Federated learning (FL) is a privacy-preserving machine learning algorithm that enables multiple clients to collaborate. To respond to non-independent and identically distributed (non-IID) environments between clients, personalized FL (PFL) has been actively investigated. The typical PFL model consists of two parts: 1) the head (i.e., classifier) for the final classification and 2) the body (i.e., feature extractor) for extracting representations from local datasets. The head is maintained separately in each client for personalization; the body is aggregated for generalization. FedSDP introduces a bridge layer, called a personalized layer, between the head and the body to preserve individual, non-shared local prototypes for each client. A personalized layer decouples the body and head, strengthening the generalization and personalization, respectively. Based on this architecture, this study proposes a new PFL framework, Federated Self-Derived Prototypes (FedSDP), to dynamically balance personalization and generalization. To this end, we introduce two dynamic adjustments for generating self-derived prototypes: 1) global-local similarity weight (GL-Sim Weight) and 2) personalization early stopping indicator (P-Stop Indicator). GL-Sim Weight based on the similarity between the global and local prototypes is utilized to adjust the degree of personalization of each local model. PStop Indicator is calculated based on the changed degree of local parameters in each client, determining the early stopping for personalization in the client and further concentrating on generalization. Our comprehensive experiments demonstrate that FedSDP outperforms existing state-of-the-art FL frameworks, showing superior effectiveness in non-IID settings. Our code and data are available at https://github.com/bigbases/FedSDP.

Terahertz (THz) communications have become an important technology for achieving 6G networks that require higher data rates. To overcome severe signal attenuation in THz communications, highly directional beamforming technique using extremely large-scale antenna array (XL-array) is widely used at the base station (BS). Since the channel exhibits near-field characteristics at higher frequencies and wider antenna apertures, the BS must perform near-field beam training to obtain distance information as well as angle information of the channel. A major issue of conventional near-field beam training is the considerable beam training latency due to the large overhead incurred by angle and distance domain search. To address this issue, we propose a subarray-based near-field beam training that estimates the user location by performing frequency-dependent beam training on partitioned array antennas. In the first step, BS simultaneously searches multiple directions by generating multiple frequency-dependent beams using phase shifters (PSs) and true time delays (TTDs). In the second step, BS estimates the user angle and distance information by exploiting the extracted directions from subarrays. Numerical results show that the proposed scheme achieves 28% improvement in the data rate and 99% reduction in the latency.

To integrate a variety of data-demanding and delay-sensitive applications, millimeter-wave (mmWave) massive multiple-input multiple-output (MIMO) system has emerged as a key enabler for 6G wireless communications. To fully exploit mmWave massive MIMO system, acquisition of accurate channel information is of great importance while such is challenging due to short coherence time of mmWave channel. Therefore, to obtain accurate channel information in a real-time manner, we propose sensing-aided multi-modal channel prediction technique (SMCPT). By analyzing sensing images, we can accurately and quickly extract the positions of mobile devices as well as the positions, orientations, and materials of entities affecting signal propagation, thus enhancing channel prediction accuracy. From practical experiments, we show that SMCPT achieves more than 69% channel prediction accuracy gain over conventional schemes.

Ultra-dense networks (UDN) are anticipated to fulfill the high-performance demands for future applications of 5G and beyond communication networks. By exploiting recent advances in computer vision techniques, we propose a novel cell association technique, referred to as computer vision-based cell association (CV-CA), that circumvents the cumbersome CSI acquisition. The computer vision technique provides accurate 3D position and determines the line-of-sight communication link of all users within the UDN using RGB images. We demonstrate from the simulations that the proposed CV-CA outperforms 5G-NR and conventional cell association techniques in terms of total data rate.

To meet the ever-increasing data rate demand expected in 6G networks, terahertz (THz) ultra-massive (UM) multiple-input multiple-output (MIMO) systems have gained much attention recently. One notable aspect of these systems is that the deployment of an extremely large-scale antenna array and high transmission frequency result in an expansion of the near-field region where the electromagnetic (EM) radiation is modeled as a spherical wave. In the near-field region, the channel becomes a function of a position of a user equipment (UE) rather than the direction, giving rise to a beam focusing operation that focuses the signal power onto the specific position. However, the traditional approaches relying on the sweeping of discretized beam codewords cannot support this ultra-sharp beam focusing operation in THz UM-MIMO systems. This paper proposes a novel beam focusing technique based on sensing and computer vision (CV) technologies. The essence of the proposed scheme is to estimate the UE’s position from the vision information using the CV technique and then generates the beam heading towards the estimated position. By replacing the discretized and time-consuming beam sweeping operation with a highly precise CV-based positioning, the positioning accuracy as well as the beam focusing gain can be improved significantly. Numerical results show that the proposed scheme achieves significant positioning accuracy and data rate gains over the conventional codebook-based beam focusing schemes.

