Clustering algorithm for D2D communication in next generation cellular networks : thesis presented in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Engineering, Massey University, Auckland, New Zealand
Next generation cellular networks will support many complex services for smartphones, vehicles, and other devices. To accommodate such services, cellular networks need to go beyond the capabilities of their previous generations. Device-to-Device communication (D2D) is a key technology that can help fulfil some of the requirements of future networks.
The telecommunication industry expects a significant increase in the density of mobile devices which puts more pressure on centralized schemes and poses risk in terms of outages, poor spectral efficiencies, and low data rates. Recent studies have shown that a large part of the cellular traffic pertains to sharing popular contents. This highlights the need for decentralized and distributive approaches to managing multimedia traffic.
Content-sharing via D2D clustered networks has emerged as a popular approach for alleviating the burden on the cellular network. Different studies have established that D2D communication in clusters can improve spectral and energy efficiency, achieve low latency while increasing the capacity of the network. To achieve effective content-sharing among users, appropriate clustering strategies are required. Therefore, the aim is to design and compare clustering approaches for D2D communication targeting content-sharing applications. Currently, most of researched and implemented clustering schemes are centralized or predominantly dependent on Evolved Node B (eNB). This thesis proposes a distributed architecture that supports clustering approaches to incorporate multimedia traffic. A content-sharing network is presented where some D2D User Equipment (DUE) function as content distributors for nearby devices. Two promising techniques are utilized, namely, Content-Centric Networking and Network Virtualization, to propose a distributed architecture, that supports efficient content delivery.
We propose to use clustering at the user level for content-distribution. A weighted multi-factor clustering algorithm is proposed for grouping the DUEs sharing a common interest. Various performance parameters such as energy consumption, area spectral efficiency, and throughput have been considered for evaluating the proposed algorithm. The effect of number of clusters on the performance parameters is also discussed. The proposed algorithm has been further modified to allow for a trade-off between fairness and other performance parameters. A comprehensive simulation study is presented that demonstrates that the proposed clustering algorithm is more flexible and outperforms several well-known and state-of-the-art algorithms.
The clustering process is subsequently evaluated from an individual user’s perspective for further performance improvement. We believe that some users, sharing common interests, are better off with the eNB rather than being in the clusters. We utilize machine learning algorithms namely, Deep Neural Network, Random Forest, and Support Vector Machine, to identify the users that are better served by the eNB and form clusters for the rest of the users. This proposed user segregation scheme can be used in conjunction with most clustering algorithms including the proposed multi-factor scheme. A comprehensive simulation study demonstrates that with such novel user segregation, the performance of individual users, as well as the whole network, can be significantly improved for throughput, energy consumption, and fairness.