This is the code for paper for OFDM-guided Deep Joint Source Channel Coding for Wireless Multipath Fading Channels, which is a journal version of Deep Joint Source Channel Coding for WirelessImage Transmission with OFDM.
CIFAR-10 and CIFAR-100 will be downloaded after the first run.
CelebA can be downloaded here and should be placed at dataset/celeba/CelebA_train and dataset/celeba/CelebA_test.
OpenImage can be donwloaded here.
Please edit the files under configs to test different schemes.
__C.name = 'JSCC_OFDM' # Name of the experiment
__C.gpu_ids = [1] # GPUs to use
__C.dataset_mode = 'OpenImage' # ['CIFAR10', 'CIFAR100', 'CelebA', 'OpenImage']
__C.checkpoints_dir = './Checkpoints/' + __C.dataset_mode # Path to store the model
__C.model = 'JSCCOFDM'
__C.C_channel = 12 # Number of channels for output latents (controls the communication rate)
# Calculation of the rate (channel usage per pixel):
# C_channel / (3 x 2^(2 x n_downsample + 1))
__C.SNR = 5 # Signal to noise ratio
__C.SNR_cal = 'ins' # ['ins', 'avg']. 'ins' is for instantaneous SNR, 'avg' is for average SNR
__C.feedforward = 'OFDM-CE-sub-EQ-sub' # Different schemes:
# OFDM-CE-EQ: MMSE channel estimation and equalization without any subnets
# OFDM-CE-sub-EQ: MMSE channel estimation and equalization with CE subnet
# OFDM-CE-sub-EQ-sub: MMSE channel estimation and equalization with CE & EQ subnet
# OFDM-feedback: pre-coding scheme with CSI feedback
__C.N_pilot = 1 # Number of pilot symbols
__C.is_clip = False # Whether to apply signal clipping or not
__C.CR = 1.2 # Clipping ratio if clipping is applied
__C.lam_h = 50 # Weight for the channel reconstruction loss
__C.gan_mode = 'none' # ['wgangp', 'lsgan', 'vanilla', 'none']
__C.lam_G = 0.02 # Weight for the adversarial loss
__C.lam_L2 = 100 # Weight for image reconstruction loss
Run train.py to train the model. The trained model will be saved under Checkpoints.
Run test.py to test the model. Some recovered images will be saved under Images.
@misc{yang2021ofdmguided,
title={OFDM-guided Deep Joint Source Channel Coding for Wireless Multipath Fading Channels},
author={Mingyu Yang and Chenghong Bian and Hun-Seok Kim},
year={2021},
eprint={2109.05194},
archivePrefix={arXiv},
primaryClass={eess.SP}
}
@INPROCEEDINGS{9500996,
author={Yang, Mingyu and Bian, Chenghong and Kim, Hun-Seok},
booktitle={ICC 2021 - IEEE International Conference on Communications},
title={Deep Joint Source Channel Coding for Wireless Image Transmission with OFDM},
year={2021},
volume={},
number={},
pages={1-6},
doi={10.1109/ICC42927.2021.9500996}
}