Future sixth-generation (6G) wireless networks are envisioned as a backbone for mission-critical applications, including autonomous transportation, smart grids, space communications, and tactical defense systems. Ensuring continuous service delivery under adversarial conditions is a fundamental goal of Zero Trust Architectures (ZTA) for Future G. This paper presents an integrated framework that combines stochastic cyberattack modeling with cost-constrained survivability optimization to enhance resilience in 6G mission-critical networks. We extend the methodology to 6G radio access networks (RAN) and core infrastructures by building upon validated survivability models for cyber-physical systems (CPSs) and stochastic attack rate prediction. We also incorporate redundancy allocation strategies and series-parallel reliability modeling. Cyber-attacks are characterized by using stochastic processes such as Poisson and negative binomial distributions to predict attack rates and dynamically update trust scores within ZTA policy engines. The proposed approach is supported by comparative analysis, including numerical insights from prior validated studies and literature benchmarks. The framework aligns with 3GPP security enhancements for 6G and offers a pathway toward unbreakable resilience in Zero Trust-enabled mission-critical networks.
Radio astronomy (RA) studies the universe by detecting extremely weak radio signals using highly sensitive antennas. The increasing deployment of 5G base stations (BSs) in neighboring frequency bands creates a risk of out-of-band (OOB) interference, threatening to disrupt RA. To address this risk, one would have to impose a limit on 5G deployment, which could in turn adversely affect its area coverage. In this paper, we present Varuna—an algorithm designed to maximize 5G network coverage while ensuring the OOB interference threshold at passive RA site is satisfied. Varuna operates by selectively adjusting the transmit power of 5G BSs based on a penalty-driven metric that balances coverage loss against interference reduction. Through simulation experiments on a real-world Very Large Baseline Antenna (VLBA) site and using the Irregular Terrain Model (ITM) propagation model, we demonstrate that Varuna achieves nearly maximum spatial coverage while offering OOB interference protection for RA.
Satellites serve as a foundational element of modern society, supporting a wide range of critical functions across communication, navigation, environmental monitoring, and national security. The growing accessibility of low-cost software-defined radio (SDR) systems has heightened the vulnerability of legacy satellite systems, which often lack robust cryptographic protections, to spoofing and replay attacks. In this paper, we propose a radio fingerprinting system centered on an autoencoder-based deep learning model enhanced with an attention mechanism to capture structured and/or temporal patterns in satellite data and improve attack detection accuracy. The model was trained and evaluated on a dataset of 1.7 million satellite downlink messages, and its performance was benchmarked against a state-of-the-art satellite radio fingerprinting system. Experimental results show that our approach outperforms the state-of-the-art fingerprinting system by 13% in ROC Area Under Curve (AUC) and 13% in Equal Error Rate (EER).
In this paper we develop a novel disruption-resilient approach for real-time, high-resolution sensor data delivery over multiple wireless channels for military autonomous systems such as drones, autonomous vehicles and robots. We design two innovative neural multiple description codecs (neural MDCs) which compress and encode images into multiple independently decodable and mutually refineable streams. Our approach not only achieves high compression efficiency, but also enables the effective use of multiple diverse radio channels for real-time delivery of high-resolution sensor data while ensuring disruption resiliency. Using benchmark image/video sensor datasets as well as real-world 5G traces, we evaluate and demonstrate the efficacy of both neural MDC codecs for high-resolution sensor data streaming over multiple radio channels under various jamming scenarios.
Allowing third-party applications on Radio Access Network (RAN) Intelligent Controllers (RICs) within the Open-RAN (O-RAN) framework introduces conflicting interactions that are often difficult to detect in advance. These conflicts, can lead to performance degradation and instability in O-RAN if not identified efficiently. Existing conflict detection and mitigation solutions in the literature assume that the conflicts are known beforehand, which is not always accurate due to the complex and often hidden relationships between control parameters and Key Performance Indicators (KPIs).In this paper, we propose a novel Recurrent Neural Network (RNN) to detect both known and unknown conflicts in O-RAN xApps as specified in the O-RAN standards. We model the xApps, control parameters, and KPIs with nodes and edges to create graph structures and use the hidden non-Euclidean geometric properties of the Riemannian manifold to train the RNN model. The performance of this proposed model is validated using evaluation metrics and compared with benchmarks. Results demonstrate that the proposed RNN model, leveraging Riemannian geometric properties, can achieve 100% of the F1-score provided by an optimal solution in just 20 iterations.
