Dissagregated NGRAN in 5G NTN networks

This repo is intended for the work "LSTM-based Resource Prediction for Disaggregated RAN in 5G Non-Terrestrial Networks"

Build docker images for Free5GC core network:

Build Free5GC base image from directory: ./build_docker_images/f5gc_base

In the directory: ./build_docker_images/f5gc_core BUild the docker images with the tag defined in the docker-compose.yaml file of the same directory.

Or instead of building all the components, you can also pull all core network functions:

docker pull henok28/f5gc-amf:v3.1.1

Build satellite emulator : opensand

The satellite emulator containers will build upon runing the docker-compose.yaml file from directory: ./build_docker_images/opensand

Three containers: Gateway, Satellite, and Terminal will be in runing state

Pull dissagregated NGRAN images: Openairinterface5G

docker pull oaisoftwarealliance/oai-gnb:develop

docker pull oaisoftwarealliance/oai-nr-ue:develop

Build Prometheus docker image from ./prometheus/Dockerfile

docker build -t custom-prometheus .

Build Grafana docker image from: ./grafana/Dockerfile

docker build -t custom-grafana .

Build CAdvisor docker image from: ./cadvisor/Dockerfile

docker build -t custom-cadvisor .

There are two architectures for F1 split and F1_E1 split

Figure 1: Emulated 5G NTN with F1 split

Architecture one with F1 split across the 5GNTN network

Figure 1: Emulated 5G NTN with F1_E1 split

Architecture one with F1_E1 split across the 5GNTN network

Start simulating the network:

Start CAdvisor container: listens at port 8080

      http://<your-host-ip>:8080   #### url for CAdvisor

Start Prometheus container: listens at port 9090

      http://<your-host-ip>:9090      #### url for Prometheus

Start Grafana container: listens at port 3000

      http://<your-host-ip>:3000     #### url for Grafana

make sure the three monitoring tools are configured properly with their IP

Start the first network for architecture one:

docker compose -f docker-compose_with_F1_split.yaml up -d

Generate a video traffic using iperf3 and run the network for 10 hours

Collect resource consumption of gNB-DU using CAdvisor, Prometheus and visuallize using Grafana

           The collected data is located in ./LSTM_data/f1 directory

Repeat the same procedure for the second architecture

Prepare the data train, test and validate the LSTM model

Results:

The code for the LSTM model taining and validation are:

f1.ipynb for the first architecture

Figure 1: Training Vs validation accuracy for the first architecture

f1_e1.ipynb for the second architecture

Figure 1: Training Vs validation accuracy for the second architecture