Generative Semantic HARQ: Latent-Space Text Retransmission and Combining

Bin Han; Yulin Hu; Hans D. Schotten

eess.SPMar 16, 2026

Generative Semantic HARQ: Latent-Space Text Retransmission and Combining

Authors: Bin Han, Yulin Hu, Hans D. Schotten

Organizations: RPTU University Kaiserslautern-Landau · Wuhan University · German Research Center for Artificial Intelligence (DFKI GmbH)

Abstract

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.

Figures & tables

Fig. 1 : The proposed framework of semantic-level HARQ as a lightweight overlay on conventional communication systems.

	Metric		RX overhead
A	VAE Uncertainty:	$q=\max\!\big(0,\,1-\frac{1}{D}\sum_{d}\sigma_{d}\big)$	Re-enc.
B	Self-Consistency:	$q=\cos(\boldsymbol{\mu},\boldsymbol{\mu}^{\prime})\cdot\delta(\boldsymbol{\mu},\boldsymbol{\mu}^{\prime})$	Re-enc.+dec.
C	Decoder Entropy:	$q=\frac{1}{N}\sum_{n}\mathbb{1}[\hat{\mathbf{s}}^{(n)}=\hat{\mathbf{s}}^{(1)}]$	$N{\times}$ dec.
D	Latent Distance:	$q=\max\!\big(0,\,1-\frac{\\|\tilde{\mathbf{z}}-\boldsymbol{\mu}\\|}{\\|\boldsymbol{\mu}\\|+\epsilon}\big)$	Re-enc.
E	Text BLEU:	$q=\text{BLEU-1}(\hat{s}_{1},\hat{s}_{2})$	Re-enc.+dec.
F	Text Similarity:	$q=J(\hat{s}_{1},\hat{s}_{2})$	Re-enc.+dec.

TABLE I : Semantic quality metrics.

	Method		Complexity
A	Weighted-Avg.:	$\hat{\mathbf{z}}=\sum_{k}w_{k}\tilde{\mathbf{z}}_{k}$ , $w_{k}=\frac{e^{q_{k}}}{\sum_{j}e^{q_{j}}}$	$\mathcal{O}(KD)$
B	Best-Only:	$\hat{\mathbf{z}}=\tilde{\mathbf{z}}_{k^{}}$ , $k^{}=\arg\max_{k}q_{k}$	$\mathcal{O}(K)$
C	MRC-Inspired:	$\hat{\mathbf{z}}=\frac{\sum_{k}q_{k}^{2}\tilde{\mathbf{z}}_{k}}{\sum_{k}q_{k}^{2}}$	$\mathcal{O}(KD)$
		$\hat{\mathbf{z}}^{(0)}\!=\!\tilde{\mathbf{z}}_{k^{*}}$ ,
D	Iterative:	$\hat{\mathbf{z}}^{(t+1)}\!=\!\alpha\hat{\mathbf{z}}^{(t)}+(1\!-\!\alpha)\tilde{\mathbf{z}}_{k}$ ,	$\mathcal{O}(KD)$
		$\alpha\!=\!\frac{q^{(t)}}{q^{(t)}+q_{k}}$

TABLE II : Semantic combining methods.

Model		Training
Enc./Dec. layers	6 / 6	Optimizer	AdamW
$d_{\text{model}}$ / $d_{\text{ff}}$	768 / 3072	Learning rate	$10^{-4}$
Attention heads	12	Batch size	16
Latent dim. $D$	256	KL annealing $\beta$	0.01 $\to$ 1.0
Max seq. length	64	Free bits $\lambda_{\text{free}}$	0.25 nats/dim
Dropout	0.1	Word dropout	0.5

TABLE III : Model and training configuration.

Panel of Figure 5 — Fig. 2 : Incremental knowledge gain vs. number of transmission attempts $K$ for aligned and misaligned configurations.

Fig. 3 : Comparison of different semantic soft combining methods.

Panel of Figure 7 — Fig. 4 : Comparison of different semantic quality metrics.

Semantic Communication

Semantic communication sends meaning- or task-relevant information so receivers can act without reconstructing every transmitted bit or signal.

See all 102 papers

Semantic Error Correction and Decoding for Short Block Codes
Apr 24, 2026Jiafu Hao, Chentao Yue, Wanchun Liu +2Semantic Communication Error-Correcting Codes
School of Electrical and Computer Engineering, The University of Sydney, Sydney, NSW 2006, Australia
When Semantic Communication Meets Queueing: Cross-Layer Latency and Task Fidelity Optimization
May 6, 2026Yalin E. Sagduyu, Tugba ErpekSemantic Communication Age of Information
Nexcepta, Gaithersburg, MD, USA
Not All Symbols Are Equal: Importance-Aware Constellation Design for Semantic Communication
May 14, 2026Albert Shaju, Christo Kurisummoottil Thomas, Mayukh Roy ChowdhurySemantic Communication
Department of Electrical and Computer Engineering, Worcester Polytechnic Institute, Worcester, MA, USA. · Nokia Bell Labs, Bengaluru, India.

Apr 24, 2026J/K move · Enter open · S save

Semantic Error Correction and Decoding for Short Block Codes

Jiafu Hao, Chentao Yue, Wanchun Liu, Branka Vucetic+1

School of Electrical and Computer Engineering, The University of Sydney, Sydney, NSW 2006, Australia

Semantic Communication Error-Correcting Codes

Open paper AskPDF arXiv

This paper presents a semantic-enhanced receiver framework for transmitting natural language sentences over noisy wireless channels using multiple short block codes. After ASCII encoding, the sentence is divided into segments, each independently encoded with a short block code and transmitted over an AWGN channel. At the receiver, segments are decoded in parallel, followed by a semantic error correction (SEC) model, which reconstructs corrupted segments using language model context. We further propose the semantic list decoding (SLD), which generates multiple candidate reconstructions and selects the best one via weighted Hamming distance, and a semantic confidence-guided HARQ (SHARQ) mechanism that replaces CRC-based error detection with a confidence score, enabling selective segment retransmission without CRC overhead. All modules are designed and trained using bidirectional and auto-regressive transformers (BART). Simulation results demonstrate that the proposed scheme significantly outperforms conventional capacity-approaching short codes and long codes at the same rate. Specifically, SEC provides approximately 0.4 dB BLER gain over plain short-code transmission, while SLD extends this to 0.8 dB. Compared to transmitting the entire sentence as a single long 5G LDPC codeword, our approach significantly improves semantic fidelity and reduces decoding latency by up to 90%. SHARQ further provides an additional 1.5 dB gain over conventional HARQ.

Abstract

Semantic Communication

Explore similar work

Semantic Error Correction and Decoding for Short Block Codes

When Semantic Communication Meets Queueing: Cross-Layer Latency and Task Fidelity Optimization

Not All Symbols Are Equal: Importance-Aware Constellation Design for Semantic Communication