Nakjung Choi
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Nakjung Choi has a total of 63 co-authors. Here are the top co-authors by number of shared publications:
- Ahan Kak
Nokia Bell Labs, Murray Hill, NJ, USA
9 shared publications - Huu-Trung Thieu
Nokia Bell Labs, Murray Hill, NJ, USA
5 shared publications - Van-Quan Pham
Nokia Bell Labs, Murray Hill, USA
5 shared publications - Bo Ji
Virginia Tech, Blacksburg, Virginia, USA
4 shared publications Yuru Zhang
University of Nebraska-Lincoln Nokia Bell Labs
4 shared publicationsAhan Kak
University of Nebraska-Lincoln Nokia Bell Labs
3 shared publications
Latest Publications
- Open Access
- 22 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 471-480
Model-based Zero-shot Conflict Detection Framework via Gradients in O-RAN
- 22 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 471-480
Open AccessModel-based Zero-shot Conflict Detection Framework via Gradients in O-RAN
- Open Access
- 26 October 2026
- Published by Association for Computing Machinery (ACM)
- p. 1852-1854
Demo: FrameSight: Semantic-Aware RAN for Tile-Based Real-Time Video Streaming
- 26 October 2026
- Published by Association for Computing Machinery (ACM)
- p. 1852-1854
Open AccessDemo: FrameSight: Semantic-Aware RAN for Tile-Based Real-Time Video Streaming
- 3 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 23, 3934-3947
Flat UP: A Converged RAN-Core Architecture for the 6G User Plane
The ongoing industry shift toward Radio Access Network (RAN) disaggregation, virtualization, and cloudification has disrupted the conventional hierarchical design of cellular networks and opened the door to greater convergence between the RAN and core domains. Despite this progress, implementing such converged architectures in practice presents numerous challenges, including those related to protocol and architectural design, quality-of-service (QoS) assurance, control plane configuration, and support for emerging 6G-specific use cases. To address these challenges, this article presents the flat User Plane (UP) architecture, a novel framework for RAN-core convergence centered around a new 6G-native component: the Access User Plane Function (AUPF). The article outlines the key innovations in the newly proposed flat user plane architecture, including protocol- and feature-level design evolutions as well as enhancements to QoS provisioning. It then explores various counterpart Control Plane (CP) architectures, analyzing the impact of the new design on different 3GPP CP procedures. A concrete, system-level prototype implementation of the AUPF is developed, accompanied by a comprehensive over-the-air evaluation to assess both fundamental network performance metrics and user plane Quality of Experience (QoE). Additionally, multiple deployment models are examined to quantify the CP signaling overhead associated with different architectural options. The results demonstrate that the proposed flat UP architecture not only improves throughput, latency performance and QoE but also reduces overall compute resource utilization when compared to the conventional hierarchical 5G user plane. The CP evaluation further provides practical insights and guidelines for real-world deployment scenarios.
- 3 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 23, 3934-3947
Flat UP: A Converged RAN-Core Architecture for the 6G User Plane
The ongoing industry shift toward Radio Access Network (RAN) disaggregation, virtualization, and cloudification has disrupted the conventional hierarchical design of cellular networks and opened the door to greater convergence between the RAN and core domains. Despite this progress, implementing such converged architectures in practice presents numerous challenges, including those related to protocol and architectural design, quality-of-service (QoS) assurance, control plane configuration, and support for emerging 6G-specific use cases. To address these challenges, this article presents the flat User Plane (UP) architecture, a novel framework for RAN-core convergence centered around a new 6G-native component: the Access User Plane Function (AUPF). The article outlines the key innovations in the newly proposed flat user plane architecture, including protocol- and feature-level design evolutions as well as enhancements to QoS provisioning. It then explores various counterpart Control Plane (CP) architectures, analyzing the impact of the new design on different 3GPP CP procedures. A concrete, system-level prototype implementation of the AUPF is developed, accompanied by a comprehensive over-the-air evaluation to assess both fundamental network performance metrics and user plane Quality of Experience (QoE). Additionally, multiple deployment models are examined to quantify the CP signaling overhead associated with different architectural options. The results demonstrate that the proposed flat UP architecture not only improves throughput, latency performance and QoE but also reduces overall compute resource utilization when compared to the conventional hierarchical 5G user plane. The CP evaluation further provides practical insights and guidelines for real-world deployment scenarios.
