arXiv subjects

Chung G. Kang

Publications and source records attributed to Chung G. Kang.

15 recordsLinked to original sources

arXiv2026

Generative AI-Enabled Mission-Aware Radio Orchestration for RIS-Assisted LEO Satellite ISAC Systems

Mission-adaptive low-Earth-orbit (LEO) satellite networks with integrated sensing and communication (ISAC) must retarget radio resources as operator goals change. To enable this adaptation from flexible operator language, we develop a generative-AI-enabled radio-orchestration framework in which a large language model (LLM) maps each mission into a structured policy comprising communication, sensing, and fairness weights, mandatory quality-of-service thresholds, power-allocation guidance, and solver initialization. Deterministic validation and physical-layer optimization then enforce feasibility and realize the policy through beam, power, and reconfigurable intelligent surface (RIS) configuration. This mixed-timescale design uses generative AI for semantic adaptation at the mission timescale while retaining conventional wireless optimization at the faster channel timescale. We compare zero-shot (LLM-ZS) and in-context (LLM-ICL) operation on familiar and held-out compositional missions. On held-out instructions, LLM-ZS and LLM-ICL achieve $91.7\%$ and $94.4\%$ priority-order accuracy, respectively, with ICL mainly improving numerical calibration. Their downstream radio-performance difference is statistically unresolved because both usually recover the hard constraints that determine admissible actions. Accordingly, LLM-ZS is the low-context default, while LLM-ICL is useful for semantically difficult missions requiring finer calibration. Explicit alternating optimization preserves the qualitative ordering when active beams and RIS phases are optimized directly. The results show how generative AI can enhance next-generation radio orchestration without replacing feasibility-critical physical-layer optimization.

cs.NI↗

arXiv2026

Segment-Wise Flow Matching for Vision-Aided mmWave V2I Beam Prediction

This paper proposes a vision-conditioned flow matching (FM) framework for beam prediction in millimeter-wave vehicle-to-infrastructure links. Instead of modeling discrete beam-index sequences, the proposed method learns the temporal evolution of normalized beam receive power vectors through a continuous vector field governed by an ordinary differential equation, enabling smooth dynamics and efficient sampling. By imposing FM over beam-state transitions and jointly optimizing beam prediction and flow consistency, the proposed framework provides a unified model for future beam prediction. Experimental results show that the proposed FM-based model significantly improves beam prediction performance over baselines, approaches the performance of large language model-based methods, and reduces predictor-side inference latency by about $6.9\times$ on GPU and $2.8\times10^3\times$ on CPU, respectively.

eess.SP↗

arXiv2025

Multimodal Radio and Vision Fusion for Robust Localization in Urban V2I Communications

Accurate localization is critical for vehicle-to-infrastructure (V2I) communication systems, especially in urban areas where GPS signals are often obstructed by tall buildings, leading to significant positioning errors, necessitating alternative or complementary techniques for reliable and precise positioning in applications like autonomous driving and smart city infrastructure. This paper proposes a multimodal contrastive learning regression based localization framework for V2I scenarios that combines channel state information (CSI) with visual information to achieve improved accuracy and reliability. The approach leverages the complementary strengths of wireless and visual data to overcome the limitations of traditional localization methods, offering a robust solution for V2I applications. Simulation results demonstrate that the proposed CSI and vision fusion model significantly outperforms traditional methods and single modal models, achieving superior localization accuracy and precision in complex urban environments.

eess.SP↗

arXiv2025

A Variational Bayesian Detector for Affine Frequency Division Multiplexing

This paper proposes a variational Bayesian (VB) detector for affine frequency division multiplexing (AFDM) systems. The proposed method estimates the symbol probability distribution by minimizing the Kullback-Leibler (KL) divergence between the true posterior and an approximate distribution, thereby enabling low-complexity soft-decision detection. Compared to conventional approaches such as zero-forcing (ZF), Linear minimum mean square rrror (LMMSE), and the message passing algorithm (MPA), the proposed detector demonstrates lower bit error rates (BER), faster convergence, and improved robustness under complex multipath channels. Simulation results confirm its dual advantages in computational efficiency and detection performance.

