Source-linked AI summary
An E2E Simulator for 5G NR Networks
Natale Patriciello, Sandra Lagen, Biljana Bojovic, Lorenza Giupponi
TL;DR
The paper addresses the need for a versatile, full-stack, open-source E2E simulator for extracting insights from evolving 5G NR specifications. It extends ns-3 to support the NR RAN and evaluates the result through calibration and realistic E2E experiments. The experiments show that the simulator can reveal unexpected effects of NR numerologies on E2E latency and device performance.
Problem
A versatile, full-stack, open-source E2E simulator is needed to study newly specified 5G NR networks and support interoperability studies.
Method
The paper extends ns-3 with an NR network simulator covering full-stack operation, MAC and PHY models, bandwidth-part management, and E2E evaluation.
Results
The simulator produces expected calibration results and exposes unexpected effects of different NR numerologies on overall E2E latency.
Takeaways & Limitations
Full-stack, high-fidelity simulation can highlight NR behaviors that other simulator types may not reveal.
Takeaways & Limitations
The current implementation lacks RRC-based BWP configuration and reconfiguration, and its PHY layer abstraction is still being developed.
Abstract
from arXiv · showhide
As the specification of the new 5G NR standard proceeds inside 3GPP, the availability of a versatile, full-stack, End-to-End (E2E), and open source simulator becomes a necessity to extract insights from the recently approved 3GPP specifications. This paper presents an extension to ns-3, a well-known discrete-event network simulator, to support the NR Radio Access Network. The present work describes the design and implementation choices at the MAC and PHY layers, and it discusses a technical solution for managing different bandwidth parts. Finally, we present calibration results, according to 3GPP procedures, and we show how to get E2E performance indicators in a realistic deployment scenario, with special emphasis on the E2E latency.
1. Introduction
The paper presents an open-source ns-3 NR simulator designed for full-stack E2E evaluation and interoperability studies. It combines high-fidelity protocol-stack modeling with PHY abstraction, 3GPP calibration, and realistic usage examples.
- NR scope: The paper describes NR support spanning features such as multiple numerologies, dynamic TDD, mmWave frequencies, mini-slots, variable TTIs, and bandwidth-part multiplexing.NR also introduces changes across the protocol stack, including SDAP and mandatory control/user-plane separation in the core network.
- Motivation: The simulator addresses the limits of analytic methods by modeling application-to-PHY behavior and E2E interactions with traffic, the core network, and mobility.Analytic methods simplify sender, receiver, network segments, and layers, limiting result generality.
- Evaluation: 3GPP calibration and realistic deployment examples demonstrate how the simulator can produce E2E performance indicators, including E2E delay across NR numerologies.The authors report using it to evaluate processing and decoding times in E2E delay for different NR numerologies.
- Contribution: The simulator uses PHY-layer abstraction and high-fidelity implementations from MAC through the application layer to support cross-layer and E2E performance evaluation.It is also intended as a platform for assessing NR coexistence with other technologies.
- Open distribution: The simulator is distributed under GPLv2 to foster 5G development and research around an openly available tool.The license choice follows from its derivation from the GPLv2-licensed ns-3 simulator.
2. Scientific Contribution
The paper contributes an open-source NR extension to ns-3, designed for reproducible, full-stack network evaluation and interoperability scenarios. It revises the mmWave foundation toward 3GPP-compliant NR operation while adding flexible MAC, frame, uplink, processing-timing, and bandwidth-part capabilities.
- Motivation: The simulator addresses limited reproducibility and restricted inspectability in existing research and industry simulation tools.The authors emphasize source-code freedom, reproducibility, collaborative development, and open innovation.
- Scientific contribution: The NR module is an open-source extension within ns-3, enabling configuration, tracing, and integration with models for multiple technologies.This supports interoperability experiments and multi-RAT coexistence scenarios involving technologies such as LTE and WiFi.
- Scientific contribution: The NR module is upstreamed to ns-3 and can reuse framework features including Direct Code Execution and real-time execution with equipment.These capabilities support realistic TCP/IP applications and movement from simulation toward emulation.
