Kwang Soon Kim

While machine learning is widely used to optimize wireless networks, training a separate model for each task in communication and localization is becoming increasingly unsustainable due to the significant costs associated with training and deployment. Foundation models offer a more scalable alternative by enabling a single model to be adapted across multiple tasks through fine-tuning with limited samples. However, current foundation models mostly rely on large-scale Transformer architectures, resulting in computationally intensive models unsuitable for deployment on typical edge devices. This paper presents a lightweight foundation model based on simple Multi-Layer-Perceptron (MLP) encoders that independently process input patches. Our model supports 4 types of downstream tasks (long-range technology recognition, short-range technology recognition, modulation recognition and line-of-sight-detection) from multiple input types (IQ and CIR) and different sampling rates. We show that, unlike Transformers, which can exhibit performance drops as downstream tasks are added, our MLP model maintains robust generalization performance, achieving over 97% accurate fine-tuning results for previously unseen data classes. These results are achieved despite having only 21K trainable parameters, allowing an inference time of 0.33 ms on common edge devices, making the model suitable for constrained real-time deployments.

6G hyper reliable low-latency communication (HRLLC) services with high data rates require higher levels of quality of service (QoS) guarantees along with high spectral efficiency (SE). To meet these demands, extreme massive multi-input multi-output (E-MIMO) scheduling with large spatial dimensions by leveraging channel distribution information (CDI) is critical. However, a practical E-MIMO CDI acquisition may suffer from diverse channel environments, large CDI parameter sizes, and marginal channel estimation, resulting in large CDI feedback overhead, significant QoS mismatch, and SE loss. In this paper, an AI-integrated E-MIMO scheduling method is proposed to resolve these issues by jointly designing compact CDI management, AI-based reliable scheduling performance prediction, and scheduling optimization, which enables efficient CDI reporting and pre-configured resource allocation for providing HRLLC QoS with high SE. Its design and performance advantages are demonstrated in a correlated Rayleigh E-MIMO scenario.

Hydra radio access networks (H-RANs) are poised to revolutionize the way data is gathered, transmitted, and analyzed, ushering in an era of unprecedented connectivity, efficiency, and intelligence. By proactively managing changes and addressing potential disruptions, H-RAN networks aim to provide consistent service quality, even under varied operating conditions. This means that networks can automatically detect issues, assess their impact, implement corrective measures, and modify operating parameters in response to variations in user behavior or environmental factors. However, conventional RANs typically implement a uniform strategy when dealing with diverse communication scenarios. This one-size-fits-all methodology limits their ability to autonomously adjust to the unique requirements of the millimeter and terahertz wave (MMW/THz) communications environment, which is characterized by high sensitivity to environmental factors. As a result, these systems face significant challenges in maintaining a reliable and efficient communication link under varying conditions. Furthermore, conventional RANs typically require considerable manual tuning and configuration, making them time-consuming and costly. Indeed, while the potential of H-RAN networks encompasses numerous improvements and innovations across various aspects of next-generation technologies, this paper addresses the initial access implementation and data transmission approaches under the non-line-of-sight (NLoS) of the level₁ case study. Our solution contains several novelties that enhance both overhead and model accuracy. To this end, we have developed a novel adaptive environment-aware blockage beamforming codebook structure. This codebook consists of edge-line-of-sight (ELoS) vectors and reconfigurable intelligent surface (RIS) vectors, RIS-LoS (SLoS). These elements are integral to optimizing beamforming in scenarios where direct line-of-sight (LoS) is obstructed by level₁ blockage scenario. Furthermore, we employed a multi-sparse input and multi-task learning (SMTL) framework in the Hydra distributed unit (H-DU) artificial intelligence and machine learning (AI/ML) (AI/ML D-engine), where each task is tailored to be executed in a particular environment based on online feedback. The SMTL model employs an approach to handling sparse and heterogeneous input data while performing multiple tasks based on the current network condition. More specifically, a novel task₂ is developed in this study to generate a list of recommended beamforming vectors and patterns designed to address the challenges arising from blockage scenarios of level₁ in MMW/THz environments.

To enable widespread use of electric vehicles (EVs), large-scale public charging stations with fast chargers are being planned in places such as shopping malls and office car parks. Operators of public charging stations need to utilize EV scheduling algorithms that can satisfy charging demands with a minimum number of simultaneous charging sessions. In this paper, we propose EV charging scheduling algorithms that meet the charging demands and deadlines of EV users while minimizing the number of simultaneous charging sessions. An uncertainty-aware deep learning (DL) framework is also used to predict EV arrivals at a charging station. The predicted EV arrivals in turn are used to help the charging operator estimate how many charging sessions to order from the grid. Our DL model not only predicts the mean EV arrival rates but also the upper limits of EV arrivals, which enhances robustness against uncertainty in EV arrivals and helps estimate the maximum charging demand for a given interval. Moreover, to overcome the challenge of insufficient EV charging data for DL models, we construct a synthetic data model that takes into account multiple factors influencing EV arrivals, such as weather, events, weekdays, and weekends. Both online and offline approaches in the design of EV scheduling algorithms are utilized. The performances of the proposed algorithms are evaluated in terms of active charging sessions used to serve EV users. We also compare their performance with a baseline algorithm which is an offline optimal algorithm based on a mixed integer linear problem formulation.

