Performance Evaluation of CDL Channel Models in a 5G NR Downlink End-to-End Simulation

MEng (IIB) Dissertation — University of Cambridge, Department of Engineering (2024–2025)

A MATLAB-based end-to-end 5G NR downlink link-level simulation framework for evaluating Physical Downlink Shared Channel (PDSCH) performance under 3GPP-standardised Clustered Delay Line (CDL) channel models, with results benchmarked against 9 industry vendors.

System Architecture

End-to-end processing chain: DL-SCH encoding → PDSCH → Precoding → CP-OFDM → CDL/TDL Channel → Receiver (timing sync, demodulation, channel estimation, decoding) with CSI feedback and HARQ retransmission loops.

About

This project was completed as a 4th-year Master of Engineering (MEng/IIB) dissertation at the University of Cambridge in collaboration with Nokia Bell Labs. It develops a configurable, 3GPP-compliant simulation framework to evaluate how standardised CDL fading environments — defined in 3GPP TR 38.901 — impact downlink throughput and block error rate (BLER) under realistic conditions including HARQ, imperfect CSI feedback, and practical channel estimation.

The work is motivated by ongoing 3GPP RAN4 standardisation efforts (Release 19, TR 38.753) to define spatial channel models for demodulation performance requirements. Simulation results are compared against industry data from Apple, BT, Ericsson, Huawei, MediaTek, Nokia, Qualcomm, Samsung, and ZTE, providing independent academic validation of vendor-reported performance.

Author: Ian Cho (Pembroke College) Supervisors: Prof. Albert Guillen i Fabregas (Cambridge), Alexander Hamilton (Nokia Bell Labs)

Key Features

3GPP-compliant PDSCH simulation following TS 38.211, TS 38.212, and TS 38.214
CDL channel models (CDL-A through CDL-E) from TR 38.901 with configurable delay spread, angular spread, and Doppler
Up to 8-layer SU-MIMO with Type-1 single-panel and multi-panel codebook precoding
HARQ retransmissions — 8 parallel processes, RV sequence {0, 2, 3, 1}, soft combining with LDPC decoding
CSI feedback modes — RI/PMI/CQI codebook reporting, AI-based CSI compression, and perfect CSI
Channel estimation — perfect (genie-aided) and practical (DM-RS least-squares) estimators
Subarray virtualisation — maps 512 physical antenna elements to 8 CSI-RS virtual ports using beamsteering vectors (3GPP Option 1Y)
Parallel computing — parfor SNR sweeps across up to 31 workers on AWS EC2 (r6i.8xlarge)
Industry benchmarking — results compared against 9 vendors from 3GPP RAN4 meetings #113–#115

Key Results

CDL Channel Profile Comparison

Different CDL profiles produce significantly different throughput characteristics depending on angular spread and deployment scenario:

CDL-C (dense urban NLOS, 300 ns delay spread) yields the highest throughput due to rich angular spread and strong multipath
CDL-E (strong LOS) is limited by single dominant eigenmode at lower SNRs
CDL-A (urban microcell NLOS) underperforms due to wide angular spread misaligning with the simulated narrow-beam configuration

Feature Impact Analysis

Feature	SNR Gain at 30% Throughput	SNR Gain at 70% Throughput
HARQ retransmissions	+2.0 dB	+1.5 dB
Perfect vs. LS channel estimation	+2.0 dB	+2.0 dB
Ideal vs. realistic assumptions	+5.0 dB	+8.0 dB
10 Hz vs. 100 Hz Doppler	+2.0 dB	+5.0 dB

Industry Alignment — 4-Layer, Option 1Y

Throughput vs. SNR comparison for 4-layer MIMO (Option 1Y, CDL-C, 10 Hz Doppler) — our simulation ("This Work") benchmarked against Nokia, Samsung, MediaTek, and Ericsson from 3GPP RAN4 #113–#115.

At 70% throughput, simulation results are concordant with industry averages (within 1.7 dB span)
At 30% throughput, a 5.2 dB divergence is attributed to the absence of ray splitting (a TR 38.753 feature not present in TR 38.901)
Results most closely align with Nokia and MediaTek at low SNR ranges

8-Layer MIMO with Dual Codewords

8-layer dual-codeword throughput comparison (CDL-C, 10 Hz Doppler) — our simulation vs. Nokia, MediaTek, and Ericsson. The characteristic "dipping" in mid-SNR regions is caused by HARQ retransmission dynamics with dual codewords.

CDL Model Architecture

Internal architecture of the 3GPP TR 38.901 CDL model, showing AAV configuration and propagation modelling components.

