Cognitive Semantic Augmentation (CSA) LEO Satellite Networks for Earth Observation

This is a Pytorch implementation of CSA as proposed in the paper Cognitive Semantic Augmentation (CSA) LEO Satellite Networks for Earth Observation\

Requirements

The codes are compatible with the packages:

• pytorch 1.8.0

• torchvision 0.9.0a0

• numpy 1.23.1

• tensorboardX 2.4

The code can be run on the datasets EuroSAT

Run experiments

Training the CSA model

python train2.py

Evaluating the trained CSA model or federated learning

python3 evaluate_fl.py --mod apsk --dataset EuroSAT --latent_d 32 --save_root ./results-fl --name EuroSAT-num_e16-latent_d32-modapsk-psnr12.1-lam0.05

Block diagram of CSA LEO

TXRX diagram of CSA

Top1 accuracy of confusion matrix using DTJSCC based on 16APSK Rician channel where PSNR=12dB and K=128.

Top1 accuracy of CSA satellite networks using DT-JSCC over 16APSK LEO Rician channel while DT-JSCC training at 4dB and CSA/federated learning training at 12dB

Top1 accuracy comparison of CSA and non-CSA DT-JSCC K=32 systems while PSNR=12dB and 16APSK over LEO Rician channel.

Top1 accuracy using DT-JSCC based on 16APSK LEO Rician and LEO Rayleigh channel

Citiation

@article{chou2024cognitivesemanticaugmentationleo,
      title={Cognitive Semantic Augmentation LEO Satellite Networks for Earth Observation}, 
      author={Hong-fu Chou and Vu Nguyen Ha and Prabhu Thiruvasagam and Thanh-Dung Le and Geoffrey Eappen and Ti Ti Nguyen and Duc Dung Tran and Luis M. Garces-Socarras and Juan Carlos Merlano-Duncan and Symeon Chatzinotas},
      year={2024},
      eprint={2410.21916},
      archivePrefix={arXiv},
      primaryClass={cs.NI},
      url={https://arxiv.org/abs/2410.21916}, 
}