Next-generation sensor and radio access networks (NG-SRANs) namely, Hydra radio access networks (H-RANs) represent a significant evolution in the telecommunications and sensor ecosystem landscape in anticipation of 6G deployment and beyond. H-RAN’s vision derives its strength from integrating various technologies and networks into a single central network with the widespread incorporation of artificial intelligence (AI) technologies throughout the network. As a result, H-RAN’s unique features and characteristics can serve as a baseline for innovating new applications and significantly enhance the overall functions of conventional open radio access networks (O-RANs). However, among the many improvements and innovations that the H-RAN architecture promises in its functionality, this paper focuses on the initial access implementation "Task1" approach. Our solution contains several novelties that enhance both overhead and model accuracy. To this end, we define a novel intelligent perception network inspired by the knowledge distribution idea for collaborative H-RAN networks. We develop sparse multi-task learning (SMTL) as part of the AI/ML D-engine for federated learning to perform multiple tasks simultaneously. The SMTL is designed to select the optimal solution from a list of recommended solutions, namely "Tasks". In the simulation, figures of merit include metrics such as top-k validation accuracy, beam selection accuracy, throughput ratios, beam sweep time, latency, and initial access times, which are used to evaluate the performance and efficiency of the proposed technologies. Simulation results demonstrate that by exploiting contextual information from distributed collaborative SRUs, and UE sends its own sensing information via a physical random-access channel in addition to using SMTL, our H-RAN-based initial access scheme can achieve 82.9% throughput of an exhaustive beam search (EBS) based-O-RAN network without any beam search overhead and 96.7% by searching among as few as 5 beams. Compared to the conventional MMW 5G-NR solution, our proposed method significantly minimizes the beam search time needed to reach the desired throughput.

An edge intelligence-aided Internet-of-Things (IoT) network has been proposed to accelerate the response of IoT services by deploying edge intelligence near IoT devices. The transmission of data from IoT devices to the edge nodes leads to large network traffic in the wireless connections. Federated Learning (FL) is proposed to solve the high computational complexity by training the model locally on IoT devices and sharing the model parameters in the edge nodes. This paper focuses on developing an efficient integration of joint edge intelligence nodes depending on investigating an energy-efficient bandwidth allocation, computing Central Processing Unit (CPU) frequency, optimization transmission power, and the desired level of learning accuracy to minimize the energy consumption and satisfy the FL time requirement for all IoT devices. The proposal efficiently optimized the computation frequency allocation and reduced energy consumption in IoT devices by solving the bandwidth optimization problem in closed form. The remaining computational frequency allocation, transmission power allocation, and loss could be resolved with an Alternative Direction Algorithm (ADA) to reduce energy consumption and complexity at every iteration of FL time from IoT devices to edge intelligence nodes. The simulation results indicated that the proposed ADA can adapt the central processing unit frequency and power transmission control to reduce energy consumption at the cost of a small growth of FL time.

Ultra-reliable and Low-latency Communications (URLLC) is expected to be one of the most critical characteristics in Beyond fifth-Generation (B5G) cellular networks with stringent low latency and high-reliability requirements. The Deep Reinforcement Learning (deep-RL) framework has been applied to predict the optimization of a Resource Block (RB) and minimize Power Allocation (PA) to guarantee a high End-to-End (E2E) reliability and low E2E latency under rate constraints. This paper proposes a novel Policy Gradient-based Actor-Critic Learning (PGACL) algorithm to optimize the policy gradient for optimal rate allocation to solve the RB, minimize power, and guarantee a solution for URLLC scheduling. The purpose of a PGACL algorithm is to provide a good policy with a closer convergence rate and a low computational cost depending on the reduced action space for every user. URLLC systems need to operate in highly reliable systems and account for extreme network conditions. Therefore, we proposed the refiner Generative Adversarial Networks (GANs) that apply enough extreme events for the deep-RL agent to generate synthetic data with high reliability similar to real data based on the regulated number of extreme events in the dataset. This refiner GAN method enables a deep-RL approach to generate large amounts of data practically used in real-time operation. Simulation results showed that the proposed deep-RL for refiner-GAN can omit the transient training time and develop deep-learning based on a controlled set of unlabeled real traffic at a relatively short time. Furthermore, the refiner GAN demonstrated 99.9999% reliability and E2E latency of less than 1.4ms.

The convergence of Artificial Intelligence (AI) can overcome the complexity of network defects and support a sustainable and green system. AI has been used in the Cognitive Internet of Things (CIoT), improving a large volume of data, minimizing energy consumption, managing traffic, and storing data. However, improving smart packet transmission scheduling (TS) in CIoT is dependent on choosing an optimum channel with a minimum estimated Packet Error Rate (PER), packet delays caused by channel errors, and the subsequent retransmissions. Therefore, we propose a Generative Adversarial Network and Deep Distribution Q Network (GAN-DDQN) to enhance smart packet TS by reducing the distance between the estimated and target action-value particles. Furthermore, GAN-DDQN training based on reward clipping is used to evaluate the value of each action for certain states to avoid large variations in the target action value. The simulation results show that the proposed GAN-DDQN increases throughput and transmission packet while reducing power consumption and Transmission Delay (TD) when compared to fuzzy Radial Basis Function (fuzzy-RBF) and Distributional Q-Network (DQN). Furthermore, GAN-DDQN provides a high rate of 38 Mbps, compared to actor-critic fuzzy-RBF’s rate of 30 Mbps and the DQN algorithm’s rate of 19 Mbps.