated through an artificial 1.5 m/s speed.
  • MAC and AMC: The AWGN curve is shifted approximately 5 dB left relative to the LENA test because the tests use different CQI mappings and target error probabilities.

B. Hybrid ARQ Retransmission

The module provides configurable HARQ with soft combining and schedulers for variable-TTI mmWave transmissions. Its retransmission and scheduling mechanisms address rapid channel variation, latency, fairness, and bandwidth utilization.

  • HARQ: HARQ with soft combining enables fast retransmissions with incremental redundancy, improving decoding success and transmission efficiency.HARQ retransmissions receive priority over new transmissions, but can add latency.
  • HARQ: The mmWave HARQ implementation supports multiple uplink and downlink processes per user, configurable process counts, larger codewords, and flexible-TTI retransmissions.Each transport block permits up to three transmission attempts, while NumHarqProcesses controls simultaneous stop-and-wait processes.
  • Schedulers: Variable-TTI schedulers allocate time-domain OFDM symbols rather than frequency-domain resource blocks, using CQI or SINR and buffer status to estimate required symbols.This common procedure computes each user’s MCS and minimum required symbol count before scheduling data.
  • Schedulers: Round Robin distributes available symbols among active users and reallocates unused symbols when a user needs fewer symbols than its share.HARQ retransmissions are automatically scheduled using available symbols.
  • Schedulers: Fixed-TTI configuration can be inefficient in multi-user cells when a slot spans an entire subframe and schedules only one UE.The SymPerSlot setting controls the fixed slot length.
  • Schedulers: PF balances high-SINR prioritization with cell-edge service, while EDF prioritizes packets by relative deadlines for latency evaluation.MR maximizes throughput by serving highest-SINR users but can provide extremely poor fairness and leave users unscheduled.

VIII. RLC LAYER

The module extends LTE RLC and ns–3 networking components for mmWave operation, adding retransmission compatibility, queue management, dual connectivity, and mobility procedures. Its dual-connectivity implementation supports integrated LTE–mmWave operation, handovers, and multiple data-transfer modes.

  • RLC Layer: The mmWave RLC layer inherits LTE RLC entities, modifies acknowledged-mode retransmissions for mmWave PHY/MAC compatibility, and optionally supports Active Queue Management.AQM can be enabled through EnableAQM, with CoDel as the default scheme.
  • RLC Layer: RLC timers are shortened for mmWave frames, including PollRetransmitTimer changing from 20 ms to 2 ms.The implementation also addresses retransmission re-segmentation when lower-layer transmission opportunities are too small.
  • Dual Connectivity: Dual-stack UEs can connect simultaneously to LTE and mmWave eNBs, supporting multi-connectivity with legacy radio access technologies.The implementation assumes integrated LTE and mmWave core networks sharing backhaul, X2, and S1 connectivity.
  • Dual Connectivity: The McUeNetDevice provides a dual lower-layer stack with separate LTE and mmWave PHY/MAC layers and an RRC instance for each link.A shared EpcUeNas interfaces with both RRC entities, while LTE RRC manages dual-connectivity control functions.
  • Dual Connectivity: Fast Switching activates only one RAT for data at a time, whereas throughput-oriented dual connectivity can transmit simultaneously over both RATs.The module allows different flow-control algorithms to be tested in the simultaneous-transmission mode.
  • Mobility: The LTE eNB coordinates surrounding mmWave eNBs using SINR-based association information and can control cell selection and mobility operations.Supported procedures include secondary-cell handover, fast switching, and X2-based inter-RAT handovers.

X. USE CASES

The use-case section demonstrates how to configure ns–3 mmWave simulations for protocol and end-to-end network studies. It describes setup steps spanning attributes, stack installation, core-network connectivity, mobility, obstacles, and traffic applications.

  • Purpose: The examples illustrate the module’s utility for analyzing novel mmWave protocols and testing higher-layer protocols such as TCP over 5G mmWave networks.The section also points readers to additional studies using the module.
  • Preparation: Users should know ns–3 and can start from basic scripts supplied in the module’s examples folder.The documentation and example scripts provide a basis for designing simulation scripts.
  • Simulation Setup: Simulation setup begins by configuring attributes and creating an MmWaveHelper to build channel objects, install stacks, attach UEs, and control traces.The helper also supports creation of the MmWavePhyMacCommon object.
  • Simulation Setup: End-to-end scenarios additionally create a core network and connect its PGW to a remote host with the internet stack installed on UEs and the host.The MmWavePointToPointEpcHelper creates the core network and provides the PGW pointer.
  • Scenario Configuration: Mobility setup specifies eNB and UE positions and velocities, while buildings and obstacles can be added through ns–3 mobility and buildings helpers.Applications are then configured on UEs and, for end-to-end scenarios, the remote host.

B. Multi-User Scheduling Simulation

The multi-user experiments examine how scheduling policies affect throughput, utilization, fairness, and latency under two traffic-load configurations. The results show that user count and arrival rate determine whether slot availability or channel capacity becomes the main bottleneck.

