Junil Choi

This paper investigates a practical precoding scheme for rate-splitting multiple access (RSMA) in multi-user multiple-input single-output (MU-MISO) systems where the nonlinear characteristics of power amplifiers are explicitly taken into account. To enhance performance in the presence of distortion caused by nonlinear power amplifiers, we propose an iterative accelerated gradient-based precoding scheme. The proposed precoding scheme is further integrated with a novel power scaling function that dynamically adjusts transmit power to satisfy practical power consumption constraints while maximizing an achievable sum rate. The gradient of the achievable sum rate for the considered system is derived to support efficient gradient-based optimization. We also prove that the proposed iterative precoding scheme converges to a critical point. Simulation results demonstrate that the proposed precoding scheme consistently outperforms modified conventional schemes including distortion-aware and weighted minimum mean square error beamforming, across both overloaded and underloaded scenarios. In particular, the proposed scheme demonstrates consistent performance in high-power regimes, where conventional schemes exhibit severe degradation due to distortion. Furthermore, the proposed precoding scheme exhibits faster convergence compared to the conventional benchmark.

Precise channel state knowledge is crucial in future wireless communication systems, which drives the need for accurate channel prediction without additional pilot overhead. While machine-learning (ML) methods for channel prediction show potential, existing approaches have limitations in their capability to adapt to environmental changes due to their extensive training requirements. In this paper, we introduce the channel prediction approaches in terms of the temporal channel prediction and the environmental adaptation. Then, we elaborate on the use of the advanced ML-based channel prediction to resolve the issues in traditional ML methods. We also analyze the training process, dataset characteristics, and the influence of source tasks and pre-trained models on channel prediction performance, demonstrating their effects under different adaptation samples and environments. Furthermore, we propose practical model selection criteria based on latency constraints, data availability, and environmental dynamics to support effective deployment in wireless communications. Finally, we discuss open challenges and possible future research directions of ML-based channel prediction.

Millimeter wave (mmWave) multiple-input multiple-output (MIMO) communication systems require accurate channel estimation for effective beamforming. Conventional compressed sensing (CS) algorithms utilize the inherent sparsity information of channels but suffer from basis mismatch problems for on-grid methods and high computational complexity for off-grid methods. Recently, generative model-based CS frameworks have shown promising results by directly learning the distribution of channel realizations, eliminating the need for prior sparsity information. However, these approaches rely on non-convex optimization in latent space that may converge to suboptimal solutions and exhibit generalization problems across diverse channel environments. Moreover, existing generative approaches have not adequately addressed the quantization effects of practical finite-resolution analog-to-digital converters (ADCs). To address these limitations, we propose a novel generative CS (GCS) algorithm using a conditional variational autoencoder (CVAE) that learns fundamental path components of mmWave MIMO channels rather than complete channel realizations while considering finite-resolution quantization. The CVAE is trained on single-path measurements with spatial frequency conditioning variables, enabling generalization across diverse channel environments without requiring environment-specific training data. Our algorithm directly encodes quantized received pilot signals and angular conditioning variables into latent representations, which are then fed into a decoder to output reconstructions conditioned on the angular parameters, with the decoder trained to be robust to perturbations from additive and quantization noise. The algorithm sequentially identifies path components by searching through spatial frequency codebooks with different conditioning values and applying gradient-based refinement for off-grid accuracy. Numerical results demonstrate that the proposed GCS algorithm outperforms conventional CS algorithms across various scenarios.

In this paper, we propose a novel expected sum rate maximization algorithm called robust beam split and angular uncertainty (RoBA) for a terahertz (THz) hybrid precoding downlink system. To mitigate the effects of beam split and angular uncertainty that occur in a practical wideband THz system, we introduce an array covariance matrix at each sub-carrier. This matrix captures the impact of frequency-dependent array responses and statistical angular errors, enabling tractable optimization of the expected sum rate. The resulting constant modulus analog precoder design problem is solved using the Riemannian conjugate gradient (RCG) algorithm. Simulation results show that the proposed RoBA algorithm significantly improves the expected sum rate compared to existing benchmarks.

This paper investigates a simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS)-aided frequency division duplexing (FDD) communication system. Since downlink (DL) and uplink (UL) signals are transmitted simultaneously in FDD, the transmission and reflection coefficients of the STAR-RIS should be optimized to support both transmissions. Considering when a single base station with multi-antenna supports multiple users, we formulate a weighted sum-rate maximization problem to consider the relative priority between the DL and UL, as well as among users. To effectively tackle the formulated non-convex optimization problem, we adopt an alternating optimization framework that decomposes it into tractable sub-problems, where the beamformers at the base station, the UL power allocation for users, and the transmission and reflection coefficients of the STAR-RIS are iteratively updated. Within this framework, successive convex approximation-based technique is proposed to optimize the transmission and reflection coefficients of the STAR-RIS. Simulation results demonstrate that the proposed optimization framework achieves superior performance compared to benchmarks and effectively balances transmission priority.

In this paper, the millimeter wave (mmWave) wireless power transfer (WPT) system is considered where the wireless charging station (WCS) operates without prior knowledge of the wireless charging device (WCD)’s location. A hierarchical beam selection algorithm is proposed, which employs hierarchical beams with varying beamwidths to determine the optimal beam for the WCD while mitigating error propagation (EP) across levels. The proposed algorithm consists of the beam identification process and the verification process. In the beam identification process, the WCS identifies the beam that achieves the highest received signal strength indicator (RSSI). In the verification process, the WCS validates the selected beam using the verification beam set, which is constructed based on the beam with the second-highest RSSI. This process mitigates EP across hierarchical levels. The proposed algorithm is implemented on the commercial mmWave RF module, and its performance is evaluated through experiments in the indoor laboratory environment.

In this paper, we configure a cell-free massive multiple-input multiple-output (MIMO) systems with reconfigurable intelligent surface (RIS). Considering dominant line-of-sight (LoS) paths and array structures of antennas of access points (APs) and elements of RIS, spatially correlated Rician fading is employed for channel models. We analyze channel estimation and data transmissions to compute statistics of aggregate channels and signals. Based on the statistical CSI, we derive closed-form expressions of uplink and downlink spectral efficiencies to evaluate system performance. Simulation results verify the analytical expressions accord with numerical experiments. Effects with a large number of system parameters are examined for several channel scenarios, e.g., full links, no direct links, or correlated Rayleigh fading, and utility of optimizing RIS phase shifts is also analyzed.

Federated learning (FL) and federated distillation (FD) are distributed learning paradigms that train UE models with enhanced privacy, each offering different trade-offs between noise robustness and learning speed. To mitigate their respective weaknesses, we propose a hybrid federated learning (HFL) framework in which each user equipment (UE) transmits either gradients or logits, and the base station (BS) selects the per-round weights of FL and FD updates. We derive convergence of HFL framework and introduce two methods to exploit degrees of freedom (DoF) in HFL, which are (i) adaptive UE clustering via Jenks optimization and (ii) adaptive weight selection via a damped Newton method. Numerical results show that HFL achieves superior test accuracy at low SNR when both DoF are exploited.