CuiHaoXuan/ns3-oran

A module that can be used to model and simulate O-RAN-like behavior in ns-3.

ORAN ns-3 Module

A module that can be used in ns-3 to simulate and model behavior based on the O-RAN specifications.

Table of Contents

• Project Overview
  • Contact
• Model Description
• Features
• Minimum Requirements
  • Optional Dependencies
• Installation
  • Clone (Recommended)
  • Download ZIP
  • Connecting the Module Quickly
    • Cmake
    • Code
  • Connecting the Module Safely
    • Cmake
    • Code
• Updating
  • Clone
  • ZIP
• ML Support
  • ONNX
  • PyTorch
• Documentation
• Running the Examples
  • Random Walk Example
  • LTE to LTE Distance Handover Example
  • LTE to LTE Distance Handover With Helper Example
  • LTE to LTE Distance Handover With LM Processing Delay Example
  • LTE to LTE Distance Handover With LM Query Trigger Example
  • Keep-Alive Example
  • Data Repository Example
  • Multiple Network Devices Example
  • LTE to LTE ML Handover Example

Project Overview

This project has been developed by the National Institute of Standards and Technology (NIST) Communications Technology Lab (CTL) Wireless Networks Division (WND).

This project includes open source, third party dependencies. For details of the licenses of these dependencies see THIRD_PARTY_LICENSES.md

Certain equipment, instruments, software, or materials, commercial or non-commercial, are used in this project. Such usage is not intended to imply recommendation or endorsement of any product or service by NIST, nor is it intended to imply that the software, materials, or equipment identified are necessarily the best available for the purpose.

This project is considered feature complete, and will be maintained on a 'best effort' basis.

Contact

To report a bug, please open a GitHub Issue. The point of contact for this project is Evan Black (evan.black@nist.gov)

Model Description

The oran module for ns-3 implements the classes required to model a network architecture based on the O-RAN specifications. These models include a RAN Intelligent Controller (RIC) that is functionally equivalent to O-RAN's Near-Real Time (Near-RT) RIC, and reporting modules that attach to simulation nodes and serve as communication endpoints with the RIC in a similar fashion as the E2 Terminators in O-RAN.

These models have been designed to provide the infrastructure and access to data so that developers and researchers can focus on implementing their solutions, and minimize the time and effort spent on handling interactions between models. With this in mind, all the components that contain logic that may be modified by end users have been modeled hierarchically (so that parent classes can take care of common actions and methods and leave child models to focus on the logic itself), and at least one example is provided, to serve as reference for new models.

The RIC model uses a data repository to store all the information exchanged between the RIC and the modules, as well as to serve as a logging endpoint. This release provides an SQLite storage backend for the data repository. The database file is accessible after the simulation and can be accessed by any SQLite-compatible tool and interface.

Modeling of the reporting and communication models for the simulation nodes has been implemented using existing traces and methods, which means there is no need to modify the models provided by the ns-3 distribution to make use of the full capabilities of this module.

Features

This release of the oran module contains the following features:

• Near-RT RIC model, including:
  • Data access API independent of the data repository backend
  • SQLite database repository implementation for Reports, Commands, and logging
  • Support for Logic Modules that serve as O-RAN's xApps
  • Separation of Logic Modules into default (only one, mandatory) and additional (zero to many, optional)
  • Support for addition and removal of Logic Modules during the simulation
  • Conflict Mitigation API
  • Logic Module and Conflict Mitigation logic logging to the data repository
  • Periodic invocation of the Logic Module's algorithms
• Periodic reporting of metrics from the simulation nodes to the Near-RT RIC
• Reporting capabilities for node location (any simulation node) and LTE cell attachment (LTE UE nodes)
• Generation and execution of LTE-to-LTE handover Commands
• Activation and deactivation for individual components and RIC
• Periodic node registration and de-registration with the RIC
• Integration with ONNX Runtime and PyTorch to support Machine Learning (ML)

Minimum Requirements

• ns-3.41
• SQLite 3.7.17

Optional Dependencies

• ONNX Runtime 1.14.1
• PyTorch 2.2.2

Installation

Clone (Recommended)

Clone the project into a directory called oran in the contrib directory of a supported version of ns-3.

