• in-band full-duplex
• non-orthogonal multiple access
• radio access network slicing
• reinforcement learning

This thesis addresses the problem of optimizing the radio resource allocation in 5G and beyond networks.
The future generations of wireless communication will experience a continuous increase of throughput requirements, as well as the rise of new services concerning massive connectivity or reliable low-latency communications.
However, the spectrum available is limited and crowded, and intelligent scheduling strategies are decisive to enable the feasibility of new communication technologies.
With this aim in mind, we investigate non-orthogonal communications, able to multiplex more than one device to a single time-frequency resource, providing higher spectral efficiency at the cost of increased signal process complexity. Also, we study the ability of optimization strategies in enabling the coexistence of different services on the same physical network.

The first part of this thesis focuses on optimization processes for non-orthogonal multiple access (NOMA) - for single-input-single-output (SISO) and multiple-input-multiple-output (MIMO) cases -, and in-band full-duplex (FD) communications. We model and solve the problem of maximum throughput having a limited power budget; also, we resolve the minimum power consumption problem requiring a minimum data rate for communication. The designed non-orthogonal allocation schedulers increase the spectral efficiency of the system, outperforming the orthogonal multiple access (OMA) strategy. Moreover, we provide better than state-of-the-art solutions, while their complexity is comparable or even lower.

The second part of this thesis regards enabling the concept of Radio Access Network slicing, focusing on the coexistence of enhanced Mobile BroadBand (eMBB) and Ultra-Reliable Low-Latency Communication (URLLC) kinds of traffic.
In practice, we are interested in how to allocate part of the available resources to devices asking for very different requirements.
We investigate this topic using two different approaches: an analytical comparison of the impact of NOMA and OMA used as the multiple access paradigm for slicing resources; the capability of artificial intelligence in dealing with the resource slicing problem in a dynamic environment.
We design feasible and optimal algorithms in the analytical study, finding that NOMA can outperform the OMA scheme for almost any case of interest.
We also demonstrate that Reinforcement Learning is a promising technique for dynamic resource allocation, able to deal with intermittent traffic better than the state-of-the-art scheduling strategy.