This PhD dissertation investigates the modeling, enablers and performance of Sub-Band Full Duplex (SBFD) to improve uplink (UL) performance in 5G-Advanced and future 6G networks. SBFD enables simultaneous UL and downlink (DL) transmissions at the 5G base station (gNB) in separate frequency sub-bands within the same carrier, offering increased UL transmission opportunities in the time domain to improve user experience, particularly for coverage-limited users, without damaging the DL performance. The research is structured into three major contributions, including interference modeling and performance evaluation, interference mitigation and radio resource management (RRM) enhancements, and machine learning (ML)-based dynamic adaptation of radio frame configurations with options for SBFD-slots in realistic network settings.

The first contribution focuses on system-level modeling of performance determining effects of SBFD and its actual performance. An interference model is developed to identify dominant interference components and to determine the required level of gNB self-interference cancellation for different deployment cases. Key performance indicators (KPIs), such as user experienced throughput and packet latency, are compared with baseline static TDD. The analysis highlights trade-offs between increased UL throughput and the challenges of gNB self-interference and cross-link interference (CLI) between gNBs, and UE-to-UE interference, which can notably impact DL performance in dense deployments.

The second contribution addresses the identified challenges related to gNB self-interference and gNB-to-gNB CLI in high-power gNB scenarios. To mitigate this, a gNB downlink transmit power control mechanism is introduced within SBFD slots. This method aims to reduce interference and improve the UL performance, particularly in wide-area deployments. To support this, RRM enhancements are proposed to enable accurate link adaptation and resource allocation in response to SINR variations resulting from dynamic power adjustments. The proposed mechanisms are shown to improve UL performance while maintaining acceptable DL quality.

Given that SBFD performance is highly dependent on traffic load and network conditions, the final contribution of the thesis introduces a dynamic radio frame configuration scheme, including combination of DL-only, UL only, and SBFD-slot types. A deep reinforcement learning (DRL) model based on the soft actor–critic (SAC) algorithm is developed to adaptively select the most suitable radio frame configurations based on UL and DL real 4-time buffer status. The model incorporates multi-step learning to improve robustness and policy convergence under dynamic traffic conditions. The proposed approach shows it is feasible and beneficial to have dynamic selection between TDD and SBFD frame structures according to the varying network conditions and traffic types, including extended reality (XR), confirming its potential for supporting different services

In urban macro and dense urban deployments, SBFD can achieve up to a fourfold improvement in uplink throughput for cell-edge users under low to medium offered loads, given gNB self-interference mitigation levels of 149 dB and 145 dB, respectively, can be realized. In low-power indoor scenarios, SBFD requires less stringent self-interference mitigation of 90 dB, making it more practically feasible. As the optimal frame configuration varies depending on load conditions and network characteristics, employing machine learning-driven adaptive SBFD operation is recommended to fully realize its benefits. These findings lay the groundwork for ongoing 6G research, guiding the development of more intelligent and efficient duplexing mechanisms.

After the defence there will be a small reception at Fredrik Bajers Vej 7B/2-104