Abstract:
Nonorthogonal multiple access (NOMA) su ers from intra-beam interference in addition to inter-beam interference. Therefore, its performance is highly interference limited. However, implementation of NOMA on millimeter wave systems promises to overcome this issue. Intra-beam and inter-beam interference are diminished respectively by high path loss and directional beam characteristics of millimeter waves. Consequently, higher system output can be achieved by combining NOMA and the millimeter wave communications. Nonetheless, inter-cell interference should also be taken into consideration as mobile networks have a tendency to be denser with each new generation of mobile communication systems. Especially in millimeter wave systems, cell ranges are fairly smaller than conventional systems. User clustering and power allocation methods can be used e ectively to diminish the negative impact of the interference. In this work, user clustering in multi cellular downlink millimeter wave NOMA systems is investigated. K-medoids, an unsupervised machine learning algorithm is used to cluster users with the goal of maximizing the system output while keeping the total transmission power and quality of service (QoS) constraints in consideration. The simulation results show that K-medoids user clustering successfully clusters physically pre-clustered users and increases the system output in comparison to several other user clustering methods. Additionally, performance improvement over traditional orthogonal multiple access (OMA) is also demonstrated.
