PaperlayerView exact revision on arXiv

Source-linked AI summary

New Radio Physical Layer Abstraction for System-Level Simulations of 5G Networks

Sandra Lagen, Kevin Wanuga, Hussain Elkotby, Sanjay Goyal, Natale Patriciello, Lorenza Giupponi

arXiv:2001.10309v2cs.ITeess.SPmath.IT

TL;DR

NR’s flexible allocations, coding, and MCS configurations complicate PHY abstraction for system-level simulation. The paper calibrates EESM and SINR-BLER tables with an NR-compliant link-level simulator, then integrates the model into ns-3; the calibrated abstraction is reported insensitive to NR numerology, and adaptive MCS with HARQ has no application-layer packet losses through 70 m.

  • Problem

    NR’s variable allocations, coding features, and MCS configurations make PHY abstraction complex for system-level simulation.

    Source: §I · p. 1

  • Method

    The model calibrates EESM and generates SINR-BLER lookup tables with an NR-compliant link-level simulator, then integrates the abstraction into ns-3.

    Source: §I · p. 2

  • Results

    The calibrated L2SM is insensitive to NR numerology, and adaptive MCS with HARQ has no application-layer packet losses up to 70 m.

    Source: C. Results with adaptive MCS · p. 6, §V · p. 7

  • Takeaways & Limitations

    The reported numerology insensitivity generalizes the model’s application to all NR numerologies and frequency ranges.

    Source: §V · p. 7

Abstract

from arXiv · show

A physical layer (PHY) abstraction model estimates the PHY performance in system-level simulators to speed up the simulations. This paper presents a PHY abstraction model for 5G New Radio (NR) and its integration into an open-source ns-3 based NR system-level simulator. The model capitalizes on the exponential effective signal-to-interference-plus-noise ratio (SINR) mapping (EESM) and considers the latest NR specification. To generate it, we used an NR-compliant link-level simulator to calibrate the EESM method as well as to obtain SINR-block error rate (BLER) lookup tables for various NR configurations. We also illustrate the usability of the developed model through end-to-end simulations in ns-3, under different NR settings of modulation and coding schemes, hybrid automatic repeat request combining methods, and link adaptation approaches.

I. INTRODUCTION

NR’s flexible radio and coding features make PHY performance modeling complex, while system-level simulators need a faster alternative to detailed PHY processing. The paper presents an EESM-based NR abstraction calibrated with a compliant link-level simulator and integrated into ns-3.

I. INTRODUCTION

  • NR performance validation is needed before real implementations, and this paper presents a PHY abstraction model for that simulation context.The paper outlines EESM, the NR abstraction model, ns-3 end-to-end results, and conclusions.Source: §I · pp. 1–2
  • System-level simulators replace detailed PHY processing with link-to-system mapping that estimates transport BLER from precomputed tables, saving execution time.For frequency-selective channels, the mapping compresses per-subcarrier SINRs to an effective SINR, then uses an SINR-BLER lookup table.Source: §I · p. 1
  • EESM is one of several SINR-mapping methods, and 3GPP recommends a single-parameter EESM method despite no consensus on one method across researchers.Other methods include MIESM, capacity ESM, and logarithmic ESM.Source: §I · p. 1
  • NR’s variable resource allocations, wider bandwidths, coding schemes, and MCS configurations increase the complexity of modeling its PHY performance.The model must account for features including LDPC coding, multiple lifting sizes and base graphs, and modulation orders up to 256-QAM.Source: §I · pp. 1–2
  • The paper calibrates EESM using an NR-compliant link-level simulator, generates SINR-BLER tables for varied NR settings, and integrates the model into ns-3.The authors report that the calibrated L2SM is insensitive to NR numerology and make the implementation publicly available.Source: §I · p. 2

II. EXPONENTIAL EFFECTIVE SINR MAPPING

The paper uses EESM to compress per-RB SINRs into an effective SINR, with the retransmission-combining calculation depending on the HARQ method. The described assumptions keep the same MCS and RB count across retransmissions.

A. Effective SINR for single transmission

  • For single transmissions, EESM maps the SINRs of allocated RBs to one effective SINR using a calibrated parameter β.The mapping is calibrated so the effective SINR corresponds to the BLER of an equivalent AWGN channel.Source: A. Effective SINR for single… · p. 2
  • The single-transmission equation assumes the same MCS is used over all RBs.This is described as a common assumption in 4G and 5G systems.Source: A. Effective SINR for single… · p. 2

B. Effective SINR for combined retransmissions

  • For simplicity, retransmissions of a HARQ process use the same MCS and number of RBs, although RB positions and per-RB SINRs may vary.The paper considers both HARQ-CC and HARQ-IR, whose EESM calculations differ.Source: B. Effective SINR for combined… · p. 2
  • With HARQ-CC, retransmissions repeat the same coded bits, so the ECR remains unchanged and corresponding resource SINRs are summed before EESM.The combined SINRs are computed over the RBs selected for combining.Source: B. Effective SINR for combined… · p. 2
  • With HARQ-IR, retransmissions carry different coded bits, so both effective SINR and ECR are recomputed after each retransmission.The ECR uses the information-bit count and coded-bit counts across retransmissions; the SINR calculation uses the RBs maintained for combining.Source: B. Effective SINR for combined… · p. 2

III. NR PHY ABSTRACTION

The NR abstraction maps allocated-RB SINRs and transmission context to transport BLER through staged processing. It uses HARQ-specific history and NR-compliant link-level calibration and lookup tables.

