Delay Constrained Hybrid CRAN: A Functional Split Optimization Framework

Abstract

Hybrid cloud radio access network (CRAN) architecture supports the realization of splitting/virtualization of the baseband processing unit (BBU) functions processing between the central cloud, central office that has large processing capacity and efficiency, and the edge cloud, an aggregation node which is closer to the user, but usually has less processing efficiency. In our previous work, we studied the impact of different split points (which decide the amount of communication processing at center and edge clouds) on the system's energy and midhaul link bandwidth consumptions. Midhaul is defined as the part of the network connecting center cloud to edge cloud. In this study, we propose an optimization framework to incorporate the end-to-end delay, from the central cloud to the end user, under different/flexible function split points. For all services requests, all users have the same services' size (in Mbits) that are associated with different delay requirements based on the service. Different service/delay requirements enforce different function splits. Our proposed optimization framework minimizes both system's power and bandwidth consumption with guaranteed end-to-end latency performance (control plane is not considered). The required split decision is significantly dependent on the processing power efficiency ratio between processing units at edge and central clouds. As shown in our previous results, Hybrid CRAN achieves about 55 percentage power saving compared fully distributed system, at the expense of midhaul bandwidth consumption. Simulation results showed that as the delay requirements decrease, average allowable latency reduces from 65msec to 45msec, this power consumption (achieved by relaxed delay case) increases by average of 33 percentage, at 20 resource blocks per user.