- Scientific contribution: The contribution adds automatic NR frame configuration, OFDMA access, redesigned MAC scheduling, grant-based uplink access, NR processing timings, and multiple-BWP support.MAC schedulers jointly consider time- and frequency-domain resources for TDMA and OFDMA with variable TTI.
3. NR module overview
The NR module extends ns-3 for end-to-end simulations, connecting core-network traffic through the gNB and UE RAN devices. It adds NR-specific RAN classes and virtual BWP-management layers while reusing existing ns-3 components.
- The NR module supports E2E simulations of 3GPP-oriented cellular networks within ns-3.
- The RAN is organized around NRGnbNetDevice and NRUeNetDevice, with MAC classes interfacing with LTE RLC through SAPs and schedulers configured through LTE RRC.
- NR PHY classes handle directional DL and UL communication while reusing MmWaveSpectrumPhy models for interference, SINR, MI-based errors, and HARQ soft combining.
- NRGnbBwpM and NRUeBwpM are virtual proxy layers that multiplex and demultiplex flows across BWPs with separate PHY and MAC configurations.
- The BWP managers can assign each bearer to one BWP or distribute a flow among multiple BWPs according to configured rules.
4. Physical Layer
The physical layer adapts ns-3 to NR’s numerology-dependent and flexible slot structure. It implements slot-based transmission, scheduling, timing parameters, and configurable processing delays while supporting TDD and multiple numerologies.
- NR uses numerology defined by subcarrier spacing and cyclic prefix, producing different slot, symbol, and frame configurations.
- Unlike LTE’s fixed subframe scheduling, NR supports flexible slot lengths, slots per subframe, OFDM symbol lengths, and numerology-dependent slot scheduling.
- For normal cyclic prefix, each slot contains 14 OFDM symbols and each resource block contains 12 subcarriers.
- The model supports TDD and derives slots, symbol periods, subcarrier spacing, and PRB counts from the configured numerology µ and total bandwidth.
- The implementation introduces slot granularity and adapts PHY transmission, MAC scheduling, resource allocation, processing delays, and PHY–MAC interaction accordingly.
- Transmission, reception, and allocation operate per slot, while variable-length TTIs are handled through dedicated variable-TTI functions and scheduling structures.
- The gNB MAC-to-PHY processing delay is numerology-dependent and defaults to 2 slots, whereas UE transport-block decoding defaults to 100 us.
5. MAC Layer
The NR MAC layer supports flexible variable-TTI multiple-access schemes, dynamic scheduling for downlink and uplink, and an NR-specific grant-based uplink handshake.
- Multiple Access Schemes: The MAC layer supports TDMA with variable TTI, pure OFDMA, and OFDMA with variable TTI, with configurable scheduling policies.Resources can be assigned using variable numbers of OFDM symbols and RBGs within a slot.
- Scheduling Schemes: Dynamic scheduled-based access is supported for both downlink and uplink, while uplink grant-free access remains future work.The implemented uplink grant-based procedure uses scheduling requests and grants.
- Scheduling Schemes: The scheduler represents each allocation with a starting symbol, duration, and RBG bitmap for frequency- and time-domain resource assignment.These parameters align with NR resource-allocation procedures for downlink and uplink.
- UL Handshake: In the uplink handshake, a UE sends an SR, receives a gNB UL grant, transmits data or a BSR after K2 slots, and may receive further grants.The first scheduling assignment is blind because the gNB initially lacks the UE buffer size.
- UL Handshake: The NR module introduces SR transmission on PUCCH and BSR handling in conjunction with MAC PDUs, distinguishing it from earlier ns-3 modules.The BSR informs the scheduler about data stored in the RLC buffers.
6. Bandwidth Part Manager
The simulator manages multiple NR bandwidth parts by assigning different numerologies and traffic types to frequency-divided regions. Its BWP architecture supports flexible routing and reuse, but relies on static, globally matched configurations and omits inter-BWP interference.
