[Paper Review] Proportional-Integral Clock Synchronization in Wireless Sensor Networks

Kasım Sinan Yıldırım, Ruggero Carli+1 more|arXiv (Cornell University)|Oct 29, 2014

Network Time Synchronization Technologies30 references3 citations

TL;DR

This paper proposes PISync, a novel distributed clock synchronization algorithm for Wireless Sensor Networks (WSNs) based on a Proportional-Integral (PI) controller that minimizes clock offset and frequency differences using only local feedback. The method achieves low memory and CPU overhead, minimal message exchange, and scalable performance with slow-growing synchronization errors across increasing network diameters.

ABSTRACT

In this article, we present a new control theoretic distributed time synchronization algorithm, named PISync, in order to synchronize sensor nodes in Wireless Sensor Networks (WSNs). PISync algorithm is based on a Proportional-Integral (PI) controller. It applies a proportional feedback (P) and an integral feedback (I) on the local measured synchronization errors to compensate the differences between the clock offsets and clock speeds. We present practical flooding-based and fully distributed protocol implementations of the PISync algorithm, and we provide theoretical analysis to highlight the benefits of this approach in terms of improved steady state error and scalability as compared to existing synchronization algorithms. We show through real-world experiments and simulations that PISync protocols have several advantages over existing protocols in the WSN literature, namely no need for memory allocation, minimal CPU overhead and code size independent of network size and topology, and graceful performance degradation in terms of network size.

Motivation & Objective

• Address the high computational and memory overhead of existing time synchronization protocols in WSNs.
• Overcome scalability limitations of flooding-based and fully-distributed protocols due to reliance on time history storage and regression.
• Develop a lightweight, distributed synchronization algorithm that maintains accuracy without storing past time information.
• Ensure robustness and scalability in large or dynamic WSNs with minimal resource consumption.
• Provide theoretical and empirical validation of performance across different network topologies and sizes.

Proposed method

• Design a Proportional-Integral (PI) controller to dynamically adjust local clock offsets and frequencies using real-time feedback.
• Implement two variants: a flooding-based protocol (FloodPISync) and a fully distributed protocol (AvgPISync) using neighbor-to-neighbor communication.
• Use only current synchronization error measurements—no historical data storage—reducing memory and computational load.
• Apply PI control law: update local clock by combining proportional feedback on current offset and integral feedback on accumulated error.
• Integrate the PI controller into a distributed consensus framework to enable self-calibration across the network.
• Modify the JProwler simulator to model hardware clock drifts (±50 ppm) and implement PISync, PulseSync, and GTSP for comparative evaluation.

Experimental results

Research questions

• RQ1 Can a PI-based control approach achieve accurate clock synchronization in WSNs with minimal memory and CPU overhead?
• RQ2 How does the performance of PISync compare to existing protocols like PulseSync and GTSP in terms of global and local skew across varying network diameters?
• RQ3 Does the fully distributed AvgPISync protocol outperform flooding-based variants in local synchronization accuracy despite slower global information propagation?
• RQ4 To what extent does PISync maintain low synchronization error as network size and diameter increase?
• RQ5 Can PISync achieve scalable performance without requiring time history storage or complex regression computations?

Key findings

• PISync protocols exhibit very slow growth in both global and local skew with increasing network diameter, demonstrating strong scalability.
• The fully distributed AvgPISync protocol achieves lower local skew than FloodPISync in larger networks due to better exploitation of local neighborhood information.
• FloodPISync and PulsePISync achieve lower global skew than AvgPISync due to faster dissemination of time information via flooding.
• PulsePISync outperforms FloodPISync in global skew due to faster time information dissemination using pulse-based flooding.
• PISync protocols require no memory allocation for time history, minimal CPU overhead, and have code size independent of network size and topology.
• Simulation results show PISync maintains high synchronization accuracy over 5-day real-world simulations (5×10⁵ seconds) across different network diameters, confirming robustness and scalability.