A spectral clustering algorithm for intelligent grouping in Dense Wireless Networks

Abstract

The density of wireless networks has been increasing with the popularization of mobile devices. Dense wireless networks (DWN) present challenges such as the current spectral scarcity and the growing demand for capacity. The Restricted Access Window (RAW) mechanism was introduced by the IEEE 802.11ah amendment to improve DWN performance. RAW restricts the number of stations that can access the channel by arbitrarily separating them into groups. K-Means clustering has shown potential to find more efficient groups using the geographical coordinates of each station. However, due to the mobile and dynamic nature of such networks, location information is difficult to obtain in practice. In this paper, we consider the use of spectral clustering to increase the performance of DWN with hidden terminals. We discuss how a spectral clustering algorithm that generates RAW groups can be implemented in practice without the geographic location of each node. We also compare the performance of the spectral clustering algorithm with the standard grouping method used in IEEE 802.11ah, with the K-Means clustering (i.e., based on node location information), and with the hidden matrix-based regrouping (HMR) algorithm. Simulation results considering several density levels, different traffic patterns, and different propagation models indicate that spectral clustering significantly outperforms both the standard grouping and HMR in terms of collision rate, throughput, and delay. It also closely approximates – and sometimes surpasses – the performance of the K-Means clustering while being much more practical to implement because it does not require knowledge on nodes’ geographical coordinates.