.png){kind=icon}
Return
Deep unfolding-based trainable adaptive quantization for diffusion least mean square algorithm with error compensation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.sigpro.2026.110733.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
To mitigate the substantial communication overhead incurred by information exchange in distributed adaptive filtering, several communication-efficient methods, including partial transmission, event-triggered communication, and quantization, are developed based on the diffusion least mean square (DLMS) algorithm. However, they may suffer from degraded estimation accuracy when communication resources are highly constrained. To address this issue, this paper proposes a trainable adaptive quantization error compensation diffusion least mean square (TAQ-EC-DLMS) algorithm. Building on a deep unfolding architecture, the proposed TAQ-EC-DLMS algorithm unfolds the iterative process, integrating error compensation and the DLMS algorithm into a layered network. Meanwhile, the key parameters introduced in TAQ-EC-DLMS are treated as trainable variables and optimized through training. This design not only reduces communication overhead, but also preserves estimation accuracy. The mean convergence condition and steady-state mean square deviation (MSD) expression are derived to support theoretical performance analysis. Simulation results verify the theoretical analysis and demonstrate the superior performance of TAQ-EC-DLMS in terms of communication efficiency and steady-state accuracy.
Keywords:
Distributed adaptive filter
Quantization
Deep unfolding
Communication efficiency
Error compensation
Journal
IF:
3.6
Papers:
10.0K
Citations:
1.7W
Organization
Cited Papers
No cited papers available