TY - GEN
T1 - Low Complexity OFDM-Guided DJSCC for Multipath Fading Channels using Tensor Train Decomposition with Fine-Tuning
AU - Xu, Man
AU - Lam, Chan Tong
AU - Liang, Yuanhui
AU - Ng, Benjamin K.
AU - Im, Sio Kei
N1 - Publisher Copyright:
© 2023 IEEE.
PY - 2023
Y1 - 2023
N2 - We proposed a low complexity OFDM-Guided deep joint source-channel coding (DJSCC) model for wireless multipath fading channels using fine-tuned tensor train (TT) decomposition. Using the optimal rank of the TT decomposition, we fine-tuned the small number of intermediate channels that lead to incomplete feature extraction after the TT decomposition. The original OFDM-guided DJSCC model is first hierarchically TT decomposition, followed by fine-tuning the size of the intermediate convolution layers to improve the performance. From the experimental results, we found that the smaller the signal-to-noise ratio (SNR) value, the smaller the loss of peak signal-to-noise ratio (PSNR) and the structural similarity index measure (SSIM) performance. The best compressed model using TT decomposition with fine-tuning loses about 4% of PSNR performance and 2% of the SSIM performance, with a compression ratio of about 17% and a reduction of the number of parameters by 6 times of the original OFDM-guided DJSCC model.
AB - We proposed a low complexity OFDM-Guided deep joint source-channel coding (DJSCC) model for wireless multipath fading channels using fine-tuned tensor train (TT) decomposition. Using the optimal rank of the TT decomposition, we fine-tuned the small number of intermediate channels that lead to incomplete feature extraction after the TT decomposition. The original OFDM-guided DJSCC model is first hierarchically TT decomposition, followed by fine-tuning the size of the intermediate convolution layers to improve the performance. From the experimental results, we found that the smaller the signal-to-noise ratio (SNR) value, the smaller the loss of peak signal-to-noise ratio (PSNR) and the structural similarity index measure (SSIM) performance. The best compressed model using TT decomposition with fine-tuning loses about 4% of PSNR performance and 2% of the SSIM performance, with a compression ratio of about 17% and a reduction of the number of parameters by 6 times of the original OFDM-guided DJSCC model.
KW - deep joint source-channel coding
KW - multipath fading channel
KW - orthogonal frequency division multiplexing
KW - tensor train decomposition
UR - http://www.scopus.com/inward/record.url?scp=85174744221&partnerID=8YFLogxK
U2 - 10.1109/ICSIP57908.2023.10270968
DO - 10.1109/ICSIP57908.2023.10270968
M3 - Conference contribution
AN - SCOPUS:85174744221
T3 - 2023 8th International Conference on Signal and Image Processing, ICSIP 2023
SP - 853
EP - 859
BT - 2023 8th International Conference on Signal and Image Processing, ICSIP 2023
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 8th International Conference on Signal and Image Processing, ICSIP 2023
Y2 - 8 July 2023 through 10 July 2023
ER -