Semantic Communication Latency-Fidelity Tradeoff in Image Transmission
Research published on arXiv (2605.05514) investigates semantic image transmission through block Rayleigh fading combined with AWGN, utilizing a multi-task semantic autoencoder. This autoencoder not only reconstructs images but also forecasts labels from the incoming waveforms. The latent dimension acts as a control variable across layers, influencing both semantic accuracy and channel resource allocation. While larger latent representations enhance inference precision, they also lead to longer service times, increased channel usage, and heightened queueing delays. To balance this tradeoff, online semantic-rate controllers adjust the latent dimension with each update.
Key facts
- arXiv paper 2605.05514 studies semantic image transmission over block Rayleigh fading with AWGN.
- A multi-task semantic autoencoder jointly reconstructs images and predicts labels.
- Latent dimension is a cross-layer control variable for semantic fidelity and channel resource usage.
- Larger latent representations improve inference accuracy but increase delay.
- Online semantic-rate controllers adapt latent dimension per update.
Entities
Institutions
- arXiv