Censoring Diffusion for Harvesting WSNs

In this paper, we analyze energy-harvesting adaptive diffusion networks for a distributed estimation problem. In order to wisely manage the available energy resources, we propose a scheme where a censoring algorithm is jointly applied over the diffusion strategy. An energy-aware variation of a diffusion algorithm is used, and a new way of measuring the relevance of the estimates in diffusion networks is proposed in order to apply a subsequent censoring mechanism. Simulation results show the potential benefit of integrating censoring schemes in energy-constrained diffusion networks.

💡 Research Summary

This paper addresses the critical challenge of distributed parameter estimation in energy-harvesting wireless sensor networks (WSNs), where the goal is to maintain estimation accuracy while ensuring long-term network sustainability through intelligent energy management. The authors propose a novel integrated framework called Censoring D-ATC (CD-ATC), which synergistically combines a diffusion-based estimation algorithm with an adaptive censoring mechanism.

The estimation core relies on the Decoupled Adapt-then-Combine (D-ATC) diffusion strategy. In this method, each node first performs a local adaptation step using its own measurements to update an intermediate estimate. Subsequently, it combines this intermediate estimate with the estimates received from its neighbors in the previous time step to produce its final updated estimate. D-ATC is chosen for its robustness in asynchronous network conditions.

The key innovation for energy efficiency is the incorporation of a censoring scheme. Instead of blindly broadcasting its updated estimate at every time step, each node decides whether to transmit based on the “relevance” or “importance” of its current information. The paper introduces a novel importance function, defined as the estimated reduction in the mean-squared error (MSE) within the node’s neighborhood if its estimate were to be shared. Each node computes a local, exponentially-smoothed MSE estimate and periodically exchanges this scalar value with neighbors to calculate the importance metric.

The energy model considers a finite battery, costs for sensing and transmission, and stochastic energy harvested from the environment. The censoring decision is governed by the Adaptive Balanced Transmitter (ABT) algorithm. This algorithm compares the calculated importance value against a dynamically learned threshold. The threshold is adaptively adjusted using a stochastic gradient method to balance the energy consumption (when transmitting) with the energy harvesting profile, aiming for long-term energy neutrality.

Simulations were conducted on a network topology consisting of two subnetworks with different noise levels connected by a bridge node. The results demonstrate that the proposed CD-ATC scheme achieves a network performance (in terms of Mean-Square Deviation) very close to an unconstrained, energy-unlimited D-ATC baseline. More importantly, it significantly outperforms a non-selective D-ATC (NSD-ATC) that always transmits, especially under low energy harvesting rates. The analysis of the evolving censoring thresholds reveals an intelligent behavior: after convergence, nodes in noisy areas adopt higher thresholds, transmitting only highly informative updates, while nodes in clean areas maintain lower thresholds, sharing information more frequently. This dynamic, data-aware censoring effectively manages the network’s energy budget without compromising estimation accuracy.

In conclusion, the paper presents a significant step towards practical, energy-aware diffusion networks. By intelligently censoring communications based on the informational value of estimates and the available energy budget, the proposed framework successfully decouples estimation performance from sheer communication volume, paving the way for sustainable operation in resource-constrained WSNs. The work also suggests potential future research in designing sparse combination schemes for diffusion networks.