Terahertz (THz) communication is envisaged as an attractive way to attain abundant spectrum resources for 6G wireless communications. One main difficulty of the THz communications is the severe attenuation of signal power caused by the high diffraction and penetration losses and atmospheric absorption. To compensate for the severe path loss, a beamforming technique realized by the massive multiple-input multiple-output (MIMO) has been widely used. Since the beamforming gain is maximized only when the beams are appropriately aligned with the signal propagation paths, acquisition of accurate beam directions is of paramount importance. A major issue of the conventional beam management schemes is the considerable latency being proportional to the number of training beams. In this paper, we propose a THz beam management technique that simultaneously generates multiple frequency-dependent beams using the true time delay (TTD)-based phase shifters. By closing the gap between the frequency-dependent beamforming vectors and the desired directional beamforming vectors using the TTD-based signal propagation network called intensifier, we generate very sharp training beams maximizing the beamforming gain. From the numerical results, we demonstrate that the proposed scheme achieves more than 70% reduction in the beam management latency and 60% increase in the data rate.

Terahertz (THz) multiple-input multiple-output (MIMO) beamforming is a key technology to support immersive mobile services in 6G communication systems. In the beamforming vector generation, the analog phase shifter which generates the phase invariant to the signal frequency have been widely used. However, since the spatial directions of the subcarrier channels are functions of the subcarrier frequency in the wideband THz systems, the beamforming techniques based on the analog phase shifters suffer from a severe beamforming gain loss. In this paper, we propose a novel beamforming technique that exploits the true time delay (TTD)-based phase shifters to generate multiple frequency-dependent beamforming vectors for the wideband THz systems. Intriguing feature of the proposed scheme is to exploit a deliberately designed TTD-based signal propagation network called calibrator to bridge the gap between the desired beamforming vectors and the frequency-dependent beamforming vectors. In doing so, the signal power is concentrated onto the mainlobe so that the generated beamforming vectors can achieve the maximum beamforming gain. From the numerical results, we demonstrate that the proposed scheme achieves more than 80% data rate gain over the conventional beamforming schemes.

Daily peak load forecasting (DPLF) is critical in smart grid applications for security analysis, unit commitment, and scheduling of outages and fuel supplies. Although excellent single machine learning methods using tree-based ensemble learning or deep learning have shown satisfactory performance for DPLF, there is still room for improvement. This study proposes a hybrid tree-based ensemble learning model, called HYTREM, for robust DPLF. We first collected two commercial buildings’ energy consumption data from publicly available datasets. We then performed data preprocessing, such as input variable configuration, for the HYTREM modeling. We divided both datasets into training and test sets and generated the prediction values of several tree-based ensemble learning models, such as gradient boosting machine, extreme gradient boosting, Cubist, and random forest (RF), for each set as novel input variables. We reconstructed datasets using the Boruta algorithm to select all the relevant features and built an online RF model trained on these datasets using time-series cross-validation for day-ahead DPLF. The experimental results showed that the HYTREM performed a better performance than tree-based ensemble and deep learning methods in building-level DPLF in terms of the mean absolute percentage error and normalized root mean square error.

In this paper, we address the Online Unsupervised Domain Adaptation (OUDA) problem and propose a novel multi-stage framework to solve real-world situations when the target data are unlabeled and arriving online sequentially in batches. Most of the traditional manifold-based methods on the OUDA problem focus on transforming each arriving target data to the source domain without sufficiently considering the temporal coherency and accumulative statistics among the arriving target data. In order to project the data from the source and the target domains to a common subspace and manipulate the projected data in real-time, our proposed framework institutes a novel method, called an Incremental Computation of Mean-Subspace (ICMS) technique, which computes an approximation of mean-target subspace on a Grassmann manifold and is proven to be a close approximate to the Karcher mean. Furthermore, the transformation matrix computed from the mean-target subspace is applied to the next target data in the recursive-feedback stage, aligning the target data closer to the source domain. The computation of transformation matrix and the prediction of next-target subspace leverage the performance of the recursive-feedback stage by considering the cumulative temporal dependency among the flow of the target subspace on the Grassmann manifold. The labels of the transformed target data are predicted by the pre-trained source classifier, then the classifier is updated by the transformed data and predicted labels. Extensive experiments on six datasets were conducted to investigate in depth the effect and contribution of each stage in our proposed framework and its performance over previous approaches in terms of classification accuracy and computational speed. In addition, the experiments on traditional manifold-based learning models and neural-network-based learning models demonstrated the applicability of our proposed framework for various types of learning models.