5G and beyond networks are expected to support a wide range of services, with highly diverse requirements. Yet, the traditional “one-size-fits-all” network architecture lacks the flexibility to accommodate these services. In this respect, network slicing has been introduced as a promising paradigm for 5G and beyond networks, supporting not only traditional mobile services, but also vertical industries services, with very heterogeneous requirements. Along with its benefits, the practical implementation of network slicing brings a lot of challenges. Thanks to the recent advances in machine learning (ML), some of these challenges have been addressed. In particular, the application of ML approaches is enabling the autonomous management of resources in the network slicing paradigm. Accordingly, this paper presents a comprehensive survey on contributions on ML in network slicing, identifying major categories and sub-categories in the literature. Lessons learned are also presented and open research challenges are discussed, together with potential solutions.

In this work, we propose a robust planning tool that allocates power statically in homogeneous and heterogeneous cellular networks with non-regular base station (BTS) placement, to mitigate interference and improve overall performance. Each BTS will use the total available spectrum, but it will divide it into multiple sub-bands, and each BTS will transmit with a specific pre-computed power on each sub-band. We refer to such a power allocation as a power map. Our offline planning tool computes a robust power map for a given topology, by solving a non-convex, non-linear optimization problem, through simple transformations, based on geometric programming. The power map is computed based solely on the network topology, and it is made available to all BTSs that use it throughout the network operation to perform scheduling using a fast quasi-optimal online algorithm that we propose. We evaluate our planning tool for different homogeneous and heterogeneous networks (HetNets), first in a static setting where scheduling is performed optimally and then in a dynamic setting when scheduling is performed with our online scheduler. Results show that our solution significantly outperforms a classical equal power/fixed frequency reuse scheme in terms of sum-rate, by up to 30% in homogeneous networks and by up to 70% in HetNets.

Next-generation mobile networks are expected to meet the requirements of a wide range of new vertical services. Hence, the network slicing concept has been introduced, in which Mobile Virtual Network Operators (MVNOs) are allowed to provide various types of services over the same physical infrastructure, owned by an Infrastructure Provider (InP). To cope with an ever-changing traffic demand, MVNOs seek to pre-allocate/reconfigure the resources at the base stations in an anticipatory manner, based on traffic demand predictions. Ideally, conducting per-slice traffic forecasting requires information that is likely to disclose MVNO confidential information (i.e., business strategy or private user data). To secure data ownership while conducting traffic forecasting, we propose the Federated Proximal Long Short-Term Memory (FPLSTM) framework, which allows MVNOs to train their local models with their private dataset at each base station; subsequently, an associated InP global model can be updated through the aggregation of the local models. The results obtained by training the models on a real-world dataset indicate that the forecasting performance of our proposed approach is as accurate as state-of-the-art centralized solutions, while improving data privacy. To enable scalability, we further propose the Information-based Clustering FPLSTM (IC-FPLSTM) and Random Clustering FPLSTM (RC-FPLSTM) frameworks, dealing with large-scale cellular networks. These solutions demonstrate computation and communication cost efficiency significantly above the state-of-the-art.

The development of virtualization techniques enables an architectural shift in mobile networks, where resource allocation, or even signal processing, become software functions hosted in a data center. The centralization of computing resources and the dynamic mapping between baseband processing units (BBUs) and remote antennas (RRHs) provide an increased flexibility to mobile operators, with important reductions of operational costs. Most research efforts on Cloud Radio Access Networks (CRAN) consider indeed an operator perspective and network-side performance indicators. The impact of such new paradigms on user experience has been instead overlooked. In this paper, we shift the viewpoint, and show that the dynamic assignment of computing resources enabled by CRAN generates a new class of mobile terminal handover that can impair user quality of service. We then propose an algorithm that mitigates the problem, by optimizing the mapping between BBUs and RRHs on a time-varying graph representation of the system. Furthermore, we show that a practical online BBU-RRH mapping algorithm achieves results similar to an oracle-based scheme with perfect knowledge of future traffic demand. We test our algorithms with two large-scale real-world datasets, where the total number of handovers, compared with the current architectures, is reduced by more than 20%. Moreover, if a small tolerance to dropped calls is allowed, 30% less handovers can be obtained.