DCS-JSCC: Leveraging Deep Compressed Sensing into JSCC for Wireless Image Transmission
Mohammad Amin Jarrahi, Eirina Bourtsoulatze, Vahid Abolghasemi · 2024
This paper presents a novel approach integrating deep compressed sensing (DCS) into joint source-channel coding (JSCC) for efficient image transmission. Leveraging the capabilities of DCS, the proposed method offers enhanced compression and resilience to channel noise in wireless image transmission systems. A key component of the proposed method is utilising a convolutional neural network (CNN) structure to implement a block-based DCS technique for image compression. The proposed encoder consists of a well-designed CNN-based structure to capture structural information of the input image which is subsequently mapped to a complex-value domain. Finally, the proposed decoder deals with channel noise and reconstructs the original image. Applying the proposed CNN-based sampling matrices and reconstruction capabilities helps the proposed algorithm enhance image compression and reconstruction in wireless transmission systems. The CIFAR-10 and Kodak datasets are used to evaluate the performance of DCS-JSCC, showing a significant improvement in terms of Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM), across different channel Signal-to-Noise Ratios (SNRs) and channel bandwidth values in comparison with state-of-art JSCC frameworks. Experimental evaluations demonstrate the effectiveness of the proposed method in achieving superior compression levels and maintaining image quality under varying channel conditions.