- Numerology and BWP design: FDM of numerologies divides total bandwidth into URLLC and eMBB BWPs, trading shorter slots for latency against longer slots for throughput.The configuration uses Bu + Be ≤ B, with separate PRB counts Nu and Ne; PRB width varies with numerology.
- Limitations: Runtime BWP reconfiguration is unavailable because RRC BWP messaging and required physical-channel configuration changes are not implemented.The limitation prevents semi-static BWP reconfiguration in the current mmWave/NR channel model.
- Architecture: The implementation extends ns-3 LTE Carrier Aggregation by installing separate MAC and PHY instances per carrier.NrGnbDevice and NrUeDevice are extended while inheriting the LTE CA architecture.
- Traffic and control: A BwpManager routes traffic and signaling to BWPs using QoS requirements and a static bearer-to-bandwidth-part lookup.Scheduling and HARQ operate per BWP, while the manager maps each message to the corresponding carrier.
- Limitations: All simulated gNBs and UEs share the same BWP configuration, and contiguous BWPs/CCs require guard bands because inter-BWP interference is not modeled.The BWP configuration must match between each gNB and its attached UEs.
- Reuse and applications: The architecture can support RAN slicing through orthogonal BWPs or be reused for Carrier Aggregation, with the main distinction residing in BWP-manager logic.Both use cases share the same implementation blocks but differ in how the managers handle flows.
7. Use Cases
The paper evaluates the simulator through 3GPP calibration and a realistic multi-application 5G scenario. Calibration matches the reference region for SNR, while E2E experiments expose how numerology, processing, and signaling affect latency and goodput.
- Calibration: 3GPP calibration uses indoor-hotspot multi-antenna procedures and compares simulated SINR and SNR distributions with company reference curves.The deployment contains 12 gNBs and 120 randomly placed indoor UEs in a 50 m × 120 m area.
- Calibration: SINR is close to the lowest 3GPP reference curve, while SNR lies entirely within the calibration region and perfectly matches the average in the first setup.The comparison covers Office-Open without shadowing and Shopping-Mall with shadowing enabled.
- Realistic deployment: The realistic scenario combines smartphones, sensors, and IP cameras with UDP and TCP traffic, using random UE positions for statistical significance.Sensor flows transmit 500-byte payloads every 2.5 ms, while video nodes transmit continuously at 10 Mb/s.
- Numerology effects: Reducing slot time with higher numerology can reduce RTT and increase TCP goodput, reaching almost 200 Mb/s at numerology 4.The simulations compare numerologies from µ = 0 to µ = 4 using TCP goodput and UDP one-way delay.
- Latency: Decoding delays from 0 ms to 0.1 ms and 0.5 ms increase overall E2E latency approximately linearly without affecting other phases.The 0 ms decoding-latency case performs best, while fixed decoding delays are added almost unchanged to the end-to-end delay.
- Latency: Higher numerology can unexpectedly increase sensor latency when additional Scheduling Requests offset the benefit of shorter slots.From µ = 2 to µ = 3, increased SR overhead produces higher plotted latency for sensor flows.
8. Future Work
Future development plans target NR-specific PHY abstraction, modulation and coding support, spatial multiplexing, FDD, and more complete mini-slot operation. The roadmap also updates error modeling and beam and stream capabilities to broaden simulator coverage.
- The simulator will extend mini-slot operation by placing PDCCH in the first symbol of the TTI allocation.
- NR PHY abstraction will be added through L2SM and an error model, following the ns-3 LTE module’s approach.The abstraction predicts MAC-layer TBLER using BLER-versus-SINR lookup tables and updated CB-to-TB mappings.
- Support for 256-QAM requires updated MCSs, MI mapping, and new BLER-versus-SINR lookup tables.The work is planned in parallel with the PHY layer abstraction.
- The current simulator supports one beam and one stream per UE, while future extensions target multi-UE transmission and multiple streams per UE.These changes require expanded interference and power-distribution models, plus updated PHY abstractions and precoders.