1. cd into the contrib directory of ns-3

cd contrib/

2. Clone the project from one of the below URLs

# Pick one of the below
# HTTPS (Choose this one if you're uncertain)
git clone https://github.com/usnistgov/ns3-oran.git oran

# SSH
git clone git@github.com:usnistgov/ns3-oran.git oran

3. (Re)configure and (Re)build ns-3

# --enable-examples is optional, see `Running the Examples`
# for how to run them
./ns3 configure --enable-examples
./ns3

Download ZIP

If, for whatever reason, git is not available, download the project and unzip it into the contrib directory of ns-3.

Note: Updates will have to be performed manually using this method.

1. Download the ZIP of the project from the url below

wget https://github.com/usnistgov/ns3-oran/archive/refs/heads/master.zip

2. Unzip the file into the ns-3 contrib/ directory

unzip ns3-oran-master.zip

3. Rename the resulting directory to oran, as ns-3 will not accept a module named differently than its directory

mv ns3-oran-master oran

Connecting the Module Quickly

If you are linking your module/program to the oran module add the following to your CMakeLists.txt(CMake).

CMake (Quick Connect)

For CMake, add oran's target ${liboran} to your libraries_to_link list.

# Example
build_lib_example(
  NAME your-example-name
  SOURCE_FILES example.cc
  LIBRARIES_TO_LINK
    ${liboran}
    # ...
)

# Module
build_lib(
  LIBNAME your-module-name
  SOURCE_FILES
    # ...
  HEADER_FILES
    # ...
  LIBRARIES_TO_LINK
    ${liboran}
    # ...
)

Connecting the Module Safely

You may wish for your module to not have a hard dependency on the oran module. The following steps will link the oran module, but still allow you to build and run your module without the oran module present.

CMake (Safe Connect)

Check for oran in the ns3-all-enabled-modules list to confirm that the module is present.

Note: If the module that links to the oran module is in the src/ directory, then you will need to add the ENABLE_ORAN C++ define yourself when you check for the presence of the module.

# Create a list of your required modules to link
# 'core' and 'mobility' are just examples here
  set(libraries_to_link "${libcore};${libmobility}")

# Check if the `oran` module is in the enabled modules list
if("oran" IN_LIST ns3-all-enabled-modules)
  # If it is there, then it is safe to add it to the library list
  list(APPEND libraries_to_link ${liboran})

  # N.B. if the module you are linking to is in the `src/` directory
  # of ns-3, then (at least for now), you must also add the C++ define
  # yourself, like this.
  #
  # There is no harm in repeated definitions of the same value, so there is no
  # need to guard this statement
  add_definitions(-DENABLE_ORAN)
endif()

# Use the `libraries_to_link` list as your dependency list

# Module
build_lib(
  LIBNAME your-module 
  SOURCE_FILES
    # ...
  HEADER_FILES
    # ...
  LIBRARIES_TO_LINK
    ${libraries_to_link}
)

# Example
build_lib_example(
  NAME your-scenario
  SOURCE_FILES scenario.cc
  LIBRARIES_TO_LINK
    ${libraries_to_link}
)

Code (Safe Connect)

In addition to the variable in the build environment, the module also defines a C++ macro also named ENABLE_ORAN. This macro may be used in C++ code to check for the presence of the oran module.

Note: If you are using CMake and the module is in the src/ directory, you may have to add this definition yourself (scratch/ and module examples are fine). See the CMake build system section for more information.

// Guard the include with the macro
#ifdef ENABLE_ORAN
#include <ns3/oran-module.h>
#endif

// ...

int main ()
{
  // ...

  // Guard any oran references in code with the macro as well
#ifdef ENABLE_ORAN
  auto oranHelper = CreateObject<OranHelper> ();
  // ...
#endif
}

Updating

Clone

To update the cloned module, move to the module's root directory and perform a git pull.

# From the ns-3 root
cd contrib/oran
git pull

ZIP

To update a ZIP installation, remove the old module and replace it with the updated one.

