The performance evaluation of ai based resource allocation algorithms for donwlink NOMA systems

Abstract

From first generation of mobile networks to 5G, different multiple access techniques such as Frequency Division Multiple Access(FDMA), Time Division Multiple Access(TDMA), Code Division Multiple Access(CDMA), and Orthogonal Frequency Division Multiple Access are used. Except from CDMA, all these techniques uses same frequency/time slot for serving a single user. In contrast, CDMA makes it possible for several users to be served by a single frequency/time slot by using various distinctive, user-specific spreading sequences. For 5G and later, these traditional methods does not satisfy the needs for the new system requirements. With the increase of mass communication and IoT technologies, different requirements such as low latency, high throughput and wide range have emerged. Various techniques such as Multiple-Input Multiple-Output(MIMO) and Non-orthogonal multiple access (NOMA) have proposed to meet these requirements. In MIMO, data is transmitted simultaneously as various signals over a single channel using different antennas. On the receiver side, these signals are recombined with the help of the MIMO module tuned with the same number of antennas. In NOMA, the users are multiplexed in the power or code domain by using the same time slot and the same frequency band. NOMA provides efficient resource allocation and bandwidth utilization. There are three important services in 5G and beyond wireless mobile networks: enhanced mobile broadband (eMBB), ultra reliable low latency communication (URLLC), and massive machine type communication (mMTC). Each service has its own challenging requirement such as the URLLC service needs reliable data transfer with the end-to-end delay of less than 1 ms while the mMTC service requires supporting a huge amount of devices simultaneously connected to the network. NOMA is one of important technologies to provide the spectral efficient utilization of limited bandwidth resources for 5G and beyond networks. As the number of IoT devices increase significantly, NOMA becomes more important to support the mMTC service by allowing multiple users share the same frequency resource simultaneously, where different power levels are assigned for the users within the same frequency resource. In this thesis, we develop three different artificial intelligence (AI) based resource and power allocation algorithms, namely Hill Climbing (HC), Simulated Annealing (SA), and Genetic Algorithm, for downlink NOMA systems. In the proposed approach, one of the AI algorithms is used to determine the NOMA user groups along with the frequency resource block for each group. Then, the optimum power allocation is performed to maximize the geometric mean of the user throughputs. The simulation experiments are performed to compare and contrast the performance of these three AI algorithms. The numerical results demonstrate that the GA provides the best results while the HC performs the worst.