To Split or Not to Split? From the Perspective of a Delay-Aware Data Collection Network Structure

Abstract

Abstract-Collecting data from massive numbers of individual nodes is always a challenging task in wireless sensor networks. The duration of a data collection process, which can greatly affect the detection capabilities of a network, should be reduced whenever possible. For scenarios where only a single cluster is allowed, the delayaware data collection network structure can minimize the duration of a data collection process. The aim of this paper is to explore the possibilities of improving the original delay-aware network structure by splitting the single tree structure into multiple clusters. Analyses on the conditions and effects of splitting the aforementioned structure are presented. Based on the analyses, two novel network splitting algorithms using k-means clustering algorithms are proposed. Simulation results show that the proposed network splitting algorithms may further reduce the duration of a data collection process. With the help of the k-means algorithms, communication distance among sensor nodes can be further reduced especially for networks with large numbers of wireless sensor nodes.

A k-Means-Based Formation Algorithm for the Delay-Aware Data Collection Network Structure

“…The T DCP of the whole network is therefore governed by that of its largest cluster, i.e. T DCP = log 2 |N | 2 + 1 = log 2 N , which concurs with findings in [2], [3].…”

“…Nguyen [10], the very first distributed clustering algorithm providing non-overlapping multi-hop clusters with energy concerns. In [3], Cheng and Ganganath are the first who attempted to exploit the geographical locations of sensor nodes in order to facilitate the formation of DADCNS. In their first attempt, a network is divided into sub-clusters by a kmeans algorithm.…”

“…In the simulations, performance of the original DADCNS will be used as references. The DADCNS will be constructed as a single cluster and multiple clusters using the top-down network formation approaches proposed in [2] and [3], respectively. In order to evaluate the effect of |N | to the performances of networks with different network structures, |N | is varied from 3 to 99 with a step-size of 3.…”

“…In the simulations, the duration of a DCP (T DCP ) and the total squared communication distance (Ψ) are used as performance indicators [3]. T DCP is expressed as the total number of time-slots required by a BS to collect data from all the nodes in the network.…”

“…(b) A network organized into the DADCNS using the topdown approach in [2] and [3]. clusters separately are less likely to yield long communication links.…”