The rise of smart factories is transforming manufacturing and service delivery using robots and unmanned ground vehicles (UGVs) for advanced tasks like autonomous navigation. While over-the-top (OTT) solutions reduce local compute and energy demands by offloading machine learning (ML) tasks to remote servers, they lack task-aware optimization. This limitation is critical for complex applications like autonomous navigation, where performance depends on dynamic device and network conditions. To bridge this gap, we propose Context-aware Adaptive inference for Machine-Task Execution in Complex environments (CAMTEC). It incorporates performance metrics from compute, energy, network, and ML to generate the context of machine task performance (e.g., UGV navigation). CAMTEC utilizes a deep reinforcement learning (DRL) agent to learn from this context information to determine the optimal inferencing mode (local, remote, or split) and vehicle speed to ensure optimal performance of an off-the-shelf object detection-based navigation algorithm. Extensive real-world testbed evaluations demonstrate CAMTEC’s efficacy in achieving collision-free autonomous navigation. CAMTEC adheres to a computationally optimal allocation strategy throughout the vehicle’s trajectory, ensuring that the actual-to-target speed variation remains below 20% even under battery-constrained conditions. Furthermore, even when network bandwidth is severely limited, CAMTEC ensures that the UGV converges to the optimal path while keeping the actual-to-target speed variation below 30%.
The increasing use of Artificial Intelligence (AI) and Big Data in defense demands high-throughput, real-time data processing capabilities, making High-Performance Computing (HPC) systems essential. To support the resulting computational demands, dedicated HPC systems have been deployed in air-gapped environments for processing surveillance data from drones, satellites, cameras, etc. However, access zone nodes-despite physical isolation-pose a critical security risk, enabling potential data exfiltration through compromised or impersonated systems. The emergence of io uring, a high-performance asynchronous I/O framework in Linux, has significantly improved data throughput, achieving rates up to 100 Gbps. Yet, it disrupts traditional observability tools like auditd, which depend on system call interception, thereby impeding flow-level logging and accountability. To address this, we propose DART - a scalable, low-latency flow-record logging solution compatible with both io uring and traditional I/O. By attaching lightweight eBPF programs to the Linux Traffic Control (TC) layer, DART enables reliable, high-performance monitoring in environments where conventional audit mechanisms fall short.
Widespread wireless communication demands robust security management to prevent adversaries from exploiting ubiquitous radio signals. Traditionally, such security is enforced through mitigation strategies implemented at higher layers using symmetric and/or asymmetric cryptography. However, this introduces increasing complexity, particularly due to the key derivation and exchange mechanisms required. This challenge is further exacerbated in private networks, which often necessitate configuration and deployment at the tactical or ad-hoc level in hostile environments, where physical interception or cryptanalysis by adversaries is a significant risk.To address this, we propose a solution that performs key derivation by exploiting 5G waveforms and integrated sensing and communication (ISAC). This approach enables the extraction of highly correlated data between legitimate parties by leveraging the reciprocity of radio channels, while preventing eavesdroppers from replicating the key derivation process.In this demonstration, we show that incorporating physical-layer information introduces geographical awareness between communicating parties, without compromising the entropy and keys generation rate. Furthermore, we demonstrate that our method effectively prevents eavesdroppers from obtaining the information needed to compromise the derived keys. Finally, we implement the system using software-defined radio within the 5G FR1 frequency bands, achieving a 100% key agreement rate with high entropy across a range of environments.
An iterative maximum-a-posteriori (MAP) single carrier (SC) detector is presented. We take note that to develop a low-complexity detector one should first obtain a list of candidate samples of the transmitted data symbols that closely match with the received signal. We thus develop a stochastic list generator for this purpose. This leads to a low-complexity, highly parallel detector with excellent performance. We find that the proposed detector has superior performance over the celebrated MMSE detector with soft interference cancellation. We also show that while the latter detector may fail to converge as the data symbols in a quadrature amplitude modulation (QAM) constellation grows, the proposed detector remains stable and leads to a good performance.
The Aerial Experimentation and Research Platform for Advanced Wireless (AERPAW) is an outdoor testbed hosted at NC State University, providing remote experimenters access to programmable radios and programmable vehicles. A key aspect of AERPAW is its real-time digital twin used for experiment development. This demonstration will showcase the capabilities of AERPAW’s digital twin by presenting a search and rescue scenario, involving a transmitter on the ground that needs to be localized using a receiver installed at an unmanned aerial vehicle (UAV) in the air based on the received signal strength at the UAV. The demonstration will be fully interactive, starting with "hiding" the transmitter at a random location in the field, and inviting the participants to "drive" the UAV (using a keyboard) and estimate the location of the hidden transmitter based on periodic received signal strength measurements.