# From the ns-3 root
cd contrib
rm -Rf oran

# Use this command, or download manually
wget https://github.com/usnistgov/ns3-oran/archive/refs/heads/master.zip \
  -O ns3-oran-master.zip
unzip ns3-oran-master.zip

# Make sure the directory in the ns-3 contrib/ directory is named `oran`
mv ns3-oran-master oran

ML Support

ML is not required to use this module, however, we provide a means of integration for both ONNX and PyTorch so that it is possible to use ML to make inferences when running simulations. Therefore, users who wish to simulate O-RAN based solutions that leverage ML may do so using this module. It does however require the extra step of making at least one of these libraries accessible to our module so that we can link and compile against it. This also means that while these tools may be installed and accessible system wide, this is not a requirement as it is possible to have a locally compiled version of the library and source files in a user's working space. Please note that when using these tools with our module, the expectation is that the user will have already trained and created an ML model outside of the O-RAN simulations, but once this is done, that model may be used via the ONNX or PyTorch C++ API from the simulation. We provide a class for each tool to demonstrate the use of both ONNX and PyTorch, respectively:

• OranLmLte2LteOnnxHandover
• OranLmLte2LteTorchHandover

OranLmLte2LteOnnxHandover is defined by the files:

• 'model/oran-lm-lte-2-lte-onnx-handover.h'
• 'model/oran-lm-lte-2-lte-onnx-handover.cc'

OranLmLte2LteTorchHandover is defined by the files:

• 'model/oran-lm-lte-2-lte-torch-handover.h'
• 'model/oran-lm-lte-2-lte-torch-handover.cc'

There is also the LTE to LTE ML Handover Example that we will discuss later that demonstrates the use of these two classes using existing ML models included with this module.

ONNX

At the time that this documentation was created, not very many linux distributions provide ONNX packages. However, the use of ONNX for this module is attractive since models that are created using other tools, such as PyTorch, can be exported to ONNX. Therfore, integration with ONNX was desired with the hopes that it can be used to provide the flexibility needed to support more than one type of ML model without having to integrate each individually. To make use of ONNX with this module, one simply needs to download the ONNX libraries that are distributed on the ONNX webiste (https://onnxruntime.ai/), and export the location of the extracted library to the LIBONNXPATH environment variable. For example,

# Change to home directory
cd ~

# Download the library files
wget "https://github.com/microsoft/onnxruntime/releases/download/v1.14.1/onnxruntime-linux-x64-1.14.1.tgz"

# Extract the library files
tar xzf onnxruntime-linux-x64-1.14.1.tgz

# Create environment variable with library location so that cmake knows where
# to find it
export LIBONNXPATH="/home/$(whoami)/onnxruntime-linux-x64-1.14.1"

At this point, the user should be able to navigate to their working directory of ns-3 and run

./ns3 configure

The output of this command should include the text "find_external_library: OnnxRuntime was found," indicating that the library and necessary source files were discovered.

PyTorch

PyTorch is widely available through most linux distribution package bases. Therefore, a user should simply be able to install the desired "python-pytorch" package with no further steps being required. However, if the user cannot or does not wish to install the package, one simply can download the PyTorch libraries that are distrubited on the PyTorch webiste (https://pytorch.org/), and export the location of the extracted library to the LIBTORCHPATH environment variable. For example,

# Change to home directory
cd ~

# Download the library files
wget "https://download.pytorch.org/libtorch/cpu/libtorch-cxx11-abi-shared-with-deps-2.2.2%2Bcpu.zip"

# Extract the library files
unzip libtorch-cxx11-abi-shared-with-deps-2.2.2+cpu.zip

# Create environment variable with library location so that cmake knows where
# to find it
export LIBTORCHPATH="/home/$(whoami)/libtorch"

At this point, the accessibility of the library can be verified by navigating to the working directory of ns-3 and running the following command.

./ns3 configure

The output of this command should include the text "find_external_library: Torch was found," indicating that the library and necessary source files were discovered.

Documentation

Sphinx is required to build the documentation.

To run Sphinx to build the documentation, cd into the doc directory in the module and run make [type] for the type of documentation you wish to build.

# From the ns-3 root directory
cd contrib/oran/doc

# HTML (Several Pages)
make html

# HTML (One Page)
make singlehtml

# PDF
make latexpdf

# To list other options, just run make
make

The built documentation can now be found in doc/build/[type].

Running the Examples

Listed below are the commands to run the examples provided with the module.

Random Walk Example

Example with a very simple topology in which ns-3 nodes move randomly and periodically report their position to the Near-RT RIC.

