Machine Learning Based Improved User-Pairing and Power Allocation with Imperfect-Successive Interference Cancellation for Downlink NOMA-UAV System

Non-orthogonal multiple access-unmanned aerial vehicles (NOMA-UAV) are the enabling technologies for 5G network in increasing the spectral efficiency as well as enhancing the computing capability for the massive machine type communication (mMTC) devices. To support mMTC devices, machine learning (ML) based algorithms can efficiently exploit the user pairing (UP) scheme for NOMA-UAV system. In this paper, we have proposed an improved user pairing (IUP) scheme using k-mean clustering algorithm and derived the lower and/or upper bound conditions on power allocation coefficient under imperfect successive interference cancellation (Im-SIC). The objective of the proposed scheme is to maximize the system sum-rate capacity under the minimum transmission rate constraints and total remaining UAV transmission power. Based on the user distance from the base station and channel condition of the users, a hybrid approach (where paired users utilize the NOMA technique and unpaired users will utilize the OMA technique) is used to maximize the system throughput for a NOMA-UAV system. Simulation results reveals that the proposed user pairing scheme achieve the high spectral efficiency and ensure good fairness among users compare to the other user pairing schemes such as Near-Far (NF), Near-Median (NM) and Random user pairing schemes.