- 25 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25 (8) , 12506-12520
LEVEL: Leveraging RAN-Core Convergence for the Private Wireless User Plane
With its wide applicability across a variety of enterprise verticals, the private wireless (PW) industry is growing at a rapid pace. However, existing hierarchical cellular network architectures are sub-optimal for most PW use cases due to their high computational overhead, lack of compatibility with on-premises enterprise applications, and complex management structure. To that end, this paper espouses the notion of RAN-Core Convergence for the PW user plane as an alternative to the traditional RAN-Core hierarchy through the introduction of LEVEL, a new converged approach to cellular network design. With a specific emphasis on the user plane, key highlights include a unified system architecture and protocol stack design, new primitives for interacting with the cellular control plane, and a programmable approach to QoS management. The paper also includes a high-performance systems-level prototype of the LEVEL User Plane (LEVEL-UP), deployed on an over-the-air experimental testbed. The resulting performance evaluation showcases that LEVEL-UP seamlessly handles traffic in excess of $\mathrm{\rm{20}~Gbps}$, while achieving a $60\%$ reduction in compute utilization and $30\%$ reduction in energy consumption when compared to state-of-the-art solutions. Furthermore, latency benchmarks reveal a delay reduction of at least $70\%$. Finally, in typical high mobility environments, LEVEL-UP helps cut bufferbloat and reduces delays from seconds to milliseconds, thereby serving as an ideal companion for the burgeoning PW industry.
- 25 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Mobile Computing
- Vol. 25 (8) , 12506-12520
LEVEL: Leveraging RAN-Core Convergence for the Private Wireless User Plane
With its wide applicability across a variety of enterprise verticals, the private wireless (PW) industry is growing at a rapid pace. However, existing hierarchical cellular network architectures are sub-optimal for most PW use cases due to their high computational overhead, lack of compatibility with on-premises enterprise applications, and complex management structure. To that end, this paper espouses the notion of RAN-Core Convergence for the PW user plane as an alternative to the traditional RAN-Core hierarchy through the introduction of LEVEL, a new converged approach to cellular network design. With a specific emphasis on the user plane, key highlights include a unified system architecture and protocol stack design, new primitives for interacting with the cellular control plane, and a programmable approach to QoS management. The paper also includes a high-performance systems-level prototype of the LEVEL User Plane (LEVEL-UP), deployed on an over-the-air experimental testbed. The resulting performance evaluation showcases that LEVEL-UP seamlessly handles traffic in excess of $\mathrm{\rm{20}~Gbps}$, while achieving a $60\%$ reduction in compute utilization and $30\%$ reduction in energy consumption when compared to state-of-the-art solutions. Furthermore, latency benchmarks reveal a delay reduction of at least $70\%$. Finally, in typical high mobility environments, LEVEL-UP helps cut bufferbloat and reduces delays from seconds to milliseconds, thereby serving as an ideal companion for the burgeoning PW industry.
- Open Access
inRAN: Interpretable Online Bayesian Learning for Network Automation in Open Radio Access Networks
Emerging AI/ML techniques have been showing great potential in automating network control in open radio access networks (Open RAN). However, existing approaches heavily rely on blackbox policies parameterized by deep neural networks, which inherently lack interpretability, explainability, and transparency, and create substantial obstacles in practical network deployment. In this paper, we propose inRAN, a novel interpretable online Bayesian learning framework for network automation in Open RAN. The core idea is to integrate interpretable surrogate models and safe optimization solvers to continually optimize control actions, while adapting to non-stationary dynamics in real-world networks. We achieve the inRAN framework with three key components: 1) an interpretable surrogate model via ensembling Kolmogorov-Arnold Networks (KANs); 2) safe optimization solvers via integrating genetic search and trust-region descent method; 3) an online dynamics tracker via continual model learning and adaptive threshold offset. We implement inRAN in an end-to-end O-RAN-compliant network testbed, and conduct extensive over-the-air experiments with the focused use case of network slicing. The results show that, inRAN substantially outperforms state-of-the-art works, by guaranteeing the chance-based constraint with a 92.67% assurance ratio with comparative resource usage throughout the online network control, under unforeseeable time-evolving network dynamics.
