.png){kind=icon}
.png){kind=icon}
Return
Group-Level Feature-Aware Hybrid Coding for Adaptive Task-Oriented Tactile Communication
.png){kind=icon}
PRE
AI
DOI：10.1109/tccn.2026.3710171.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Tactile communication imposes stringent requirements on low-latency and reliable wireless transmission. To address this challenge, this paper proposes a deep hybrid source-channel coding (DHSCC) framework for task-oriented tactile transmission. The framework integrates residual product quantization and polar coding into an end-to-end neural network, enabling efficient feature compression and digitally compatible bitstream transmission. In particular, a capsule network-based group-level importance estimation module is designed, combined with an adaptive feature group to achieve unequal error protection (UEP) for critical tactile information and alleviate the “cliff effect” under time-varying channels. Meanwhile, a lightweight online learning mechanism is embedded to dynamically update the transmission strategy by only optimizing the selection module while freezing backbone parameters, reducing deployment overhead and enhancing channel adaptability. Simulation and hardware prototype results validate the superior performance of the proposed scheme over existing DJSCC methods on tactile perception tasks. Moreover, additional tests on channel migration and image classification further demonstrate its good generalization ability for broader task-oriented communication scenarios.
Keywords:
Tactile communication
task-oriented communication
hybrid source-channel coding
group-level importance estimation
unequal error protection
Journal
I
IF:
7
Papers:
1.5K
Citations:
5.5K