Optimal functional split selection and scheduling policies in 5G Radio Access Networks

We theoretically study the joint functional split selection and scheduling problem that arises in a Centralized Radio-Access-Network (C-RAN) architecture with a central location connected to a set of Remote Radio Heads (RRHs) through fronthaul links. We consider a static scenario where a set of LTE frames at the central location need to be transported to RRHs through the fronthaul topology. One frame is destined towards one RRH and contains data to be transmitted to users in that RRH. For each frame, a functional split needs to be selected out of a discrete set of available options, where each option corresponds to a different split of the baseband processing load between the server at the central location and the one residing at the RRH. A functional split also requires that a certain amount of data traverses the fronthaul link to the RRH. Prior to transmission, frames need to be scheduled for processing at the central location server. The total latency experienced by the frame is the sum of computation, data communication and scheduling delays, and it depends on both the schedule and the functional split selection. We seek to characterise the complexity of the joint functional split selection and scheduling policy that minimizes total latency or the maximum latency over all RRHs. The former objective becomes equivalent to a constrained shortest-path problem which is NP-Hard, although the scheduling problem for given functional splits and the functional split selection problem for given scheduling policy are polynomially solvable. The latter objective is also NP-Hard, while the problem of scheduling for given functional splits is optimally solvable through its equivalence to single-machine scheduling for maximizing lateness, and the functional split selection problem for fixed schedule is a mixed-integer linear program.