Tutor: CESANA MATTEO
Advisor: REDONDI ALESSANDRO ENRICO CESARE Major Research topic
:Cloud-empowered DSP Leveraging Big Data Analytics in C-RAN ArchitectureAbstract:
The fast development of Network Function Virtualization (NFV) technologies, as one of the pillars of emerging 5G and beyond networks, is spreading out to all segments of mobile networks, including both, Core Network (CN) and Radio Access Network (RAN). Once virtualized the network functions are scalable, flexible and though could be computed in different parts of the network. The full flexibility is supported by the development of powerful Cloud systems that are deployed closer and closer to the end-users, creating in such a way Edge cloud resources and the concept of Multi-access Edge Computing (MEC). From the other side, the generation of enormous amounts of network monitoring data offers an opportunity to improve and optimize any tasks related to network resource management by analyzing those datasets and extracting meaningful network knowledge.
In this work, we mainly focus on the RAN network exploiting virtualization techniques within the Cloud-RAN (C-RAN) architecture in combination with outcomes of mobile network monitoring datasets analysis at different levels of the network. We define and estimate computational requirements and constraints of computing virtualized RAN (vRAN) functions in the Cloud, for which we create an experimental platform built out of virtual machines specially designed for processing of vRAN functions. Through the detailed overview of C-RAN architecture, we investigate several design solutions by mainly tackling the problem of centralizing Digital Signal Processing functions to be computed in the cloud with General Purpose Processing (GPP) hardware.
The latest advances in the network traffic data analysis are indicating that behavioral patterns of mobile users are predictable in space and time as a high correlation between current and previous usage of mobile network infrastructure is present. Those meaningful insights about the spatial and temporal network utilization, together with the support of the NFV concept, motivate mobile network optimization to follow the actual requirements of mobile users.
For those reasons, we performed a big data analysis and exploit its outcomes in order to provide an optimization framework that allocates computational resources of virtualized RAN functions. The analysis is carried out on realistic city-wide and highway area datasets, in order to demonstrate mobile network traffic variabilities over time and space.
Moreover, for the purposes of providing a dynamic allocation of virtual resources, we compare different forecasting algorithms, analyzing in detail the impact of differently sampled datasets and prediction horizons.
The knowledge gained from the big data analysis and forecasting algorithms is further used to improve the proposed optimization framework that allocates virtualized RAN resources in a dynamic and optimal manner over time by following the variations of actual network traffic loads.
Our proposed solution of allocating virtual RAN resources by exploiting the data analysis provides higher savings in the mobile network compared to the conventional C-RAN approaches. Additionally, the framework can be potentially used by network operators to support the requirements of various mobile applications and network services requested by the mobile users as a part of 5G and beyond network use cases.