eess.SP↗

arXiv2025

BeamLLM: Vision-Empowered mmWave Beam Prediction with Large Language Models

In this paper, we propose BeamLLM, a vision-aided millimeter-wave (mmWave) beam prediction framework leveraging large language models (LLMs) to address the challenges of high training overhead and latency in mmWave communication systems. By combining computer vision (CV) with LLMs' cross-modal reasoning capabilities, the framework extracts user equipment (UE) positional features from RGB images and aligns visual-temporal features with LLMs' semantic space through reprogramming techniques. Evaluated on a realistic vehicle-to-infrastructure (V2I) scenario, the proposed method achieves 61.01% top-1 accuracy and 97.39% top-3 accuracy in standard prediction tasks, significantly outperforming traditional deep learning models. In few-shot prediction scenarios, the performance degradation is limited to 12.56% (top-1) and 5.55% (top-3) from time sample 1 to 10, demonstrating superior prediction capability.

cs.LG↗

arXiv2025

M2BeamLLM: Multimodal Sensing-empowered mmWave Beam Prediction with Large Language Models

This paper introduces a novel neural network framework called M2BeamLLM for beam prediction in millimeter-wave (mmWave) massive multi-input multi-output (mMIMO) communication systems. M2BeamLLM integrates multi-modal sensor data, including images, radar, LiDAR, and GPS, leveraging the powerful reasoning capabilities of large language models (LLMs) such as GPT-2 for beam prediction. By combining sensing data encoding, multimodal alignment and fusion, and supervised fine-tuning (SFT), M2BeamLLM achieves significantly higher beam prediction accuracy and robustness, demonstrably outperforming traditional deep learning (DL) models in both standard and few-shot scenarios. Furthermore, its prediction performance consistently improves with increased diversity in sensing modalities. Our study provides an efficient and intelligent beam prediction solution for vehicle-to-infrastructure (V2I) mmWave communication systems.

cs.CL↗

arXiv2023

Multi-Agent Reinforcement Learning for Distributed Joint Communication and Computing Resource Allocation over Cell-Free Massive MIMO-enabled Mobile Edge Computing Network

To support the newly introduced multimedia services with ultra-low latency and extensive computation requirements, resource-constrained end user devices should utilize the ubiquitous computing resources available at network edge for augmenting on-board (local) processing with edge computing. In this regard, the capability of cell-free massive MIMO to provide reliable access links by guaranteeing uniform quality of service without cell edge can be exploited for seamless parallel processing. Taking this into account, we consider a cell-free massive MIMO-enabled mobile edge network to meet the stringent requirements of the advanced services. For the considered mobile edge network, we formulate a joint communication and computing resource allocation (JCCRA) problem with the objective of minimizing energy consumption of the users while meeting the tight delay constraints. We then propose a fully distributed cooperative solution approach based on multiagent deep deterministic policy gradient (MADDPG) algorithm. The simulation results demonstrate that the performance of the proposed distributed approach has converged to that of a centralized deep deterministic policy gradient (DDPG)-based target benchmark, while alleviating the large overhead associated with the latter. Furthermore, it has been shown that our approach significantly outperforms heuristic baselines in terms of energy efficiency, roughly up to 5 times less total energy consumption.

cs.NI↗

arXiv2023

Deep Learning-based Data-aided Activity Detection with Extraction Network in Grant-free Sparse Code Multiple Access Systems

This letter proposes a deep learning-based data-aided active user detection network (D-AUDN) for grant-free sparse code multiple access (SCMA) systems that leverages both SCMA codebook and Zadoff-Chu preamble for activity detection. Due to disparate data and preamble distribution as well as codebook collision, existing D-AUDNs experience performance degradation when multiple preambles are associated with each codebook. To address this, a user activity extraction network (UAEN) is integrated within the D-AUDN to extract a-priori activity information from the codebook, improving activity detection of the associated preambles. Additionally, efficient SCMA codebook design and Zadoff-Chu preamble association are considered to further enhance performance.

cs.IT↗

arXiv2022

On the Performance of Deep Learning-based Data-aided Active User Detection for GF-SCMA System

