Search papers, labs, and topics across Lattice.
This paper introduces a semantic-level hybrid automatic repeat request (HARQ) framework for text communication using a Transformer-VAE codec as an overlay on conventional protocols. The stochastic encoder generates diverse latent representations across retransmissions, providing incremental knowledge without protocol redesign. At the receiver, a soft quality estimator triggers retransmissions, and a quality-aware combiner merges latent vectors, demonstrating the effectiveness of Weighted-Average or MRC-Inspired combining with self-consistency-based HARQ triggering.
Achieve reliable semantic text communication by retransmitting and combining latent representations, sidestepping the need for explicit protocol design.
Semantic communication conveys meaning rather than raw bits, but reliability at the semantic level remains an open challenge. We propose a semantic-level hybrid automatic repeat request (HARQ) framework for text communication, in which a Transformer-variational autoencoder (VAE) codec operates as a lightweight overlay on the conventional protocol stack. The stochastic encoder inherently generates diverse latent representations across retransmissions-providing incremental knowledge (IK) from a single model without dedicated protocol design. On the receiver side, a soft quality estimator triggers retransmissions and a quality-aware combiner merges the received latent vectors within a consistent latent space. We systematically benchmark six semantic quality metrics and four soft combining strategies under hybrid semantic distortion that mixes systematic bias with additive noise. The results suggest combining Weighted-Average or MRC-Inspired combining with self-consistency-based HARQ triggering for the best performance.
