Fully distributed clock synchronization in wireless sensor networks under exponential delays☆
Introduction
A variety of applications in wireless sensor networks (WSNs), such as object tracking, machine health monitoring, home appliance control, etc. [1], [2], require cooperative execution of distributed tasks amongst a set of sensor nodes. To execute these tasks, nodes are usually required to run on a common time frame. However, this is far from trivial due to many factors, including imperfection of oscillators, environmental changes, and unknown delays in message delivery. Therefore, global clock synchronization constitutes an important prerequisite in many services of WSNs.
One critical component of clock synchronization is the modeling of the random network delays that perturb the message-exchange process. Existing probabilistic models for random delays in WSNs include Gaussian, exponential, Gamma, Weibull, and log-normal [3], [4], [5]. Not a single model dominates, because different models are justified in different applications. For example, it was observed in [6] that when the point-to-point hypothesis reference connection topology is of interest, the cumulative link delay in WSN is approximately represented as a single server M/M/1 queue, and the random delays should be modeled as exponential random variables. Furthermore, the minimum link delay algorithm [6], which was later derived in [7] based on the assumption of exponentially distributed network delays, shows good performance in practice. In addition, among all distributions with support in and a fixed mean, the exponential distribution has the maximum entropy, and thus least informative. Hence the assumption of exponential delay distribution is worth investigating.
The other crucial challenge in clock synchronization is to design algorithms enabling global clock calibration among all the nodes in the network. Earlier protocols were mainly based on pairwise synchronization. More specifically, before synchronization, a particular network structure (spanning tree or cluster) must be built, and then layer-by-layer pairwise synchronization was executed starting from a reference node. As an important component in this approach, pairwise synchronization under exponential delay distribution has been extensively studied in the past decade. For example, the maximum likelihood estimator (MLE) is studied in [7], [8], while the linear unbiased estimator using order statistics (BLUE-OS) and the minimum variance unbiased estimator (MVUE) are derived in [9] under asymmetric exponential distributed delays in uplink and downlink. Recently, max-product algorithm is used in [10], [11] to derive a message passing method for the clock offset tracking in the presence of exponential distributed random delays. Unfortunately, the clock skew is not considered in these works, resulting in potentially frequent re-synchronization. To avoid frequent offset resetting, skew compensation is adopted for a number of synchronization algorithms. The MLE for joint clock offset and skew estimation has been derived in [12], [13], while a low-complexity MLE has been proposed in [14], [15]. Furthermore, an approximate MLE for joint offset and skew estimation has been reported in [16].
However, in such pairwise-based synchronization approach, not only large communications overhead are required to maintain the specific network structure, synchronization errors also accumulate rapidly as distance from reference node increases. This hinders the scalability of the synchronization protocol and reduces its robustness to changes in the network topology. To relieve these problems, more recent synchronization algorithms were designed to work in a fully distributed way. Specifically, each node calibrates its own clock by exchanging messages with its direct neighbors and performing computations locally, with no special network structure maintained. Existing fully distributed synchronization schemes can be categorized into pulse-coupled based and packet-coupled based. For the pulse-coupled based approach, physical pulses are transmitted from different nodes to achieve a unified clock ticking rhythm [17], [18]. On the other hand, for packet-coupled based synchronization, clock reading at each node is encoded in time stamps and transmitted as data packets to its neighbors. Within the packet-coupled based category, two major ideas have been frequently used: consensus principle [19], [20], [21], [22] and factor graph based statistical inference [23], [24], [25]. Unfortunately, large mean-square-error (MSE) occurs in estimates from consensus based algorithms since the fixed message delays are not explicitly considered. Furthermore, for existing factor graph based algorithms, either only clock offset is considered [23], or the algorithms are derived explicitly for Gaussian delays [24], [25].
In this paper, based on the two-way message exchange mechanism, a network-wide joint estimator of clock offsets, clock skews and fixed delays is derived under exponential delay model. The joint maximum likelihood estimation problem is first cast into a linear programming (LP) problem, and then a distributed solution is derived based on alternating direction method of multipliers (ADMM) [26], [27]. Exploiting the special structure of the derived solution, further improvement in terms of computation efficiency is proposed. Both theoretical analysis and simulation results show that the proposed distributed algorithm approaches the MSE performance of the centralized LP solution.
The rest of the paper is organized as follows. In Section 2, the system model is first described, and then the MLE for global clock synchronization problem is cast as a LP-problem. In Section 3, a distributed synchronization algorithm based on ADMM is derived. Simulation results are presented in Section 4. Finally, conclusions are drawn in Section 5.
Notation: The operator takes the trace of matrix , and denotes the vector of N ones, while represents an identity matrix. denotes the block diagonal concatenation of input arguments. Superscript denotes the transpose operator. denotes the set of neighbors of node Si and indicates the number of elements in set . Finally, represents the Euclidean norm of vector .
