Advanced grant-free NOMA and hybrid beamforming techniques for 5G and beyond

Mobile cellular communication has evolved significantly since the advent of the first-generation networks in the 1980s. Fifth-generation (5G) networks were designed to support three critical use cases: massive machine-type communication (mMTC), enhanced mobile broadband (eMBB), and ultra-reliable low-latency communication (URLLC). These use cases continue to expand with the emergence of sixth-generation (6G) networks. This thesis advances key technologies essential for 5G and beyond, focusing on three pivotal topics in collaboration with industry partners. The first part addresses mMTC by exploring the integration of grant-free (GF) transmission with non-orthogonal multiple access (NOMA), referred to as GF-NOMA, to leverage the strengths of both techniques. A primary challenge in GF-NOMA is mitigating high packet drop rates due to decentralized user activity. To address this, we propose uplink GF-NOMA schemes incorporating hybrid automatic repeat request (HARQ) with two packet combining techniques—chase combining and incremental redundancy. Additionally, we introduce a grant-free single-transmission scheme that concurrently transmits all redundancy versions of a packet. Comparative evaluations confirm that our proposed strategies outperform conventional grant-based and existing GF-NOMA approaches in mMTC scenarios. The second part focuses on eMBB and investigates a deep learning (DL)-based hybrid beamforming (HBF) strategy to mitigate phase noise in millimeter-wave transmission systems caused by local oscillator instabilities. Our method employs a deep neural network to optimize precoding and combining matrices based on channel state information, integrating an adaptive attention mechanism that adjusts per symbol to counteract phase noise. The approach also considers practical constraints such as low-resolution phase shifters and imperfect channel estimation. Simulation results demonstrate substantial data rate improvements, outperforming conventional techniques in scenarios affected by phase noise and compounded distortions. The third part extends DL-based HBF to high-mobility scenarios where users traverse multiple environments, causing significant channel distribution shifts that necessitate efficient online fine-tuning. This work introduces low-rank adaptation (LoRA) fine-tuning for beamforming, establishing a novel online re-training paradigm applicable beyond our proposed model. Extensive simulations across various channel conditions confirm that LoRA-based fine-tuning outperforms meta-learning and direct transfer learning, further validating its effectiveness. Together, these contributions advance data-driven solutions for wireless networks, addressing critical challenges in 5G and beyond.