III. NR PHY ABSTRACTION

  • The abstraction takes per-RB SINRs, MCS selection, TBS, and HARQ history as inputs and returns MAC transport-block BLER.Its stages are SINR compression, LDPC base-graph selection, code-block segmentation, code BLER lookup, and transport-BLER mapping.Source: §III · p. 3
  • HARQ-CC history stores per-RB SINRs, whereas HARQ-IR history stores the last effective SINR and coded-bit count for each prior retransmission.The effective SINR is computed using EESM, and β is optimized with an NR-compliant link-level simulator.Source: §III · p. 3
  • After LDPC base-graph selection and any needed segmentation, code BLER is obtained from NR-compliant link-level SINR-BLER tables using effective SINR, ECR, MCS, and code-block size.The resulting code BLER and number of code blocks determine transport BLER.Source: §III · p. 3

A. MCS selection, BG selection, and code block segmentation

The model uses NR MCS selection and coding inputs to determine how transport blocks are represented and segmented for BLER estimation. Link adaptation selects the highest MCS that meets a target transport-BLER constraint.

A. MCS selection, BG selection, and code block segmentation

  • The model focuses on MCS Table1, which supports up to 64-QAM, and MCS Table2, which supports up to 256-QAM.An MCS index specifies an ECR, modulation order, and spectral efficiency.Source: A. MCS selection, BG selection, and… · p. 3
  • For link adaptation, the highest MCS index meeting a target transport-BLER constraint is selected for the channel conditions and reported through the associated CQI index.The model can be used as an error model or for this MCS-selection procedure.Source: A. MCS selection, BG selection, and… · p. 3
  • LDPC base-graph selection depends on the selected MCS ECR and TBS; segmentation occurs when the transport block exceeds the base graph’s maximum code-block size.The maximum CBS is 8448 bits for BG1 and 3840 bits for BG2; segmentation accounts for base-graph selection and lifting size.Source: A. MCS selection, BG selection, and… · pp. 3–4

B. SINR Compression (β optimization)

The authors calibrate EESM’s β with an NR-compliant link-level simulator by matching fading-channel BLER to AWGN BLER. The resulting mapping is reported as insensitive to SCS and NR numerology.

B. SINR Compression (β optimization)

  • An NR-compliant link-level simulator calibrates β using LOS and NLOS TDL channels with 30 kHz and 60 kHz SCS.The channel delay spreads range from 30 ns to 316 ns.Source: B. SINR Compression (β optimization) · p. 4
  • The optimal β minimizes the error between measured fading-channel BLER and AWGN BLER at the EESM-derived effective SINR.Calibration spans channel realizations and noise-variance values.Source: B. SINR Compression (β optimization) · p. 4
  • The calibrated effective SINR mapping is reported as insensitive to SCS, or NR numerology, and carrier frequency.The optimal β values in Table II are averaged over both simulated SCSs.Source: B. SINR Compression (β optimization) · p. 4

C. Effective SINR to code BLER mapping

The model uses link-level simulations to generate code-BLER lookup curves for NR configurations. Since BLER depends on code block size, the system-level model uses conservative bounds when matching effective SINR to the curves.

C. Effective SINR to code BLER mapping

  • Link-level simulations generate code-BLER-versus-SINR curves across MCSs, code block sizes, and corresponding LDPC base-graph choices.The simulations use 28 GHz, 120 kHz SCS, 200 MHz bandwidth, and resource allocations of 1–132 RBs and 1–10 OFDM symbols.Source: C. Effective SINR to code BLER mapping · p. 4
  • Larger code block size gives better code BLER at fixed SINR for the illustrated MCSs.The paper shows MCS23 of MCS Table1 and MCS21 of MCS Table2; MCS23 of Table1 is equivalent to MCS16 of Table2, explaining its lower SINR range.Source: C. Effective SINR to code BLER mapping · p. 5, Figure 3 · p. 5, Figure 4 · p. 5
  • Because the lookup curves are quantized and cover a subset of code block sizes, the system-level model uses lower bounds for actual code block size and effective SINR.For HARQ-IR, effective ECR is limited to the lowest ECR among MCSs with the selected modulation order.Source: C. Effective SINR to code BLER mapping · p. 5

D. Transport BLER computation

The link-level lookup curves provide code BLER without code-block segmentation, while system-level transport blocks may require segmentation. The model combines code-block BLERs to obtain transport BLER.

