چكيده به لاتين
As traffic on wireless networks increases, more bandwidth is needed. High-bandwidth millimeterwave (mmWave) communications are of interest to the fifth generation of wireless communications.
Besides, due to the rapid growth in the number of mobile devices, it is necessary to use the new multiple access technique with mmWave communication to achieve extensive connectivity and increase
access to the mmWave spectrum. Using multiplexing in the power domain with successive interference cancellations (SIC), non-orthogonal multiple access (NOMA) is an essential candidate in
fifth-generation telecommunications. The resource allocation in the mmWave-NOMA systems is
a crucial issue, the optimal allocation of which can maximize the NOMA capacity of mmWave
systems. This optimization problem is a mixed integer programming problem and, therefore, nonconvex. The mmWave-NOMA systems enable multiple users to communicate simultaneously on
a single beamforming (BF) vector, which complicates the architecture of mmWave-NOMA systems. To solve this optimization problem, machine learning methods can be used. The goal of this
project is to design the power and cluster allocated to each user so that the maximum sum rate can
be achieved. Clustering in the field of machine learning literature is also known as a grouping. To
design user clustering, one of the machine learning methods is called deep reinforcement learning.
Because accurate information and a complete model of environmental behavior in mobile environments are unknown, the model-free reinforcement learning algorithm is used to solve the stochastic
optimization problems in wireless networks. Reinforcement learning provides an effective solution
to the problems of successive decision-making when the environment is unknown, and optimal performance is learned through interaction with the environment. In this project, the resource allocation
in the downlink mmWave-NOMA systems considered for objective functions such as spectral efficiency and energy efficiency. The simulation results confirm that the proposed system performs
better in terms of spectral efficiency and energy efficiency up to about 50 percent and 20 percent
than other clustering schemes in the references.