Open AccessinRAN: Interpretable Online Bayesian Learning for Network Automation in Open Radio Access Networks
Emerging AI/ML techniques have been showing great potential in automating network control in open radio access networks (Open RAN). However, existing approaches heavily rely on blackbox policies parameterized by deep neural networks, which inherently lack interpretability, explainability, and transparency, and create substantial obstacles in practical network deployment. In this paper, we propose inRAN, a novel interpretable online Bayesian learning framework for network automation in Open RAN. The core idea is to integrate interpretable surrogate models and safe optimization solvers to continually optimize control actions, while adapting to non-stationary dynamics in real-world networks. We achieve the inRAN framework with three key components: 1) an interpretable surrogate model via ensembling Kolmogorov-Arnold Networks (KANs); 2) safe optimization solvers via integrating genetic search and trust-region descent method; 3) an online dynamics tracker via continual model learning and adaptive threshold offset. We implement inRAN in an end-to-end O-RAN-compliant network testbed, and conduct extensive over-the-air experiments with the focused use case of network slicing. The results show that, inRAN substantially outperforms state-of-the-art works, by guaranteeing the chance-based constraint with a 92.67% assurance ratio with comparative resource usage throughout the online network control, under unforeseeable time-evolving network dynamics.
- Open Access
oneTwin: Online Digital Network Twin via Neural Radio Radiance Field
Digital network twin is a promising technology that replicates real-world networks in real-time and assists with the design, operation, and management of next-generation networks. However, existing approaches (e.g., simulator-based and neural-based) cannot effectively realize the digital network twin, in terms of fidelity, synchronicity, and tractability. In this paper, we propose oneTwin, the first online digital twin system, for the prediction of physical layer metrics. We architect the oneTwin system with two primary components: an enhanced simulator and a neural radio radiance field (NRRF). On the one hand, we achieve the enhanced simulator by designing a material tuning algorithm that incrementally optimizes the building materials to minimize the twin-to-real gap. On the other hand, we achieve the NRRF by designing a neural learning algorithm that continually updates its DNNs based on both online and simulated data from the enhanced simulator. We implement oneTwin system using Sionna RT as the simulator and developing new DNNs as the NRRF, under a public cellular network. Extensive experimental results show that, compared to state-of-the-art solutions, oneTwin achieves real-time updating (0.98s), with 36.39% and 57.50% reductions of twin-to-real gap under in-distribution and out-of-distribution test datasets, respectively.
Open AccessoneTwin: Online Digital Network Twin via Neural Radio Radiance Field
Digital network twin is a promising technology that replicates real-world networks in real-time and assists with the design, operation, and management of next-generation networks. However, existing approaches (e.g., simulator-based and neural-based) cannot effectively realize the digital network twin, in terms of fidelity, synchronicity, and tractability. In this paper, we propose oneTwin, the first online digital twin system, for the prediction of physical layer metrics. We architect the oneTwin system with two primary components: an enhanced simulator and a neural radio radiance field (NRRF). On the one hand, we achieve the enhanced simulator by designing a material tuning algorithm that incrementally optimizes the building materials to minimize the twin-to-real gap. On the other hand, we achieve the NRRF by designing a neural learning algorithm that continually updates its DNNs based on both online and simulated data from the enhanced simulator. We implement oneTwin system using Sionna RT as the simulator and developing new DNNs as the NRRF, under a public cellular network. Extensive experimental results show that, compared to state-of-the-art solutions, oneTwin achieves real-time updating (0.98s), with 36.39% and 57.50% reductions of twin-to-real gap under in-distribution and out-of-distribution test datasets, respectively.
- Open Access
- 3 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 33-40
RAN-SURE: Toward Sustainable RAN Orchestration
- 3 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 33-40
Open AccessRAN-SURE: Toward Sustainable RAN Orchestration
- Open Access
- 23 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 131-140
Odin: Effective End-to-End SLA Decomposition for 5G/6G Network Slicing via Online Learning
- 23 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 131-140
Open AccessOdin: Effective End-to-End SLA Decomposition for 5G/6G Network Slicing via Online Learning
- 25 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 22 (6) , 5803-5821
HexRAN: A Programmable Approach to Open RAN Base Station System Design
In recent years, the radio access network (RAN) domain has seen significant changes with increased virtualization and softwarization, driven by the Open RAN (O-RAN) movement. However, the fundamental building block of the cellular network, i.e., the base station, remains unchanged and ill-equipped to handle this architectural evolution. In particular, there exists a general lack of programmability and composability along with a protocol stack that grapples with the intricacies of the 3GPP and O-RAN specifications. Recognizing the need for an “O-RAN-native” approach to base station design, this paper introduces HexRAN– a novel base station architecture characterized by key features relating to RAN disaggregation and composability, 3GPP and O-RAN protocol integration and programmability, robust controller interactions, and customizable RAN slicing. Furthermore, the paper also includes a concrete systems-level prototype and comprehensive experimental evaluation of HexRAN on an over-the-air testbed. The results demonstrate that HexRAN uses only 8% more computing resources compared to the baseline, while managing twice the user plane traffic, delivering control plane processing latency of under 120μs, and achieving 100% processing reliability. This underscores the scalability and performance advantages of the proposed architecture.