D. Transport BLER computation

  • System-level transport blocks may require code-block segmentation, so code BLER from the link-level lookup must be converted to transport BLER for the given TBS.The baseline link-level curves omit segmentation, making transport BLER equal to code BLER there.Source: D. Transport BLER computation · p. 5
  • The model combines the BLERs of the C segmented code blocks to compute transport-block BLER.The approximate equality relies on NR code blocks being roughly equal in size.Source: D. Transport BLER computation · p. 5

IV. END-TO-END EVALUATION IN NS-3

The NR PHY abstraction model is integrated into the open-source ns-3 NR system-level simulator and evaluated end to end under several NR settings.

IV. END-TO-END EVALUATION IN NS-3

  • The ns-3 evaluation covers fixed and adaptive MCS, HARQ-CC and HARQ-IR, and both MCS Table1 and Table2.The evaluation is presented to illustrate the model’s usability.Source: §IV · p. 5

A. Simulation Scenario

The end-to-end scenario evaluates a gNB–UE downlink under varied link distances, using UDP traffic and collecting packet delay and loss metrics.

A. Simulation Scenario

  • The Urban Micro gNB–UE link used 28 GHz, 120 kHz SCS, 100 MHz bandwidth, and distances of 10, 30, 50, and 70 m.The setup used a 4×8 gNB array, a 2×4 UE array, and a fast-fading 3GPP channel model.Source: A. Simulation Scenario · p. 5
  • Downlink UDP sent one 100-byte packet every 200 ms for 50 s; measured outcomes were end-to-end delay, application-layer packet loss, and PHY-layer transmission failures.The channel was updated every 150 ms so packets encountered different channel realizations.Source: A. Simulation Scenario · p. 5
  • In the simulated downlink UDP scenario, RLC used unacknowledged mode, with two slots of PHY-MAC processing delay (250 μs) and 100 μs decoding latency.The RLC-to-MAC reporting buffer-status timer was 1 ms.Source: A. Simulation Scenario · pp. 5–6

B. Results with fixed MCS

Adaptive MCS with HARQ combining prevents application packet loss through 70 m in the evaluated setup, whereas the tested fixed-MCS strategies cannot serve the UE at 70 m. Fixed-MCS outcomes also vary with MCS choice and HARQ method.

B. Results with fixed MCS

  • At 50 m with the same modulation order and ECR, HARQ-IR performed better than HARQ-CC in delay and application throughput.Combining retransmissions reduces the equivalent ECR under HARQ-IR, increasing robustness to channel impairments.Source: B. Results with fixed MCS · p. 6
  • For same-modulation, same-ECR settings, Table1 MCS13 and Table2 MCS11 match at any gNB-UE distance under HARQ-CC; under HARQ-IR, small differences occur at 50 m for Table1 MCS13 and Table2 MCS7.Under HARQ-IR, Table2 MCS7’s effective ECR can be reduced up to 0.37, while Table1 MCS13’s can be reduced up to 0.33, providing slightly better results.Source: B. Results with fixed MCS · p. 6
  • At 30 m, a lower MCS index gave lower delay because retransmissions began earlier with the higher index; at 50 m, MCS11 exceeded channel capacity.The 30 m delay effect was observed with both HARQ-CC and HARQ-IR.Source: B. Results with fixed MCS · p. 6

C. Results with adaptive MCS

  • Through 70 m, adaptive MCS with HARQ combining yielded no application-layer packet losses, while PHY failures at 70 m were recovered; fixed MCS could not serve the UE there.The adaptive results are reported for error-model and Shannon-bound link adaptation; fixed-MCS failure occurred because the transmission rate exceeded channel capacity.Source: C. Results with adaptive MCS · p. 6, B. Results with fixed MCS · p. 6, Table V · p. 7
  • The Shannon-bound approach selected lower MCSs and had 0% PHY packet losses, while the error-model approach had 0%–4% failures and slightly lower delay at 30 m.MCS selection was independent of HARQ method; the error-model target was transport BLER at most 0.1.Source: C. Results with adaptive MCS · pp. 6–7

V. CONCLUSIONS

The paper presents and calibrates an EESM-based NR PHY abstraction model, then integrates it into an ns-3 NR system-level simulator.

V. CONCLUSIONS

  • The model represents NR channel coding, LDPC base-graph selection, code-block segmentation, both MCS tables, and HARQ combining methods.An NR-compliant link-level simulator calibrated EESM and generated SINR-BLER lookup tables across MCS and resource-allocation settings.Source: §V · p. 7
  • Calibration showed that the L2SM is insensitive to NR numerology, which the authors say generalizes the model to all NR numerologies and frequency ranges.Source: §V · p. 7
  • The model was integrated into an ns-3 NR system-level simulator and made openly available to support reproducible research and collaborative model development.Source: §V · p. 7
Loading 2001.10309v2…