alexf1991/self_attention_oamp_mimo

This repository implements the OAMP-based architecture propsed in the paper "Self-attention for enhanced OAMP detection in MIMO Systems"

Self-attention model for MIMO

This repository implements an enhanced OAMP estimator using a self-attention neural network model.

First steps

Install all dependencies using the "environment.yml" file.

conda env create -f environment.yml

Generate data

python GIT/self-attention-mimo/scripts/generate_data.py --data_dir path/to/data

Training

Training a self-attention model

Set the "data_path" in the config.json file to your desired data path, e.g., "path/to/data/5dB"

python GIT/self-attention-mimo/scripts/train.py --model_dir path/to/model

Training multiple SNR and correlation values

Set the "data_path" in the config.json file to your desired data path including placeholders for the SNR range "XdB" , e.g., "path/to/data/XdB"

python GIT/self-attention-mimo/scripts/train_multi.py --model_dir /path/to/model/XdB/YC

Evaluation

Evaluating multiple SNR and correlation values

Set the "data_path" in the config.json file to your desired data path including placeholders for the SNR range "XdB" , e.g., "path/to/data/XdB"

python  python GIT/self-attention-mimo/scripts/evaluate.py --model_dir path/to/model/XdB/YC/oampsa

Evaluating the generalization over multiple SNR and correlation values for a model trained on a specifc SNR range

Set the "data_path" in the config.json file to your desired data path including placeholders for the SNR range "XdB" , e.g., "path/to/data/XdB"

python GIT/self-attention-mimo/scripts/evaluate_generalization.py --model_dir path/to/model/5dB/YC/oampsa

Citation

You may cite this project as:

put bib citation here!!