./ns3 run "oran-random-walk-example --verbose"

LTE to LTE Distance Handover Example

A complex scenario that shows how to configure and deploy each of the models manually, without using the oran's helper. This scenario consists of 2 LTE eNBs and 1 LTE UE that moves back and forth between both eNBs. The UE is initially attached to the closest eNB, but as it moves closer to the other eNB, the Logic Module in the RIC will issue a handover Command and the UE will be handed over to the other eNB.

./ns3 run "oran-lte-2-lte-distance-handover-example"

LTE to LTE Distance Handover With Helper Example

Functionally the same scenario as the LTE to LTE Distance Handover Example, however, in this scenario the helper is used to configure and deploy the models.

./ns3 run "oran-lte-2-lte-distance-handover-helper-example"

LTE to LTE Distance Handover With LM Processing Delay Example

Similar to the LTE to LTE Distance Handover With Helper Example, however, in this scenario the Logic Module is configured with a processing delay.

./ns3 run "oran-lte-2-lte-distance-handover-lm-processing-delay-example"

LTE to LTE Distance Handover With LM Query Trigger Example

Similar to the LTE to LTE Distance Handover With Helper Example example, however, in this scenario the Near-RT RIC is configured with a custom Query Trigger (provided in the scenario) that may initiate the Logic Module querying process as soon as Reports with certain characteristics are received by the Near-RT RIC.

./ns3 run "oran-lte-2-lte-distance-handover-lm-query-trigger-example"

Keep-Alive Example

This example showcases how the keep-alive mechanism works. A single node moving in a straight line periodically reports its position to the Near-RT RIC. However, not all of these Reports will be accepted by the Near-RT RIC because the configuration of the timing for sending Registration messages in the node means that the node will frequently be marked as ’inactive’ by the Near-RT RIC.

NS_LOG="OranE2NodeTerminator=prefix_time|warn" ./ns3 run "oran-keep-alive-example"

Data Repository Example

This example showcases how the Data Repository API can be used to store and retrieve information. This example performs all these operations from the scenario for simplicity, but the same methods can be used by any model in the simulation that can access the RIC. This will be important when developing custom Logic Modules, and can also be used in scenarios for debugging, testing and validation.

./ns3 run "oran-data-repository-example"

Multiple Network Devices Example

Similar to the LTE to LTE Distance Handover Wth Helper Example this example showcases how a node with several network devices can interact with the Near-RT RIC separately.

./ns3 run "oran-multiple-net-devices-example"

LTE to LTE ML Handover Example

Note that in order to run this example using the flag, --use-onnx-lm, the ONNX libraries must be found during the configuration of ns-3, and it is assumed that the ML model file "saved_trained_model_pytorch.onnx" has been copied from the example directory to the working directory. In order to run this example using the flag, --use-torch-lm, the PyTorch libraries must be found during the configuration of ns-3, and it is assumed that the ML model file "saved_trained_model_pytorch.pt" has been copied from the example directory to the working directory.

This example showcases how pretrained ONNX and PyTorch ML Models can be used to initiate handovers based on location and packet loss data. It consists of four UEs and two eNBs, where UE 1 and UE 4 are configured to move only within coverage of eNB 1 or eNB 2, respectively, while UE 2 and UE 3 move around in an area where the coverage of eNB 1 and eNB 2 overlap. As the simulation progresses with UEs moving and receiving data, the distances of all four UEs as well as the recorded packet loss for each UE are fed to an ML model that returns a desired configuration that indicates which eNB UE 2 and UE 3 should be attached to minimize the overall packet loss. The models that we provide are for demonstration purposes only and have not been thoroughly developed. It should also be noted that "saved_trained_model_pytorch.onnx" is the same trained model as "saved_trained_model_pytorch.pt" only it has been exported to the ONNX format.

./ns3 run "oran-lte-2-lte-ml-handover-example"

Without going into too many details, there is also a script included in the examples folder called, "oran-lte-2-lte-ml-handover-example-generate-training-data.sh," that can be used as a start to generate data using this same example to train an ML model. Furthermore, once the training data has been generated, the file "oran-lte-2-lte-ml-handover-example-classifier.py" that is also included in the example folder, can be used to produce a PyTorch ML model using the training data that is generated.