Dottorato in TECNOLOGIE DELL'INFORMAZIONE E DELLE COMUNICAZIONI (ICT)

Titolo della tesi: Rate-Splitting Multiple Access For Beyond 5G Wireless Networks

In modern wireless networks, multiple-antenna transmission opens the door for exploiting the spatial dimension of the wireless channel to serve multiple users simultaneously, thereby achieving high spectral efficiencies. Space Division Multiple Access (SDMA) is a commonly used technique to achieve these gains. However, realizing these spectral gains is contingent upon the availability of precise and timely Channel State Information at the Transmitter (CSIT). Since receivers typically lack the ability to coordinate with each other, preprocessing the user data at the transmitter becomes essential to deal with multiuser interference; hence CSIT is crucial. In a practical system, CSIT is subject to errors, and the factors contributing to imperfect CSIT include uncertainties arising from estimation and quantization errors, and delays in CSI feedback. The performance of SDMA degrades significantly when the CSIT is imperfect. To address this challenge, this thesis explores a new multiple access technique named Rate Splitting Multiple Access (RSMA), showing superior performance to SDMA in imperfect CSIT conditions. RSMA operates by employing rate splitting at the transmitter, allowing non-orthogonal transmission of split messages, and utilizing Successive Interference Cancellation (SIC) at the receivers. This thesis investigates the application of RSMA in broadcast downlink channels within a multiuser Multiple Input Multiple Output Frequency Division Duplexed (MIMO FDD) system, where, the CSIT is imperfect due to finite rate feedback. First, we investigate the RSMA in the system operating with a finite rate feedback channel modeled as a constraint on the aggregate feedback load. The system is analyzed in an overloaded setup, where the number of potential users is larger than the number of antennas at the transmitter, and user selection is required; fairness is enforced by dividing equally the aggregate feedback load between all potential users, providing thus an equal opportunity of being selected. We introduce an approximation of the sum rate of RSMA in the above scenario and use it to determine the optimal number of feedback bits per user that maximizes the sum rate. An optimal power allocation strategy for RSMA, valid in both fully loaded and overloaded systems, is also introduced and used to define an energy-efficient switching mechanism between RSMA and SDMA as a function of the feedback bits per user, capable of improving efficiency at both transmitter and receiver. The validity of both the proposed sum rate approximation and power allocation strategy is proved by numerical simulations, that are also used to investigate the impact of feedback load, SNR, and user selection strategy on RSMA vs. SDMA. Results highlight that RSMA has superior performance in the large user regime, where many potential users operate with a low number of feedback bits, and responds better than SDMA to an increase in SNR in this regime, thanks to its superior Multiuser Interference Management (MUI) management capabilities. Finally, we propose a deep learning-based user selection strategy, shifting from optimization-driven to data-driven user selection. Comparative evaluations involve traditional greedy and channel quality information-based methods. Simulation results demonstrate that deep learning-based user selection improves RSMA performance, in terms of sum rate, outperforming conventional RSMA and SDMA-assisted systems. Furthermore, deep learning-based user selection achieves near-optimal sum rate performance with reduced time complexity compared to exhaustive search.