The recent works on a deep learning (DL)-based joint design of preamble set for the transmitters and data-aided active user detection (AUD) in the receiver has demonstrated a significant performance improvement for grant-free sparse code multiple access (GF-SCMA) system. The autoencoder for the joint design can be trained only in a given environment, but in an actual situation where the operating environment is constantly changing, it is difficult to optimize the preamble set for every possible environment. Therefore, a conventional, yet general approach may implement the data-aided AUD while relying on the preamble set that is designed independently rather than the joint design. In this paper, the activity detection error rate (ADER) performance of the data-aided AUD subject to the two preamble designs, i.e., independently designed preamble and jointly designed preamble, were directly compared. Fortunately, it was found that the performance loss in the data-aided AUD induced by the independent preamble design is limited to only 1dB. Furthermore, such performance characteristics of jointly designed preamble set is interpreted through average cross-correlation among the preambles associated with the same codebook (CB) (average intra-CB cross-correlation) and average cross-correlation among preambles associated with the different CBs (average inter-CB cross-correlation).

cs.IT↗

arXiv2022

DRL-based Distributed Resource Allocation for Edge Computing in Cell-Free Massive MIMO Network

In this paper, with the aim of addressing the stringent computing and quality-of-service (QoS) requirements of recently introduced advanced multimedia services, we consider a cell-free massive MIMO-enabled mobile edge network. In particular, benefited from the reliable cell-free links to offload intensive computation to the edge server, resource-constrained end-users can augment on-board (local) processing with edge computing. To this end, we formulate a joint communication and computing resource allocation (JCCRA) problem to minimize the total energy consumption of the users, while meeting the respective user-specific deadlines. To tackle the problem, we propose a fully distributed solution approach based on cooperative multi-agent reinforcement learning framework, wherein each user is implemented as a learning agent to make joint resource allocation relying on local information only. The simulation results demonstrate that the performance of the proposed distributed approach outperforms the heuristic baselines, converging to a centralized target benchmark, without resorting to large overhead. Moreover, we showed that the proposed algorithm has performed significantly better in cell-free system as compared with the cellular MEC systems, e.g., a small cell-based MEC system.

cs.NI↗

arXiv2022

Data-aided Active User Detection with a User Activity Extraction Network for Grant-free SCMA Systems

In grant-free sparse code multiple access (GF-SCMA) system, active user detection (AUD) is a major performance bottleneck as it involves complex combinatorial problem, which makes joint design of contention resources for users and AUD at the receiver a crucial but a challenging problem. To this end, we propose autoencoder (AE)-based joint optimization of both preamble generation networks (PGNs) in the encoder side and data-aided AUD in the decoder side. The core architecture of the proposed AE is a novel user activity extraction network (UAEN) in the decoder that extracts a priori user activity information from the SCMA codeword data for the data-aided AUD. An end-to-end training of the proposed AE enables joint optimization of the contention resources, i.e., preamble sequences, each associated with one of the codebooks, and extraction of user activity information from both preamble and SCMA-based data transmission. Furthermore, we propose a self-supervised pre-training scheme for the UAEN prior to the end-to-end training, to ensure the convergence of the UAEN which lies deep inside the AE network. Simulation results demonstrated that the proposed AUD scheme achieved 3 to 5dB gain at a target activity detection error rate of $\bf{{10}^{-3}}$ compared to the state-of-the-art DL-based AUD schemes.

eess.SY↗

arXiv2021

Deep Learning-based Codebook Design for Code-domain Non-Orthogonal Multiple Access Approaching Single-User Bit Error Rate Performance

A general form of codebook design for code-domain non-orthogonal multiple access (CD-NOMA) can be considered equivalent to an autoencoder (AE)-based constellation design for multi-user multidimensional modulation (MU-MDM). Due to a constrained design space for optimal constellation, e.g., fixed resource mapping and equal power allocation to all codebooks, however, existing AE architectures produce constellations with suboptimal bit-error-rate (BER) performance. Accordingly, we propose a new architecture for MU-MDM AE and underlying training methodology for joint optimization of resource mapping and a constellation design with bit-to-symbol mapping, aiming at approaching the BER performance of a single-user MDM (SU-MDM) AE model with the same spectral efficiency. The core design of the proposed AE architecture is dense resource mapping combined with the novel power allocation layer that normalizes the sum of user codebook power across the entire resources. This globalizes the domain of the constellation design by enabling flexible resource mapping and power allocation. Furthermore, it allows the AE-based training to approach a global optimal MU-MDM constellations for CD-NOMA. Extensive BER simulation results demonstrate that the proposed design outperforms the existing CD-NOMA designs while approaching the single-user BER performance achieved by the equivalent SU-MDM AE within 0.3 dB over the additive white Gaussian noise channel.

