5G Network Simulation and Resource Allocation Optimization Using Machine Learning

---

Mentor

Bhupendra Kumar

Team Members

Mamindla Sathvika: [email protected]

Akash Maurya: [email protected]

---

Table of Contents

1. Project Overview
2. Objectives
3. System Architecture
4. Features
5. Installation
6. Configurations
7. Machine Learning Approach
8. Results and Analysis
9. Troubleshooting and Errors
10. Contributors
11. Video Drive Link

---

Project Overview

This project involves building a 5G testbed using open-source platforms like *Open5GS, **Free5GC, and **UERANSIM, alongside implementing *Machine Learning (ML) models to optimize resource allocation. The project aims to simulate real-world 5G network scenarios and apply ML techniques to enhance bandwidth usage, reduce latency, and improve resource efficiency.

---

Video

https://drive.google.com/drive/folders/1rS93xvHYyijNT6-3-vuyck3b1cC-rKhO?usp=sharing

---

Objectives

1. To establish a fully functional 5G standalone (SA) core network using Open5GS and Free5GC.
2. To simulate radio access and user equipment (UE) using UERANSIM.
3. To analyze network performance metrics such as latency, signal strength, and bandwidth usage.
4. To implement and compare *Linear Regression, **Polynomial Regression, and *XGBoost for resource allocation optimization.
5. To address and document troubleshooting and errors encountered during implementation.

---

System Architecture

The system architecture is composed of:

• 5G Core Network: Implemented using Open5GS and Free5GC, including AMF, SMF, NRF, UDM, and AUSF.
• Radio Access Network: Simulated using UERANSIM, creating virtual gNodeB and UE instances.
• Database: MongoDB is used to store subscriber information (IMSI, keys, etc.).
• Machine Learning Component: Python-based ML models (Linear Regression, Polynomial Regression, and XGBoost) are used to optimize resource allocation.

---

Features

• Simulation of real-world 5G core network and radio access using open-source tools.
• UE registration, PDU session management, and network slice configuration.
• Detailed resource allocation optimization using ML models.
• Comprehensive troubleshooting and error documentation.

---

Installation

Prerequisites

• Operating System: Ubuntu 20.04 or later.
• Dependencies: MongoDB, Python 3.8+, GCC, CMake, SCTP libraries.
• Tools: Open5GS, Free5GC, UERANSIM.

---

Go through the 2 PDFs uploaded for detailed installation steps for Open5gs and Free5gs. Download the IPYNB notebook to run the ML complete end to end Resource Allocation in 5G Project

---

Machine Learning Approach

The project utilized three models to optimize resource allocation within the 5G network:

1. Linear Regression:

   • Used to establish baseline predictions.
   • Performance: R² = 0.19

2. Polynomial Regression:

   • Captures non-linear relationships in network parameters.
   • Performance: R² = 0.59

3. XGBoost:

   • Final and most effective model.
   • Achieved R² = 0.71

Dataset Features:

The dataset included the following key metrics:

• Timestamp: Time of metric recording.
• Signal Strength (dBm): Measured power level at the receiver.
• Latency (ms): Round-trip time between UE and the server.
• Bandwidth Usage (Mbps): Data transmission rate.
• Active Users: Number of concurrent users.

---

Results and Analysis

1. Successfully set up *Open5GS, **Free5GC, and *UERANSIM for emulating the 5G core and radio access networks.
2. Achievements in ML-based resource allocation optimization:
   • Linear Regression R²: 0.19
   • Polynomial Regression R²: 0.59
   • XGBoost R²: 0.71
3. Detailed error analysis and troubleshooting of issues in Open5gs and Free5gs such as:
   • NAS registration errors ([SEMANTICALLY_INCORRECT_MESSAGE]).
   • SCTP connection issues.
   • RRC connection drops.

---

Troubleshooting and Errors

Common Issues and Fixes:

1. SCTP Connection Issues:
   • Resolved by installing libsctp-dev and ensuring proper SCTP endpoint configuration.
2. NAS Registration Failures:
   • Corrected IMSI and key values in MongoDB subscriber entries.
3. RRC Connection Drops:
   • Fixed inconsistencies in tac and gnbSearchList settings in ueransim-gnb.yaml.

For a comprehensive list of troubleshooting steps and solutions, refer to the 2 pdf's uploaded.

---

Acknowledgments

We extend our gratitude to the open-source communities behind *Open5GS, **Free5GC, and *UERANSIM for their excellent tools and documentation, which made this project possible. And also this project would have not been possible without the guidance of our professor.

---