Search

Machine learning paradigms have gained considerable interest in the field of wireless networks. In particular, clustering techniques have been widely used to implement efficient routing plans, manage massive data, and increase network performance. The K-means clustering algorithm is one of the algorithms that has attracted the most interest from the scientific community for clustering in various wireless network applications. However, this algorithm relies on parameters that need to be adjusted to achieve viable performance. Recently, a free-initialisation version of K-means has been proposed for pattern recognition. In this paper, we demonstrate the effectiveness of this method for cluster identification and energy conservation in wireless sensor networks. In addition, we present a bio-inspired scheme for efficient cluster head selection. The effectiveness of the proposed protocol is evaluated using the PyCharm simulation tool. The results show that the proposed protocol outperforms recent state-of-the-art solutions in terms of energy consumption and network lifetime.

One of the most challenging tasks in a disaster scenario is the detection and localization of victims with high accuracy and minimum delay, especially in out-of-coverage areas. In the event of a disaster that disrupts the cellular network infrastructure, emergency calls can be relayed to the core network via multi-hop D2D communications. In this paper, a localization system is proposed that uses radio measurements obtained through such D2D multi-hop assisted emergency calls to localize in-coverage and out-of-coverage devices. To address the uncertainty and gradual reception of data in real-time in this scenario, a dynamic constraint satisfaction-based Multi Victim Localization Algorithm (MVLA) is proposed. This algorithm locates multi-hop devices in a progressive propagation manner to provide fast and accurate updates on victim locations. Additionally, three modes of ${MVLA}$MVLA, namely ${MVLA}_{recent}$MVLArecent, ${MVLA}_{seq}$MVLAseq, and ${MVLA}_{all}$MVLAall are proposed. Simulation results demonstrate that ${MVLA}_{all}$MVLAall has a lower localization error compared to ${MVLA}_{recent}$MVLArecent and ${MVLA}_{seq}$MVLAseq. Moreover, ${MVLA}_{all}$MVLAall, is compared with an existing particle filtering-based localization algorithm called RSSI Monte-Carlo Boxed Localization (RSSI-MCL) under an increasing number of emergency user devices and functional gNodeBs. Results show that ${MVLA}_{all}$MVLAall significantly outperforms the RSSI-MCL method in terms of localization accuracy and computational delay.