A Reconfigurable Architecture for Packet Based 5G Transport Networks Raghu M. Rao Xilinx Inc. 2100, Logic Drive, San Jose, CA, USA raghu.rao@xilinx.com Mickaël Fontaine, Raimena Veisllari TransPacket Hoffsvein 21-23, Oslo, Norway {mickael.fontaine; raimena.veisllari}@transpacket.com Abstract— The 5G system architecture envisions a converged backhaul and fronthaul network which necessitates the fronthaul links to move to an Ethernet packet based network with preemption capabilities to ensure bounded delays. Furthermore, the introduction of the PDCP aggregation node as part of the Telco Cloud introduces a lower priority midhaul link but still requiring bounded delays. This paper presents a realistic Ethernet packet based 5G transport network with fronthaul express traffic and lower priority, preemptable, midhaul/backhaul traffic in a multi- hop network connecting the remote radio units and the baseband unit to the Telco Cloud and the packet core. The system includes an intelligent scheduler that aggregates and schedules midhaul/backhaul traffic by exploiting the inter-packet gaps without impacting the delay or PDV of the fronthaul streams. Analytical and experimental results are presented to demonstrate the effectiveness of the scheduling algorithm. Index Terms—5G, CRAN, RoE, NGFI, Fronthaul, Midhaul, Backhaul, Xhaul, Time Sensitive Networking, eCPRI, CPRI, Scheduling, Preemption. I. INTRODUCTION As 5G wireless technology gains traction, three use cases have emerged as the key to defining the air interface and the transport network. The enhanced mobile broadband (EMBB), the first of these use cases, drives the throughput at the user equipment (UE) to 10x compared to 4G. The massive machine type communication (MMTC) or massive IoT brings millions of connected devices into the network. While each of these devices could transmit and receive small bursts of data, the high total number of devices communicating small packets of data has the potential to overwhelm the transport network. The third use case is the ultra-reliable low latency communication (URLLC), which is being defined to enable applications such as remote surgery and self-driving cars that require extreme reliability and low latency communications. Given the vastly different traffic patterns that these three use cases generate, a revolutionary change in the radio access network is required, all the way from the air interface to the packet core and the transport network. The diverse nature of 5G traffic requires a dynamic and flexible network, which the existing network is incapable of providing. In particular, the fronthaul in a 4G network relies on hardwired point-to-point connections in a time division multiplexed system using the CPRI [1] protocol. While CPRI based fronthaul is capable of providing varying transmission rates, it is not dynamically configurable and is not efficient under varying traffic conditions. This has prompted the move to packet-based fronthaul, based on Ethernet technology. This move also offers the possibility to support fronthaul, midhaul and backhaul over the same Ethernet transport infrastructure. However, Ethernet is a best effort technology and requires Time Sensitive Networking (TSN) [2] to provide a flexible and dynamic transport network with low and bounded delays. In addition, deployments of Ethernet Fronthaul require very accurate timing to serve advanced radio functions and also require encapsulation of the radio signal in an Ethernet frame. There are two competing standards for encapsulating Fronthaul traffic in an Ethernet frame; 1) eCPRI from the CPRI consortium [3] and 2) Radio over Ethernet (RoE) or NGFI from the IEEE 1914 Working Group [4]. While the encapsulations themselves are different, the principle is similar in both cases. The EMBB use case relies on technologies such as Massive MIMO that greatly improve network capacity by enabling significant frequency reuse and dual/multi connectivity that uses the WiFi spectrum to provide additional links to the user equipment. This comes at the price of increased fronthaul capacity. To contain fronthaul capacity, the baseband signal chain is being partitioned, and a portion of the layer 1 is moving to the remote radio unit (RRU). In addition, other functional splits in layer 2 are also being considered. One important functional split called Option 2 splits the upper L2 with the PDCP layer into an aggregation node to enable dual/multi connectivity. This PDCP aggregation node is also an important part of the multi-access edge computing node and the now emerging Telco Cloud which can provide much quicker hand- offs and also much reduced end-to-end latency. The link from the Telco Cloud to the base station is called midhaul. All this leads to prioritized traffic in the cellular transport network with fronthaul being high priority, a midhaul with lower priority but still requiring bounded delays, and backhaul which connects the Telco Cloud to the packet core with even more relaxed delay bounds. Fronthaul traffic, due to its strict timing constraints, cannot be preempted [2], while midhaul and backhaul are preemptable. This paper describes a complete Ethernet fronthaul ecosystem providing all the required building blocks to implement the transport network for new 5G use cases. The paper further presents an Ethernet transport network incorporating fronthaul express traffic and lower priority (preemptable) midhaul/backhaul traffic in a multi-hop network