- 25 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Network and Service Management
- Vol. 22 (6) , 5803-5821
HexRAN: A Programmable Approach to Open RAN Base Station System Design
In recent years, the radio access network (RAN) domain has seen significant changes with increased virtualization and softwarization, driven by the Open RAN (O-RAN) movement. However, the fundamental building block of the cellular network, i.e., the base station, remains unchanged and ill-equipped to handle this architectural evolution. In particular, there exists a general lack of programmability and composability along with a protocol stack that grapples with the intricacies of the 3GPP and O-RAN specifications. Recognizing the need for an “O-RAN-native” approach to base station design, this paper introduces HexRAN– a novel base station architecture characterized by key features relating to RAN disaggregation and composability, 3GPP and O-RAN protocol integration and programmability, robust controller interactions, and customizable RAN slicing. Furthermore, the paper also includes a concrete systems-level prototype and comprehensive experimental evaluation of HexRAN on an over-the-air testbed. The results demonstrate that HexRAN uses only 8% more computing resources compared to the baseline, while managing twice the user plane traffic, delivering control plane processing latency of under 120μs, and achieving 100% processing reliability. This underscores the scalability and performance advantages of the proposed architecture.
- Open Access
S2M3: Split-and-Share Multi-Modal Models for Distributed Multi-Task Inference on the Edge
With the advancement of Artificial Intelligence (AI) towards multiple modalities (language, vision, speech, etc.), multi-modal models have increasingly been used across various applications (e.g., visual question answering or image generation/captioning). Despite the success of AI as a service for multi-modal applications, it relies heavily on clouds, which are constrained by bandwidth, latency, privacy concerns, and unavailability under network or server failures. While on-device AI becomes popular, supporting multiple tasks on edge devices imposes significant resource challenges. To address this, we introduce S2M3, a split-and-share multi-modal architecture for multi-task inference on edge devices. Inspired by the general-purpose nature of multi-modal models, which are composed of multiple modules (encoder, decoder, classifier, etc.), we propose to split multi-modal models at functional-level modules; and then share common modules to reuse them across tasks, thereby reducing resource usage. To address cross-model dependency arising from module sharing, we propose a greedy module-level placement with per-request parallel routing by prioritizing compute-intensive modules. Through experiments on a testbed consisting of 14 multi-modal models across 5 tasks and 10 benchmarks, we demonstrate that S2M3 can reduce memory usage by up to 50% and 62% in single-task and multi-task settings, respectively, without sacrificing accuracy. Furthermore, S2M3 achieves optimal placement in 89 out of 95 instances (93.7%) while reducing inference latency by up to 56.9% on resource-constrained devices, compared to cloud AI.
Open AccessS2M3: Split-and-Share Multi-Modal Models for Distributed Multi-Task Inference on the Edge
With the advancement of Artificial Intelligence (AI) towards multiple modalities (language, vision, speech, etc.), multi-modal models have increasingly been used across various applications (e.g., visual question answering or image generation/captioning). Despite the success of AI as a service for multi-modal applications, it relies heavily on clouds, which are constrained by bandwidth, latency, privacy concerns, and unavailability under network or server failures. While on-device AI becomes popular, supporting multiple tasks on edge devices imposes significant resource challenges. To address this, we introduce S2M3, a split-and-share multi-modal architecture for multi-task inference on edge devices. Inspired by the general-purpose nature of multi-modal models, which are composed of multiple modules (encoder, decoder, classifier, etc.), we propose to split multi-modal models at functional-level modules; and then share common modules to reuse them across tasks, thereby reducing resource usage. To address cross-model dependency arising from module sharing, we propose a greedy module-level placement with per-request parallel routing by prioritizing compute-intensive modules. Through experiments on a testbed consisting of 14 multi-modal models across 5 tasks and 10 benchmarks, we demonstrate that S2M3 can reduce memory usage by up to 50% and 62% in single-task and multi-task settings, respectively, without sacrificing accuracy. Furthermore, S2M3 achieves optimal placement in 89 out of 95 instances (93.7%) while reducing inference latency by up to 56.9% on resource-constrained devices, compared to cloud AI.