The evolution of wireless networks toward the sixth generation aim to support a widening range of services -- from high-throughput enhanced mobile broadband (eMBB) to low-latency extended reality (XR). Achieving this flexibility demands both higher link reliability under challenging propagation conditions and higher system capacity. This thesis addresses these goals through physical- and medium access control- layer enhancements: advanced receiver-side processing for improved link reliability and improved dynamic link adaptation (LA) for increased system capacity. The proposed LA framework extends conventional hybrid automatic repeat request (HARQ) by enabling LA during retransmissions, in addition to enhancing LA for the first transmission of a transport block.

To enhance link reliability in coded multiple-input multiple-output (MIMO) systems, receiver-side processing is identified as the primary domain for improvement. A deep unfolded iterative learning approach is proposed for joint detector–decoder optimization. The proposed soft-input soft-output successive interference cancellation-based MIMO receiver leverages deep learning to refine the iterative exchange of bit reliability information, improving interference cancellation efficiency and overall detection accuracy in terms of coded bit error rate, compared to existing solutions. Since symbol ordering critically affects SIC performance but is computationally intensive under conventional minimum mean squared error methods, a deep learning–driven ordering mechanism is introduced, significantly reducing SIC complexity.

To enhance downlink (DL) system capacity, LA for the first HARQ transmission is improved by mitigating the slow convergence of conventional fixed-step outer loop LA (OLLA) toward stable operating points at the gNodeB (gNB). The proposed exponential OLLA (EOLLA) introduces step adjustments that apply larger corrections when far from convergence and smaller ones near the target, while maintaining stability through system-level a priori statistics. This mechanism systematically drives modulation and coding scheme (MCS) selection toward higher efficiency, yielding approximately a 33% DL capacity gain over legacy OLLA for latency-sensitive XR traffic in indoor hotspot deployment. Practical realization is enabled by a lightweight user equipment (UE) module using simple arithmetic and lookup tables, requiring minimal additional signaling to the gNB.

Further DL system-level gains are realized through spectrally-efficient retransmissions enabled by the proposed enriched HARQ feedback (EHF) mechanism, where the UE reports a delta MCS -- indicating the adjustment relative to the previous (re)transmission -- to support adaptive retransmission LA. The most spectrally efficient delta is selected using an optimization metric and encoded into a fixed-size feedback using different quantization methods, with the advanced variant tailoring delta-MCS levels to each UE cluster and feedback resolution, resulting in significant improvements in DL physical resource block utilization. To balance feedback overhead, the flexible variant jointly optimizes delta selection and feedback size, reducing UL feedback by roughly 25% for retransmissions. It also achieves around 25% median DL throughput gain over baseline HARQ with Chase combining in dense urban-macro deployment, enhancing overall system capacity for throughput-sensitive eMBB traffic with minimal computational cost at both UE and gNB.

After the defence there will be a small reception at Fredrik Bajers Vej 7A 1.103