- FDD support is planned because it will be needed for future deployments as regulation progresses.The simulator currently supports dynamic TDD, mainly considered for higher-frequency deployment ranges.
8.2. MAC
MAC-layer future work addresses contention-based uplink access, bearer-aware scheduling, and punctured scheduling for multiplexing eMBB and URLLC traffic. These additions target NR procedures involving collisions, latency constraints, and dynamic resource use.
- UL grant-free access: UL grant-free access eliminates the SR, BSR, and UL grant handshake compared with UL grant-based access.Resources are configured through RRC signaling and activated or deactivated through DCI in the PDCCH.
- UL grant-free access: The simulator needs UL grant-free resources configured by RRC and an error model for contention-based access.Because the access is non-scheduled, collisions may arise.
- Bearer-aware scheduling: The current scheduler does not prioritize bearer flows according to guaranteed bit rate or latency deadlines.A general policy is planned to enforce such constraints for active flows.
- Punctured scheduling: Punctured scheduling lets URLLC traffic use resources already allocated to eMBB traffic and signals preemption to affected UEs.Required simulator changes include symbol-by-symbol scheduling and preemption indication in downlink control channels.
8.3. Upper layers extensions
The simulator’s RRC, RLC, and PDCP layers still rely on LTE implementations, so updates and extensions are planned for NR operation and latency-oriented requirements.
- RRC, RLC, and PDCP currently rely on LTE implementations.
- RLC and PDCP simplifications have been proposed to facilitate the targeted latency reduction of NR.
- RRC requires extensions because its original implementation was mainly designed for CONNECTED mode.Updates and extensions will be considered for inclusion.
8.4. SDAP
The simulator lacks the SDAP entity required by NR’s QoS framework. Planned SDAP support maps QoS flows to data radio bearers and carries configurable flow markings across protocol layers.
- SDAP maps distinct QoS flows into data radio bearers in the 5G QoS framework.
- The simulator currently lacks an SDAP entity between upper layers and its peer SDAP entity.The planned entity receives SDUs from upper layers and sends SDAP PDUs through lower layers.
- SDAP entities should mark QoS flows and support static or dynamic flow-to-bearer mapping through RRC.
8.5. Core Network
The core-network design incorporates 5G service-based architecture changes and prepares a compatible API for external users. The module also targets NR unlicensed-band coexistence through Listen Before Talk and ns-3’s multi-technology support.
- 5G core architecture: 5G core-network changes separate control and user planes and provide each service as a network function.
- 5G core architecture: The design incorporates the LTE ns-3 EPC functional split between PGW and SGW while preparing a compatible API for external users.
- Unlicensed operation: NR natively supports unlicensed-band operation, with ongoing work below 7 GHz and expected Release 17 extensions for mmWave bands.
- Unlicensed operation: Listen Before Talk functionalities are intended to facilitate coexistence with WiFi and WiGig in the same band.
- Unlicensed operation: ns-3’s multi-technology characteristic supports proposing this unlicensed-operation extension in the module.
9. Conclusion
The paper presents a full-stack NR simulator extension for end-to-end, standard-compliant network simulation and validates it through calibration and indoor scenarios. Its experiments assess numerology effects, parameter sensitivities, and realistic end-to-end behavior, including latency.
- Contribution and validation: The work extends the ns-3 mmWave simulation tool to support end-to-end simulation of NR networks on an up-to-date, standard-compliant platform.
- Contribution and validation: The extension was validated and calibrated in different indoor scenarios against similar proprietary 3GPP simulators and recommendations.
- Evaluation: A complex end-to-end simulation campaign assessed the impact of different NR numerologies on devices’ overall end-to-end latency.
- Evaluation: Parameter-sensitivity analysis showed that full protocol-stack, high-fidelity models can reveal unexpected behaviors that other simulator types could hardly highlight.
- Roadmap: The conclusion also presents a roadmap for additions and modifications aligned with current and future 3GPP NR simulator development.