Fast optimization of cache-enabled Cloud-RAN using determinantal point process

Abstract

Cloud radio access network (Cloud-RAN) has been considered as a potential candidate for the next generation of radio access networks. It addresses many challenges in terms of flexibility, scalability, radio resource management and energy efficiency. Caching popular contents at radio remote heads (RRHs) plays an important role for reducing fronthaul traffic congestion and delay in cache-enabled Cloud-RAN. Although, mathematical optimization methods have shown to be providing numerical solutions for addressing key signal processing issues in Cloud-RAN, the exponential complexity hinders their application in practice, particularly in large networks. Learning-based methods have become attractive to overcome the complexity issues associated with the mathematical optimization methods. Several subset selection problems have been formulated as a mixed-integer non linear program (MINLP) in wireless networks. Determinantal point process (DPP) is a probabilistic model of choosing two similar items which are negatively correlated. In this paper, we propose a DPP based-learning (DPPL) framework to obtain a subset of admitted users for cache-enabled Cloud-RAN with limited fronthaul capacity. The formulated problem of minimizing the total network cost including power and fronthaul cost while admitting as many users as possible is converted into mixed-integer second order cone programming (MI-SOCP). The subset of admitted users is obtained by learning the quality-diversity trade-off of the DPP using the optimal subsets of admitted users which are obtained by the optimization approach. We then propose an optimization algorithm to determine the beamforming and the base station-user allocation for the obtained subset of admitted users. We provide numerical results to assess the performance and complexity of the proposed DPPL algorithm and compare it with its optimization counterpart. The results reveal that the proposed DPPL can achieve a comparable performance with much lower complexity.