Section snippets
System model and maximum likelihood estimator
Consider a network with N sensor nodes . These sensors are randomly distributed in the field and can be self-organized into a network1 by establishing connections between neighbor nodes lying within each other׳s communication range. An example of 25 sensor nodes is shown in Fig. 1, where each edge represents the ability to transmit and receive packets between the
Distributed clock synchronization algorithm
First, we transform the inequality constraints in (7) into equality constraints by introducing nonnegative slack variables [27], [31]. In particular, with the definitions , , (7) can be written into a LP form as
Simulation results and discussions
In this section, numerical simulations will be presented to assess the performance of the proposed distributed clock synchronization algorithm. The measure of parameter estimation fidelity at iteration m is the average mean squared error (AMSE) of clock offsets and clock skews over the whole network: where . In all figures except Fig. 7, network of 25 nodes are randomly deployed in an area , with each node having communication radius 1.5 m (an example
Conclusions and future works
Global clock synchronization for WSN in the presence of unknown exponential delays was investigated under the two-way message exchange mechanism. The MLE for the joint estimation of clock skews, clock offsets and fixed delays was cast into a linear programming problem. A distributed message-passing synchronization algorithm based on ADMM was derived. The proposed algorithm requires communications only between neighboring sensors and the communication overhead was shown to be low. Simulation
References (38)
- et al.
Wireless sensor networksa survey
Comput. Netw.
(2002) - et al.
Wireless Sensor Networks: A Systems Perspective
(2005) - J.S. Kim, J. Lee, E. Serpedin, A robust estimation scheme for clock phase offsets in wireless sensor networks in the...
- A. Noor, A. Ahmad, E. Serpedin, H. Nounou, M. Nounou, On clock offset estimation in wireless sensor networks with...
- et al.
Clock synchronization of wireless sensor networks
IEEE Signal Process. Mag.
(2011) Analysis of synchronization algorithms with time-out control over networks with exponentially symmetric delays
IEEE Trans. Commun.
(2002)On maximum-likelihood estimation of clock offset
IEEE Trans. Commun.
(2005)- et al.
Energy-efficient estimation of clock offset for inactive nodes in wireless sensor networks
IEEE Trans. Inf. Theory
(2010) - et al.
On minimum variance unbiased estimation of clock offset in a two-way message exchange mechanism
IEEE Trans. Inf. Theory
(2010) - A. Ahmad, D. Zennaro, E. Serpedin, L. Vangelista, Time-varying clock offset estimation in two-way timing message...
A factor graph approach to clock offset estimation in wireless sensor networks
IEEE Trans. Inf. Theory
On maximum likelihood estimation of clock offset and skew in networks with exponential delays
IEEE Trans. Signal Process
Maximum likelihood estimators of clock offset and skew under exponential delays
Appl. Stoch. Models Bus. Ind.
Low complexity maximum likelihood estimators for clock synchronization of wireless sensor nodes under exponential delays
IEEE Trans. Signal Process
Novel clock phase offset and skew estimation using two-way timing message exchanges for wireless sensor networks
IEEE Trans. Commun.
A scalable synchronization protocol for large scale sensor networks and its applications
IEEE J. Sel. Areas Commun.
On the scalability of cooperative time synchronization in pulse-connected network
IEEE Trans. Inf. Theory
Global clock synchronization in sensor networks
IEEE Trans. Comput.
Cited by (21)
Online distributed detection of sensor networks with delayed information
2023, Journal of the Franklin InstituteClock synchronization with partial timestamp information for wireless sensor networks
2023, Signal ProcessingConsensus-based time synchronization via sequential least squares for strongly rooted wireless sensor networks with random delays
2022, AutomaticaCitation Excerpt :Thus far, there have been considerable amount of algorithms provided in the literature for the time synchronization problem in WSNs. Distributed time synchronization, as an important category of time synchronization algorithms, has received great concern, due to its improved robustness to node failures and dynamic network topology (Barooah & Hespanha, 2007; Freris, Borkar, & Kumar, 2009; Freris, Graham, & Kumar, 2011; Giridhar & Kumar, 2006; Luo, Cheng, & Wu, 2016; Simeone, Spagnolini, Bar-Ness, & Strogatz, 2008; Stanković, Stanković, & Johansson, 2018). In particular, inspired by the consensus theory in multi-agent systems, one promising area in distributed synchronization for WSNs, called consensus-based time synchronization (CBTS), has also attracted much attention over the years (He, Cheng, Shi, Chen, & Sun, 2014; Schenato & Fiorentin, 2011; Wang, Xiong, Chen, & Wang, 2018; Xiong & Kishore, 2008).
Distributed time synchronization algorithm based on sequential belief propagation in wireless sensor networks
2021, Computer CommunicationsCitation Excerpt :Under the assumption that the nondeterministic delay obeys Exponent distribution, Ref. [3] establishes a factor graph model for sensor networks, and combines with the message transmission rule designed by the maximum product algorithm to realize the network-wide synchronization. Ref. [7] maps the joint maximum likelihood estimation problem of clock offset, clock skew and deterministic delay to a linear programming (LP) problem, and proposes a distributed message-passing synchronization algorithm based on alternating direction method of multipliers (ADMM). However, the convergence speed of ADMM based distributed algorithm is slow.
Synchronization of data measurements in wireless sensor networks for IoT applications
2019, Ad Hoc NetworksCitation Excerpt :Some of these parameters are mechanical vibrations that occur due to the devices’ positions, unpredictable temperature fluctuations, exposure to sunlight, humidity, the unwanted presence of human activity, animals, and planting, etc. These conditions are responsible for clock skew and offset that lead to fast de-synchronization and unpredictable clock rates [7,8]. Clocks drift even on the order of seconds, over quite short time periods, which is another factor pointing out that these clocks are rather unstable.