Learning-Based Resource Allocation Scheme for TDD-Based 5G CRAN System

Provision of high data rates with always-on connectivity to high mobility users is one of the motivations for design of fifth generation (5G) systems. High system capacity can be achieved by coordination between large number of antennas, which is done using the cloud radio access network (CRAN) design in 5G systems. In terms of baseband processing, allocation of appropriate resources to the users is necessary to achieve high system capacity, for which the state of the art uses the users' channel state information (CSI); however, they do not take into account the associated overhead, which poses a major bottleneck for the effective system performance. In contrast to this approach, this paper proposes the use of machine learning for allocating resources to high mobility users using only their position estimates. Specifically, the `random forest' algorithm, a supervised machine learning technique, is used to design a learning-based resource allocation scheme by exploiting the relationships between the system parameters and the users' position estimates. In this way, the overhead for CSI acquisition is avoided by using the position estimates instead, with better spectrum utilization. While the initial numerical investigations, with minimum number of users in the system, show that the proposed learning-based scheme achieves 86% of the efficiency achieved by the perfect CSI-based scheme, if the effect of overhead is factored in, the proposed scheme performs better than the CSI-based approach. In a realistic scenario, with multiple users in the system, the significant increase in overhead for the CSI-based scheme leads to a performance gain of 100%, or more, by using the proposed scheme, and thus proving the proposed scheme to be more efficient in terms of system performance.