Power-aware optimization of baseband-function placement in cloud radio access networks

Because of the advent of highly diverse and heterogeneous services, 5G networks required the design of a new paradigm in terms of Radio Access Networks. The Centralized- RAN appeared as a suitable solution, enabling the reduction of costs by centralizing the baseband functions. Nevertheless, it requires very high bandwidth, making it unfeasible. Functional splits were proposed to divide the protocol stack and place the baseband functions in Central Units (CU) or Distributed Units (DU). It is an efficient way of enabling different types of services while reducing costs for operators. However, each split option introduces very stringent constraints in terms of latency, throughput and processing. The problem that arises is how to optimally place the CU functions in the network nodes. This paper proposes an optimization framework to minimize link and node power, subject to the split requirements. We also present models to compute bandwidth, computational effort and latency for splits 2 (PDCP-RLC) and 6 (MAC-PHY). The solution proposed was compared to the fully centralized and distributed scenarios over a realistic topology of a metro network with 24-hour traffic. Results show that the proposed CU-placement optimization reduces the power consumption in respect to both scenarios in split 2. In addition, even if the consumption is very similar to the centralized scenario in split 6, the ILP solution meets the latency requirements. Finally, when comparing the split options, the overall power consumption in split 2 is 5% lower than split 6.