Paper is focused on methods for quantitatively assessing the impact of cyber operations on the outcome of kinetic operations. We propose using computational intelligence methods, especially optimization methods for allocations cyber modules to modules of command and control. Optimization model was formulated, using concept of maturity level of command and control and cyber modules. Some examples illustrated potential influence of cyber operations on outcome of armed combat. It is anticipated that the proposed model will be used in the armed forces development planning system.
Medium power broadband millimeter wave (mmWave) power amplifiers (PAs) designed in an advanced 40-nm Gallium-Nitride (GaN) process from Hughes Research Laboratories (HRL) and a 22-nm fully depleted silicon-on-insulator (FDSOI) CMOS process from Global Foundries (GF) are characterized using 5G modulated signals in our lab. Our GaN PAs have reached continuous wave (CW) saturated output power (P-OUT,P- sat) of similar to 24 dBm, whereas our CMOS PAs reached P-OUT,P- sat similar to 14 dBm, both covering large portions of the FR2 band (e.g., 24.25-52.6 GHz) and reaching down into the upper FR3 band (e.g., 7.125-24.25 GHz). Post-layout parasitics extracted (PEX) electromagnetic (EM) simulations are used to design the PAs, and they are tested with 5G modulated waveforms using equipment in our labs, consisting of a mm-Wave probe station and two different vector signal transceivers (VSTs) from Emerson (i.e., formerly National Instruments (NI)). The VSTs used to characterize the PAs are: (1) our original VST, "VST1" (i.e., PXIe-5831), and (2) a latest VST, "VST3" (i.e., PXIe-5842 1C), that has greater maximum operating frequency and higher instantaneous signal bandwidth (IBW), and the VST3 can be used for testing in the FR4 band (e. g., 71-114.25 GHz) and at sub-THz frequencies up to similar to 300 GHz with up/downconverters. The performance of the PAs will be characterized using these VSTs, and demonstrated during the conference using the first VST3 deployed worldwide.
This paper introduces a hybrid middleware system for Unmanned Arial Vehicles (UAV) that can seamlessly alternate between Delay/Disruption Tolerant Networking (DTN) and Named Data Networking (NDN) without any changes to the application protocols. We first describe the middleware gateway controller system design, then, discuss the implementation details of the prototype built as containerized docker micro-services, and finally, show preliminary evaluations with practical UAV application use cases highlighting the efficacy of mixing the two protocols for time-sensitive data dissemination in highly dynamic heterogeneous network environments. The preliminary results highlight improved interoperability and resiliency in the hybrid protocol approach, laying the groundwork for future research in information sharing in highly contested environments.
As the demand for spectrum increases, significant research has been devoted to the development of interference-resistant waveforms. In this paper, we propose modified filterbank multicarrier spread spectrum (MFBMC-SS), a modification to the existing filterbank multicarrier spread spectrum (FBMC-SS) scheme that increases spectral and temporal diversity to be able to resolve and remove interference while operating at a low signal-to-noise ratio (SNR). Optimal interference rejection techniques for MFBMC-SS are derived. Robust order statistics and Gaussian mixture models are considered and simulated for interference parameter detection to apply to the derived rejection techniques. These simulations give a characterization for MFBMC-SS under different interference environments and with different interference rejection techniques.
Civilian technologies are developed based on specific use cases to address business requirements of underlying verticals or general enhancement of capabilities from a corresponding previous generation of technologies. This is also evident for wireless communications technologies. The 5th generation (5G) and upcoming 6th generation (6G) of wireless communications standards developed by 3rd Generation Partnership Project (3GPP), offer significant performance improvements from previous generations. In recent years, 5G and beyond technologies have garnered significant attention from the military communities to address certain capability requirements.Military technologies have traditionally developed independently from the civilian domain, due to the divergence of operational scenarios that mandate the need of stringent security and resilience overlay measures for seamless functionality in challenging environments. Commercially standardized technologies, like 5G, are generally not well-suited for such operational conditions. However, they can inherently offer wider interoperability and lower dependence on vendor-specific equipment. Additionally, improved quality of service (QoS) particularly with respect to throughput and wide spectral range are attractive features for consideration in military domain. Therefore, in order to use such commercial technologies for military operations, they should be made sufficiently robust to handle the operational requirements.In this paper, we focus on the 5G Radio Access Network (RAN) and provide an novel incremental approach to ruggedize commercial network nodes at the RAN-level. This is followed by an analysis performed by Airbus inline with this approach. A preliminary version of this approach was initially presented at the 5G/6G improving the NATO Digital Backbone Track of TIDE Sprint-43, held during 17-21 February 2025 in Helsinki.
