计算机科学
接头(建筑物)
编码(社会科学)
稳健性(进化)
传输(电信)
计算机视觉
信号处理
人工智能
实时计算
语音识别
数据传输
编码(内存)
钥匙(锁)
算法
作者
Kemi Chen,Ting Zhang,Mingkai Chen,Zhenyu Liu,Tiesong Zhao,Xiaoming Tao
标识
DOI:10.1109/tmc.2026.3711711
摘要
Efficient and reliable transmission of vibrotactile signals is critical for interactive applications motivated by scenarios such as remote haptics, robotics, and teleoperation, where low latency and high-quality objective reconstruction are essential. Although recent deep learning–based Joint Source–Channel Coding (JSCC) methods have demonstrated strong robustness in speech and image transmission, their direct application to vibrotactile signals remains suboptimal due to interaction-induced temporal non-stationarity and perception-oriented objective reconstruction requirements. This paper proposes Tactile-JSCC, a task-aware JSCC framework specifically designed for vibrotactile transmission. The proposed framework incorporates an entropy-guided adaptive attention mechanism in both the encoder and decoder to model temporal non-stationarity and dynamically prioritize interaction-critical signal regions during transmission. To further enhance reconstruction quality, a time-frequency fine decoding module is introduced to jointly exploit local temporal continuity and global spectral structure. Additionally, a perception-aligned reconstruction objective is employed to guide end-to-end optimization toward perception-oriented objective reconstruction. Extensive experiments on publicly available vibrotactile datasets demonstrate that Tactile-JSCC achieves superior average performance over representative JSCC baselines and LDPC-coded SSCC tactile codecs under high compression ratios, imperfect CSI, and adverse channel conditions, enabling robust vibrotactile transmission.
科研通智能强力驱动
Strongly Powered by AbleSci AI