[Paper Review] Non-Hierarchical Clock Synchronization for Wireless Sensor Networks

Donald R. Brown, Andrew G. Klein+1 more|arXiv (Cornell University)|Dec 6, 2012

Network Time Synchronization Technologies22 references3 citations

TL;DR

This paper proposes a non-hierarchical, distributed clock synchronization protocol for wireless sensor networks using random bidirectional pairwise message exchanges at the physical or MAC layer. It achieves monotonic expected convergence of clock drift and offset without requiring network hierarchy, local averaging, or predefined roles, with analytical and numerical validation under broad topological conditions and a critical stepsize threshold μ < N/(N−1).

ABSTRACT

Time synchronization is important for a variety of applications in wireless sensor networks including scheduling communication resources, coordinating sensor wake/sleep cycles, and aligning signals for distributed transmission/reception. This paper describes a non-hierarchical approach to time synchronization in wireless sensor networks that has low overhead and can be implemented at the physical and/or MAC layers. Unlike most of the prior approaches, the approach described in this paper allows all nodes to use exactly the same distributed algorithm and does not require local averaging of measurements from other nodes. Analytical results show that the non-hierarchical approach can provide monotonic expected convergence of both drifts and offsets under broad conditions on the network topology and local clock update stepsize. Numerical results are also presented verifying the analysis under two particular network topologies.

Motivation & Objective

• Address the limitations of hierarchical synchronization protocols like NTP and PTP in resource-constrained wireless sensor networks (WSNs).
• Overcome the high overhead and dependency on network structure in consensus-based synchronization methods that rely on local averaging of neighbor clock values.
• Develop a fully distributed, non-hierarchical synchronization approach that enables all nodes to run the same algorithm without predefined roles or topology constraints.
• Ensure robust convergence under general network topologies and clock update stepsize conditions, avoiding reliance on centralized or tree-based structures.

Proposed method

• Use random bidirectional pairwise message exchanges between nodes to exchange timing measurements, eliminating the need for a fixed network hierarchy.
• Decompose the synchronization process into two stages: drift compensation using carrier frequency measurements and offset compensation using round-trip delay estimates.
• Apply a distributed update rule based on physical-layer or MAC-layer timestamps, where each node adjusts its clock based on relative measurements from peer nodes.
• Model the system using a probabilistic messaging framework with random sender-receiver pairs, assuming uniform probability over all node pairs.
• Formulate the synchronization dynamics using a state vector β[k] representing clock deviations and apply a linear update model with stepsize μ to ensure convergence.
• Derive a sufficient condition for monotonic expected convergence using the positive definiteness of a matrix A = Qₙᵀ(R + Rᵀ − μS)Qₙ, which depends on network topology and stepsize.

Experimental results

Research questions

• RQ1 Can a non-hierarchical, fully distributed clock synchronization protocol achieve stable convergence in WSNs without relying on local averaging or network hierarchy?
• RQ2 What conditions on network topology and clock update stepsize μ ensure monotonic expected convergence of clock drift and offset?
• RQ3 How does the proposed method compare in performance and overhead to hierarchical protocols like TSPN or consensus-based protocols like FTSP and RBS?
• RQ4 Under what conditions is the synchronization algorithm guaranteed to converge, and what is the critical threshold for the stepsize μ?
• RQ5 Can the algorithm be implemented at the physical layer using existing data packets, thereby minimizing additional signaling overhead?

Key findings

• The proposed non-hierarchical synchronization method achieves monotonic expected convergence of clock drift and offset under broad conditions on network topology and clock update stepsize.
• The critical condition for convergence is that the stepsize μ must be less than N/(N−1), where N is the number of nodes in the network.
• For the equiprobable case of random pairwise exchanges, the convergence condition reduces to μ < N/(N−1), which is both necessary and sufficient for positive definiteness of the convergence matrix.
• Numerical results verify the analytical findings, showing convergence in both ring and fully connected topologies under the derived conditions.
• The method avoids the need for local averaging of neighbor clock values, reducing computational and communication overhead compared to consensus-based approaches.
• The algorithm is fully distributed and symmetric—every node runs the same update rule regardless of position or role, enabling seamless integration into existing network traffic.