.png){kind=icon}
.png){kind=icon}
Return
Zero-Delay Communications: A Predictive Semantic Communication Scheme Based on PCVAE
.png){kind=icon}
PRE
AI
DOI：10.1109/TVT.2025.3606126.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
By eliminating semantic redundancy in source information, semantic communication can achieve performance beyond the Shannon limit. In this paper, we introduce an innovative predictive semantic communication paradigm that utilizes the spatio-temporal correlations inherent in signals, to overcome the intrinsic latency associated with wireless transmission and processing, thus approximating the ideal of zero-delay communication. The methodology initiates with the semantic compression of images using the clustering vector quantization autoencoder semantic compression (CAESC) method, which significantly reduces the data volume transmitted while maintaining minimal impact on reconstruction quality. Subsequently, the scheme employs a convolutional long short-term memory (ConvLSTM) network for predicting image sequences. Empirical simulations show a 48-fold reduction in data transmission volume with minimal loss in reconstruction quality. Furthermore, our predictive methodology significantly improves the learned perceptual image patch similarity (LPIPS), showing about a 43% increase during one constant delay period over conventional approaches.
Keywords:
Zero-delay communication
semantic compression
convLSTM
predicted semantic communication
Journal
IF:
7.1
Papers:
1.8W
Citations:
6.6W