Dynamic Caching in a Hybrid Millimeter-wave/Microwave C-RAN

Placing popular content at the edge of the network close to users, known as caching, is a promising approach in the 5th generation (5G) of wireless communications in order to lower latency and congestion of fronthaul links. In this paper, we investigate the optimization of caching and fetching decisions to minimize a long-term network cost in a cloud radio access network. Importantly, in our model, the popularities of files vary over time and user requests are considered as arising from hidden-mode Markov decision processes. The primary fronthaul link is modeled as a millimeter-wave (mmWave) link; recognizing that mmWave links may be blocked, we allow for switching to a microwave link. The cache policy at each decision time can influence the network cost in the future, leading to coupled decision variables in time. To deal with the complexity of the resulting optimization problem, we introduce a dynamic programming approach to approximate the future cost of each cache state. The total network cost is minimized at each time-slot while also accounting for the future effects of the decisions taken. To reduce the complexity of calculating the future cost, an approximation approach is introduced and its accuracy is validated numerically. Simulation results confirm the effectiveness of our proposed algorithm to lower the total network cost while dealing with variable popularities.