SCAN: Semantic Communication with Adaptive Channel Feedback

TL;DR

This paper introduces SCAN, a novel adaptive feedback framework for semantic image communication that improves reliability by allocating resources based on predicted reconstruction quality, using a new metric SDOP.

Contribution

The paper proposes a new reliability metric SDOP and a novel adaptive feedback framework SCAN that enhances semantic communication reliability with reduced feedback overhead.

Findings

SCAN significantly reduces SDOP, improving transmission reliability.

The framework adaptively allocates feedback resources based on image quality predictions.

Experimental results show improved image reconstruction quality with less feedback overhead.

Abstract

In existing semantic communication systems for image transmission, some images are generally reconstructed with considerably low quality. As a result, the reliable transmission of each image cannot be guaranteed, bringing significant uncertainty to semantic communication systems. To address this issue, we propose a novel performance metric to characterize the reliability of semantic communication systems termed semantic distortion outage probability (SDOP), which is defined as the probability of the instantaneous distortion larger than a given target threshold. Then, since the images with lower reconstruction quality are generally less robust and need to be allocated with more communication resources, we propose a novel framework of Semantic Communication with Adaptive chaNnel feedback (SCAN). It can reduce SDOP by adaptively adjusting the overhead of channel feedback for images with different reconstruction qualities, thereby enhancing transmission reliability. To realize SCAN, we first develop a deep learning-enabled semantic communication system for multiple-input multiple-output (MIMO) channels (DeepSC-MIMO) by leveraging the channel state information (CSI) and noise variance in the model design. We then develop a performance evaluator to predict the reconstruction quality of each image at the transmitter by distilling knowledge from DeepSC-MIMO. In this way, images with lower predicted reconstruction quality will be allocated with a longer CSI codeword to guarantee the reconstruction quality. We perform extensive experiments to demonstrate that the proposed scheme can significantly improve the reliability of image transmission while greatly reducing the feedback overhead.