cs.IT↗

arXiv2021

Multi-sequence Spreading Random Access (MSRA) for Compressive Sensing-based Grant-free Communication

The performance of grant-free random access (GF-RA) is limited by the number of accessible random access resources (RRs) due to the absence of collision resolution. Compressive sensing (CS)-based RA schemes scale up the RRs at the expense of increased non-orthogonality among transmitted signals. This paper presents the design of multi-sequence spreading random access (MSRA) which employs multiple spreading sequences to spread the different symbols of a user as opposed to the conventional schemes in which a user employs the same spreading sequence for each symbol. We show that MSRA provides code diversity, enabling the multi-user detection (MUD) to be modeled into a well-conditioned multiple measurement vector (MMV) CS problem. The code diversity is quantified by the decrease in the average Babel mutual coherence among the spreading sequences. Moreover, we present a two-stage active user detection (AUD) scheme for both wideband and narrowband implementation. Our theoretical analysis shows that with MSRA activity misdetection falls exponentially while the size of GF-RA frame is increased. Finally, the simulation results show that about 82% increase in utilization of RRs, i.e., more active users, is supported by MSRA than the conventional schemes while achieving the RA failure rate lower bound set by random access collision.

cs.IT↗

arXiv2020

A Closed-form Transceiver Design for Interference Alignment and Cancellation (IAC) in MIMO Interference Channel

For a K-user interference channel, the degree of freedom (DoF) which can be achieved through interference alignment (IA) is constrained to signal space dimension governed by the number of Tx/Rx antennas. To overcome this problem, IA can be combined with interference cancellation (IC), involving a new receiver architecture associated with signaling over backhaul links among the different users, as another interference mitigation scheme which is referred to as interference alignment and cancellation (IAC). In our earlier work, by proposing an IAC graph, we have derived the necessary and sufficient conditions for the existence of closed-form solutions for IAC subject to the given DoF requirement for individual user. Furthermore, we have also shown that it can achieve the theoretically maximum possible DoF, which is 2M for MIMO system with M Tx/Rx antennas. Following our previous works on IAC, we aim to investigate the design criteria to obtain such closed-form transceivers when they exist. We first develop a general closed-form IAC transceiver design for any given DoF requirement of individual user and then, we specify how the optimal IAC transceiver can be designed to achieve the theoretically maximum DoF of 2M, beating the performance of conventional IA with much less computational complexity.

eess.SP↗

arXiv2019

Achievable Degrees of Freedom for Closed-form Solution to Interference Alignment and Cancellation in Gaussian Interference Multiple Access Channel

A combined technique of interference alignment (IA) and interference cancellation (IC), known as interference alignment and cancellation (IAC) scheme, has been proposed to improve the total achievable degrees of freedom (DoFs) over IA. Since it is NP-hard to solve the transceiver under a given tuple of DoFs or to maximize the total achievable DoFs in the general system configuration by IA (or IAC), the optimal transceiver cannot be obtained in polynomial time. Meanwhile, it has been known that a closed-form yet suboptimal transceiver can be designed for IAC by employing a symbol-to-symbol (STS) alignment structure. As its performance has not been known yet, we aim to derive the total DoFs that can be achieved by such suboptimal but closed-form IAC transceivers for Gaussian interference multiple access channels with K receivers and J users (transmitters), each with M antennas. Our analysis shows that the closed-form IAC transceivers under consideration can achieve a maximum total achievable DoFs of 2M, which turns out to be larger than those achieved in classical IA, e.g., 2MK/(K+1) DoFs by a specific configuration where each link has the same target DoFs. Moreover, considering the NP-hardness of deriving the maximum total achievable DoFs with the optimal IAC transceiver, its upper bound has been derived for comparison with the results of our closed-form IAC transceiver. Numerical results illustrate that its performance can be guaranteed within 20% of the upper bound when the number of multiple access channels are relatively small, e.g., K <=4.

eess.SP↗