Python code for "Joint coding-modulation for digital semantic communications via variational autoencoder"

This repository contains the original code and models for the work Joint Coding-Modulation for Digital Semantic Communications via Variational Autoencoder[1].

[1] Y. Bo, Y. Duan, S. Shao and M. Tao, "Joint Coding-Modulation for Digital Semantic Communications via Variational Autoencoder," in IEEE Transactions on Communications, doi: 10.1109/TCOMM.2024.3386577.

Requirements

• matplotlib==3.7.2
• numpy==1.23.5
• pandas==2.0.3
• scikit_learn==1.3.0
• scipy==1.13.0
• scikit-image==0.21.0
• torch==1.12.1+cu113
• torchvision==0.13.1+cu113
• tqdm==4.65.0

Training & Evaluation

This code implements 4 modulation schemes: BPSK, 4QAM, 16QAM and 64QAM.

For training, run the following command (as an example):

python main.py --mode 'train' --mod_method '64qam' --load_checkpoint 1

For evaluation, run the following command (as an example):

python main.py --mode 'test' --mod_method '64qam' --load_checkpoint 1

Results

The performance results in Fig. 2 and Fig. 6 in [1] are presented below.

1. Accuracy

SNR (dB)	BPSK	4QAM	16QAM	64QAM
18	0.8783	0.8705	0.8723	0.8750
12	0.8713	0.869	0.8765	0.8730
6	0.8771	0.8682	0.8768	0.8688
0	0.8777	0.8726	0.8732	0.8772
-6	0.8683	0.8711	0.8737	0.8738
-12	0.7735	0.8674	0.8810	0.8645
-18	0.4316	0.6148	0.6320	0.6271

2. PSNR (dB)

SNR (dB)	BPSK	4QAM	16QAM	64QAM
18	19.5620	21.4954	23.6771	25.0308
12	19.4385	21.2107	23.4145	24.2522
6	19.1311	20.7429	22.2025	22.5085
0	17.5842	18.7376	19.7264	19.7710
-6	15.2930	16.4231	16.7817	16.9260
-12	13.1355	13.6273	13.6800	13.7743
-18	12.9488	13.0387	13.0450	13.0472