  • Experiment setup: 70 UEs generate 10 Mbps each, while 7 UEs generate 100 Mbps each, yielding the same total arrival rate of 700 Mbps.These combinations target the system’s delay knee, where requested rates can no longer be serviced for most users.
  • Throughput and fairness: MR and RR exhibit the greatest disparity between high- and low-rate users in the 70-UE scenario.Their preference for high-rate users also produces poor utilization when achievable PHY rates exceed the 10 Mbps arrival rate.
  • Latency: MR provides the best delay performance by scheduling only the highest-rate users, whereas EDF balances users and achieves 1.6 ms mean UE latency.EDF’s mean latency drops below 1 ms for 60 or fewer users at the same arrival rate.
  • Load comparison: At 7 UEs and 100 Mbps per UE, latencies are much lower overall because slot availability is no longer the scheduling bottleneck and utilization is better.The total packet arrival rate remains 700 Mbps, but the higher per-user rate improves utilization for all policies.

C. Latency Evaluation for Variable and Fixed TTI Schemes

This evaluation compares fixed and variable TTI designs for low-latency multi-user mmWave traffic and also illustrates transport behavior under blockage. Variable TTI generally provides lower latency and deadline-miss rates, while congestion-control responses remain important during capacity changes.

  • Method: EDF prioritizes packets by deadline proximity and gives priority to HARQ retransmissions in the latency experiments.Traffic uses Poisson arrivals at 10 Mbps or 100 Mbps per UE, with packets targeted for delivery within 1 ms.
  • TTI comparison: Variable TTI achieves sub-ms average latency and about 10% DMR with over 60 users at 10 Mbps per UE, consistently outperforming fixed TTI.Fixed TTI exceeds 1 ms average latency and has over 60% DMR at 20 users and more than 90% at 40 or more users.
  • TTI comparison: At 100 Mbps per UE, variable TTI reduces radio-link latency by around 500 µs, or 30%, in some cases.Here, multi-user channel capacity rather than minimum slot size is the bottleneck.
  • Subframe design: The 66.67 µs subframe helps with fewer users, but the 100 µs subframe yields lower DMR in the 100 Mbps-per-UE case and reverses the trend with more users.The reversal is attributed to the lower control-to-data-symbol ratio of the longer subframe.
  • Transport behavior: CoDel drops packets when buffering delay is high, reducing the sender’s congestion window during queue buildup caused by human blockage.After the link recovers, the congestion window ramps toward capacity slowly.
  • Transport behavior: Dynamic Receive Window regulates sending rate to reduce delay without rate degradation.The receiver feeds back an optimal window, and the sender uses the minimum of the receive and congestion windows.

E. LTE-aided Multi Connectivity

The LTE-aided multi-connectivity examples evaluate mobility, blockage, and multipath transport across LTE and mmWave links. Dual connectivity can recover communication immediately during mmWave outages, while uncoupled MPTCP congestion control is more stable in NLOS conditions.

  • Scenario design: The mobility scenarios vary UE movement and LOS/NLOS conditions to evaluate mobility-management and multi-connectivity schemes.The framework supports scenarios with LTE and mmWave eNBs and changing channel conditions.
  • Dual connectivity: Dual connectivity immediately recovers communication over LTE when the mmWave link experiences an outage.A stand-alone UE must detect failure and hand over to LTE, creating an interval of zero PDCP throughput.
  • DCE integration: DCE combines ns-3 flexibility with the Linux TCP/IP stack and real applications, enabling use of protocols such as MPTCP.MPTCP can use simultaneous subflows over LTE and mmWave links at different carrier frequencies.
  • MPTCP results: Uncoupled LTE and mmWave congestion control reaches more stable throughput in NLOS conditions than coupled BALIA.The LTE subflow has much lower throughput than the mmWave subflow, so total throughput remains similar to the mmWave connection.

XII. POTENTIAL USES AND FUTURE EXTENSIONS

The framework is positioned as an extensible platform for testing new algorithms, applications, connectivity schemes, and deployment scenarios. Future work focuses on broadening modeled capabilities and addressing scalability for large, mobile, low-latency networks.

  • Extensibility: The modular framework lets researchers implement novel algorithms, procedures, and architectures while retaining backward compatibility.Scheduling and allocation strategies are examples of components that can be extended.
  • Future extensions: Planned extensions include 3GPP-inspired signaling and beam tracking, Carrier Aggregation, multi-hop access and backhaul, and vehicular channel and traffic models.These additions target new radio procedures, multi-connectivity, network topology, and high-mobility evaluation.
  • Future extensions: The roadmap includes virtual and augmented reality, public-safety scenarios, aerial communications, robotics, and related end-to-end performance studies.These environments have propagation conditions and performance requirements that differ from traditional cellular networks.
  • Scalability: Scalability remains a future challenge because large, highly mobile networks require frequent channel updates and very fast packet-scheduling interactions.The authors propose exploring low-rank channel models and cluster-computing deployment to address computational demands.
  • Current scope: The current framework supports configurable channel, PHY, and MAC models, dual connectivity, LTE-core integration, direct code execution, and example simulations.The conclusion presents these capabilities as the basis for testing custom stacks and congestion-control algorithms.
Loading 1705.02882v3…