Generative and Explainable AI for High-Dimensional MIMO-OFDM Channel Estimation in Time-Frequency-Space Domain

This work is currently under review for publication in the IEEE Transactions on Green Communications and Networking

PDF

• Manuscript

Pipelines

Installation

Prerequistes

The code use Sionna library, python version Python 3.10.

Note: To use sionna on macOS, you must install the LLVM backend to avoid initialization errors

Setup

# Create new conda enviroment
conda create python=3.10 -n cdl
conda activate cdl

# Install requirements
pip install -r requirements.txt

# initial envs
export PYTHONPATH=`pwd`
export DRJIT_LIBLLVM_PATH="/path/to/libLLVM.dylib"

Simulation

The steps bellow descrice the generate data, training channel estimation.

Dataset

Make sure the configuration setting in src/settings/config.py are correct before generate the channel dataset for training.

python src/ml/gen_data.py

Note: This script also supports generating datasets using real ray-tracing data.

For more information, see Real Scenario

Training

Make sure the configuration setting in src/settings/ml.py are correct before generate the channel dataset for training.

python src/ml/train.py

Evaluating

The evaluation metrics are measured at

• notebooks/eval.ipynb: for the CDL Channel data
• notebooks/eval_rt.ipynb: for the Real data

Notebooks

• notebooks/statistic_gbsm.ipynb: statistic the geometric characteristics of CDL channel model.
• notebooks/statistic_pipeline.ipynb: statistic the OFDM transmission pipeline and visualize evaluation results
• notebooks/data.ipynb: visualize the data information

Real Scenario

The real scenario is using OpenStreetMap for Blender. Follow this tutorial to use Sinonna RT + BLOSM and the demo notebooks/blender.ipynb. To generate the real data channel, you can follow dataset but update the channel configuration to use rt_channel.