This paper explores a new addition to the MAC protocol of a LEO network that employs cross-layer networking between the MAC and Physical Layers utilizing spatial processing of the phased array antenna for the uplinks to significantly improve frequency reuse and network performance, in particular capacity and revenue by as much as 25-times.
Network intrusion detection systems (NIDS) are essential in defending online assets from bad actors. Machine learning (ML) has proven to be an effective tool due to its ability to accurately detect complex patterns in network data. Maintaining high model performance, however, is a constant challenge, as malicious actors are non-static and the distribution, frequency and variety of attacks a network experiences constantly changes. Continual Learning (CL) has emerged as a new paradigm for ML-based NIDS to constantly adapt to changing threats and network environments, but faces many key issues, most of all the catastrophic forgetting problem. This paper provides an empirical analysis of different training strategies of CL and model-agnostic metal learning (MAML) to mitigate the effects of catastrophic forgetting within the network security domain. We show that regularization strategies prove ineffective in network security to prevent catastrophic forgetting, and that replay is the most effective way to ameliorate catastrophic forgetting. We also demonstrate that model repair with MAML, while effective for select classes, causes dramatic performance degradation to other classes. This work should prove useful in future research into the efficacy of traditional continual learning strategies in ML-NIDS, and their potential applications.
As the 5G New Radio (NR) standard is primarily designed to meet the needs of commercial cellular communications, it is not inherently robust against malicious interference sources such as jammers. To support military-grade deployments in diverse tactical environments, there is a need for a jamming-resilient physical uplink control channel (PUCCH) receiver. PUCCH plays a vital role in transmitting various control messages, collectively referred to as uplink control information (UCI), from user equipment (UE) to the base station (BS). For UCI payloads between 3 and 11 bits, small block coding (SBC) is employed. Due to their short lengths, these codes do not include a cyclic redundancy check (CRC), complicating error detection and the assurance of message reliability, especially in the presence of interference. To address this challenge, we propose a jamming-resilient PUCCH receiver that leverages a robust reliability assessment mechanism based on a threshold criterion, which bounds the probability of decoding errors. Additionally, we introduce a channel estimation error-aware equalizer tailored for DFT-s-OFDM waveforms, enabling more accurate computation of log-likelihood ratios (LLRs). Simulation results demonstrate that the proposed receiver can effectively evaluate decoding reliability without relying on CRC, even under jamming conditions, thereby enhancing the robustness of 5G control signaling in contested environments.
Decoy nodes such as honeypots defend military cyber systems from malicious attacks by detecting adversaries, gaining information on novel exploits, and introducing misinformation to adversaries. Virtual honeypots and honeynets enable cyber systems to rapidly create, reconfigure, and take down decoy nodes in order to maximize the effectiveness of the deception. This paper presents a game-theoretic framework for dynamic decoy placement. In contrast to existing game-theoretic methodologies that focus on static decoy placement, our framework formulates dynamic strategies that modify the decoy locations in response to suspicious network activity. In order to solve the game, we propose a reinforcement learning approach for computing an optimal decoy placement policy. We mitigate the computational complexity by exploiting submodular structure of the value function of the game, which enables provable optimality bounds using a sequential greedy approximation at each time step. We conduct a numerical study that demonstrates our approach is tractable for a 100-node grid network and significantly improves win probability while reducing resource cost compared to a static decoy placement algorithm.
Large Language Models (LLMs) have achieved remarkable success across a wide range of domains, yet their application to network security remains underexplored. Prior studies of network intrusion detection system (NIDS) using LLMs either rely on computationally expensive fine-tuning or use LLMs to generate explanations for traditional NIDS outputs. However, to date, no studies have conducted a comprehensive evaluation of LLM effectiveness for NIDS using diverse prompting strategies and extended text inputs. In this paper, we present NetPrompt, a prompt-only NIDS evaluation pipeline that operates without any fine-tuning. We evaluate its efficacy using one proprietary (Gemini-2.0-Flash) and two open-source LLMs (Qwen2.5-7B-Instruct, Llama-3.1-8B-Instruct), employing prompting techniques such as Zero-Shot, Few-Shot, and Chain-of-Thought across two benchmark datasets: CICIDS2017 and CICDDoS2019. We also propose iCentroid, a novel class-centric example selection algorithm that identifies representative examples to enhance contextual understanding within the Few-Shot prompting. Our experimental results show that the Gemini model consistently outperforms both open-source LLMs, achieving higher F1-scores and lower error rates. Moreover, Gemini surpasses the baseline multi-layer perceptron (MLP)-based NIDS model, making LLMs attractive for cybersecurity tasks without the need for fine-tuning.