Project Structure

NokiaIIBProject/
├── README.md                          # This file
├── LICENSE                            # MIT License
├── logbook.xlsx                       # Project logbook / diary
│
├── Final_Report/
│   ├── Cho_Ian_ic404_Final_Report.pdf # Full 53-page dissertation
│   └── Figures/                       # Architecture diagrams and result plots
│
├── PPT_Mich/                          # Michaelmas term progress presentation
├── PPT_Final/                         # Final project presentation
│
└── Simulation/
    ├── README.md                      # Simulation-specific notes
    │
    │   # Main simulation scripts
    ├── pp_throughput_withCSIwithHARQwithsubarray.m   # Final: Option 1Y, 512-element subarray (Section 4.5)
    ├── throughput_withCSIwithHARQ.m                  # Option 3 with HARQ (Sections 4.1.2–4.4)
    ├── throughput_withCSInoHARQ.m                    # Baseline without HARQ (Section 4.1.2)
    ├── pp2_throughput_withCSIwithHARQ.m              # 8-layer dual-codeword variant (Section 4.4)
    ├── pp3_throughput_withCSIwithHARQ.m              # TDL channel variant with TDL compatibility
    ├── plotPerformanceMetrics.m                      # Post-processing and plotting
    │
    ├── HARQEntity.m                   # HARQ process state machine
    ├── virtualizeChannel.m            # Subarray virtualisation (512 → 8 ports)
    ├── hCSIEncode.m / hCSIDecode.m    # CSI feedback encoding/decoding
    ├── hDLPMISelect.m / hDLPMIRandom.m # PMI selection (optimal / random)
    ├── hRISelect.m / hCQISelect.m     # RI and CQI selection
    ├── hMCSSelect.m                   # MCS selection (BLER < 10% target)
    ├── hPrecodedSINR.m                # MMSE SINR calculation
    ├── hArrayGeometry.m               # Antenna array geometry configuration
    ├── hSubbandChannelEstimate.m      # Practical DM-RS channel estimation
    ├── helperCSINet*.m                # AI CSI compression network helpers
    ├── nPMI.m                         # PMI codeword counting (TS 38.214)
    │
    ├── figures/                       # LaTeX/TikZ figure exports (via matlab2tikz)
    └── matlab2tikz-master/            # Third-party MATLAB → TikZ converter

Getting Started

Prerequisites

MATLAB R2024a or later
Required toolboxes:
- 5G Toolbox
- Communications Toolbox
- Phased Array System Toolbox
- Parallel Computing Toolbox (optional, for parfor acceleration)

Running the Simulation

Clone the repository:

git clone https://github.com/ianwh02/5G-NR-CDL-Simulation.git
cd 5G-NR-CDL-Simulation/Simulation

Open MATLAB and navigate to the Simulation/ directory.

Run the main script — choose based on your scenario:

Script	Scenario	Dissertation Section
`pp_throughput_withCSIwithHARQwithsubarray.m`	Final — Option 1Y, 512-element subarray, 4-layer	4.5
`throughput_withCSIwithHARQ.m`	Option 3 with HARQ, CSI, channel estimation	4.1.2–4.4
`throughput_withCSInoHARQ.m`	Baseline without HARQ (for measuring HARQ impact)	4.1.2
`pp2_throughput_withCSIwithHARQ.m`	8-layer dual-codeword, imperfect estimator, 50 frames	4.4
`pp3_throughput_withCSIwithHARQ.m`	TDL-C channel variant	—

Configure parameters at the top of the script (see Configuration below).
Plot results after simulation completes:
```
plotPerformanceMetrics
```
This loads saved .mat files from Simulation/results/ and generates throughput/BLER vs. SNR plots.

Configuration

Key parameters configurable at the top of each simulation script:

Parameter	Default	Description
`NFrames`	`1`	Number of 10 ms radio frames (use 50+ for statistically reliable results)
`SNRIn`	`25`	SNR range in dB (e.g., `-5:1:25` for a full sweep)
`PDSCH.NumLayers`	`4`	Number of MIMO layers (2, 4, or 8)
`DelayProfile`	`'CDL-C'`	Channel model (`'CDL-A'` through `'CDL-E'`, or `'TDL-A'` through `'TDL-E'`)
`DelaySpread`	`300e-9`	RMS delay spread in seconds
`MaximumDopplerShift`	`10`	Doppler frequency in Hz (10 = 3 km/h, 100 = 30 km/h at 3.5 GHz)
`PerfectChannelEstimator`	`true`	`true` for genie-aided, `false` for DM-RS least-squares
`CSIReportMode`	`'RI-PMI-CQI'`	CSI feedback mode (`'RI-PMI-CQI'`, `'AI CSI compression'`, `'Perfect CSI'`)
`CSIReportConfig.CodebookType`	`'Type1MultiPanel'`	Codebook type (`'Type1SinglePanel'`, `'Type1MultiPanel'`, `'Type2'`)
`CSIReportConfig.PMIModeOverride`	`'random'`	PMI selection strategy (`'best'`, `'random'`, `'fixed'`)
`TransmitAntennaArray.Size`	`[8 4 2 8 1]`	gNB antenna array dimensions [M, N, P, Mg, Ng]
`ReceiveAntennaArray.Size`	`[2 1 2 1 1]`	UE antenna array dimensions

Dissertation

The full 53-page dissertation is available at Final_Report/Cho_Ian_ic404_Final_Report.pdf.

Chapters:

Introduction — 5G NR background, motivation, and project scope
Background & Literature Review — Physical layer, MIMO techniques, CDL models, and industry context
Methodology — Simulation setup, CDL architecture, configuration parameters, subarray virtualisation
Results & Discussion — Framework validation, CDL profile comparison, Doppler analysis, MIMO layer scaling, industry alignment
Conclusion — Contributions, limitations, and future work

Acknowledgements

This project was carried out in collaboration with Nokia Bell Labs. Special thanks to:

Prof. Albert Guillen i Fabregas (University of Cambridge) — academic supervisor
Alexander Hamilton (Nokia Bell Labs) — industrial supervisor
Tugce Kobal (Nokia Bell Labs) — technical advisor
Kelvin, Krish, and Tian — fellow project collaborators

License

This project is licensed under the MIT License — see the LICENSE file for details.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

Performance Evaluation of CDL Channel Models in a 5G NR Downlink End-to-End Simulation

System Architecture

About

Key Features