چكيده به لاتين
Nowadays, big data are so important, and using cloud radio access networks (CRANs) in 5G and 6G is one of important ways for big data transmission. Toward this end, two scopes are considered in this thesis: the first is to provide efficient and reliable QoS in downlink transmission of big data for secondary users (SUs), and the second is to select SUs, schedule and allocate frequency spectrum and remote radio heads (RRHs) for those selected SUs.
Existing approaches either try to maximize the number of accepted SUs or the sum data rate of admitted SUs. Optimization by using first approach is very simple, but unfairly favors users with small data requests, whereas optimization by second approach is difficult, but that allocates most resources to users with better channel conditions. In contrast, a novel approach is provided in this thesis. In this novel approach, a new definition of the objective function is proposed that provide priority for big data requests while simultaneously maintaining a certain degree of fairness among SUs. On the other hand, each big or ordinary data request can have a different demands including data size, target bit error rate (BER), minimum signal to noise ratio (SNR), and deadline. These demands are involved in the optimization problem by considering all five V features of big data (including volume, veracity, value, velocity, and variety). Therefore, many parameters affect two conflicting demands of big data priority and fairness among Sus (in the objective function), which makes optimization very challenging. To overcome this challenge, three novel algorithms are proposed (first one is optimal and two others are sub-optimal) to maximize sum weighted data transfer. In first algorithm, using a novel objective function jointly optimizes SU selection, deadline-aware scheduling, and spectrum and RRH allocation. Then, it is demonstrated that finding the global optimum solution entails the enumeration of all colorful independent sets on a generalized interval graph, which is known to be NP-hard. To solve this problem, a dynamic programming (DP) approach is proposed, which yields the global optimum solution at a reduced computational cost. However due to high computational complexity, this proposed solution is practical for small-to-medium size networks. Moreover, it offers an optimum benchmark for any new sub-optimal algorithm.
In order to reduce complexity in the massive CRANs, two sub-optimal algorithms are proposed. The first one is offline batch (OFB) algorithm, that assumes all data requests and network information are available at the time of optimization. This assumption is bottleneck in real-time applications. Therefore, the second proposed algorithm is online real-time (ONR) scheduling, that performs admission and resource allocation on-the-fly utilizing predictions of upcoming data requests and resources. These two novel sub-optimal algorithms are rigorously analyzed and mathematical bounds on their performance compared to global optimum are derived. Moreover, their complexities are in order of polynomial respect to the number of parameters in objective function. The simulation results show proposed algorithms favor big data requests of SUs while incurring only a small degradation in the fairness index.
