.png){kind=icon}
.png){kind=icon}
Return
Feature-Sensitivity-Aware Quantization and Joint Multi-Streaming Design for Latency-Constrained Multi-Task Semantic Communications
.png){kind=icon}
PRE
AI
H
Y
X
A
DOI：10.1109/twc.2026.3718740.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Semantic communications have emerged as promising solutions for improving efficiency by transmitting relevant semantics instead of raw bits. However, for latency-critical applications, the stringent low-latency requirements pose significant challenges for semantics transfer, especially in cases with diverse user demands and limited radio resources. In this paper, we focus on the scenario of low-latency multi-streaming wireless semantic communication, where multiple users have diverse task requirements, and accordingly propose an effective multi-task semantic communication framework for efficiently delivering semantic services. More specifically, we introduce fine-grained feature-sensitivity-aware data compression to explore intrinsic uncertainty, first-order information and divergence, which allow adaptive quantization with respect to diverse low-latency demands. We aim to maximize the minimum weighted task success probability among all users via jointly optimizing feature extraction, per-feature-channel mixed-precision quantization and multi-streaming design. To facilitate the overall joint optimization, we first obtain multiple per-feature-channel quantization strategy candidates based on feature sensitivities and different bit budgets, respectively for common features shared across tasks and special features uniquely requested by specific tasks. Afterwards, we solve the joint problem of quantization strategy determination and multi-streaming design via convex problem reformulation and binary variable relaxation. Finally, from the relaxed optimum, we reconstruct a feasible joint solution for the whole design with high performance quality and close optimality. Simulation results verify that the proposed design substantially outperforms baseline methods, while the proposed optimal multi-streaming design efficiently and effectively addresses the low-latency semantic communication scheduling.
Keywords:
Multi-task semantic communication
feature-sensitivity-aware quantization
multi-streaming design
finite blocklength codes
resource allocation
Journal
IF:
10.7
Papers:
1.3W
Citations:
5.3W
Organization
Cited Papers