Joint memory and radio resource optimization in cache-aided multi-antenna communications

The rapid growth of global internet traffic in recent years has underscored the need for more efficient data delivery methods, due to the rising use of high-bandwidth applications such as video streaming, cloud computing, online gaming, and real-time communication. As a result, modern networks face increasing challenges in meeting demands for speed, scalability, and robustness to user heterogeneity. This thesis addresses these challenges by optimizing the usage of the distributed memory across the network and managing the radio resource allocation in cache-aided multi-input single-output (MISO) setups.

The first part of the thesis addresses the dynamicity challenge in cache-aided networks, while the server has direct access to the centralized data library. In dynamic networks, users can join or leave the network at any time, and the server has no prior knowledge of the active user count. This dynamic behavior calls for a transmission protocol that supports an arbitrary number of users. To this end, we propose a universal framework based on coded caching (CC), designed for cache-aided networks with a dynamic population of users. The content placement phase is designed based on users’ cache ratios rather than their number, and each user is assigned a caching profile upon entry. Accordingly, the delivery phase is designed to maximize the achievable degrees of freedom. To further reduce delivery latency at finite signal-to-noise ratio (SNR), we introduce a flexible subpacketization scheme that adjusts data fragment sizes according to cache profile imbalances, increasing the number of users served per transmission.

The second part of the thesis studies the multi-user information retrieval (MIR) in networks with distributed data, in which the base station (BS) lacks direct access to the data library and functions as an amplify-and-forward relay. The proposed MIR model is inspired by a CC-based uplink (UL)–downlink (DL) mechanism. In the UL phase, users transmit data fragments to the BS, which linearly combines these inputs to generate DL transmissions. To minimize delivery latency, UL and DL resources are optimized by decoupling the end-to-end optimization problem into two subproblems. UL parameters are first optimized to reinforce the uplink, then used in the DL to minimize end-to-end delivery time. By tuning the spatial multiplexing factor with respect to the SNR, the proposed MIR model outperforms existing approaches, effectively leveraging both spatial multiplexing and CC gains across the UL and DL phases.