Wideband terahertz (THz) massive multiple-input multiple-output (MIMO) promises extremely high data rates, but large fractional bandwidth and large arrays cause beam squint/beam split, leading to severe spectral-efficiency loss with conventional phase-shifter-only hybrid precoding. Delay–phase (true-time-delay plus phase-shifter) precoding mitigates beam split, yet existing designs often provision the delay-network granularity for full-angle coverage, resulting in unnecessary hardware and power consumption in practical sectorized deployments. This paper develops a sector-aware delay–phase precoding (SA-DPP) framework for wideband THz massive MIMO with reconfigurable intelligent surface (RIS) assistance. We first derive a sector-conditioned sufficient condition that links the required number of true-time-delay (TTD) elements per RF chain to the supported angular sector and system bandwidth, and we select the smallest feasible delay-network granularity. Given this hardware dimensioning, we construct the wideband analog beamformer in closed form and design the per-subcarrier digital precoder; for RIS-assisted links, we also configure the RIS phases using a geometry-based initialization followed by a low-complexity alternating optimization. Simulation results for RIS-assisted links show that SA-DPP achieves near-optimal wideband rate close to a TTD-based baseline while using fewer delay elements and improving energy efficiency in sector-limited channels. To address the propagation-regime concern for electrically large THz arrays, we also evaluate the original far-field-based DPP/SA-DPP designs on a spherical-wave channel generator and show that, although the representative short-range link lies in the radiating near field, the resulting rate loss remains modest. We further quantify the impact of RIS size under an aperture-scaling model and show that larger RISs improve the wideband data rate with diminishing gains due to frequency-flat RIS phase constraints. We also evaluate quantized RIS phase shifts in wideband operation and demonstrate that a small number of phase bits (e.g., 3–4 bits) suffices to approach continuous-phase performance.

Beyond the fifth generation (B5G) paradigm, including the Internet of Things (IoT), is projected to establish unprecedented scalability and real-time responsiveness. However, the large-scale deployment of such systems is prone to security attacks, which grow due to their decentralised architecture, minimal latency, and substantial data exchange. While traditional intrusion detection systems (IDS) struggle to effectively safeguard user privacy and system integrity in dynamic and distributed environments, large language models (LLMs) have recently unveiled novel domains in cybersecurity due to their proficiency in capturing intricate sequential patterns. This paper proposes a novel method for detecting intrusions specifically designed for IoT networks that utilize B5G technology. The system employs a Hierarchical Federated Learning (HFL) model in conjunction with lightweight LLMs, such as TinyLLaMA and DistilBERT, to ensure that model training remains private, scalable, and efficient across diverse devices. In particular, the architecture consists of three tiers: clients, edge aggregators, and a central server, which, as a result facilitates hierarchical model aggregation while preserving data locality. The framework integrates quantisation, knowledge distillation, and Low-Rank Adaptation (LoRA) to improve the efficiency of edge deployment. Simulation results reveal that the proposed approach outperforms the conventional federated learning baselines on the TON IoT dataset. In addition, the proposed framework effectively accommodates non-IID data distributions and reduces communication overhead by 28

This paper presents an analytical framework for the ergodic capacity (EC) of a downlink multi-user Rate-Splitting Multiple Access (RSMA) system over keyhole Nakagami- $m$ fading channels, motivated by wireless connectivity in subway tunnels and indoor corridors. In such environments, structural constrictions force all multi-path energy through a single dominant waveguide mode. This collapses the effective channel rank to unity and reduces the channel to the product of two independent Nakagami- $m$ random variables. For a multi-user network with $N$ single-antenna users, closedform expressions are derived for the EC of both common and private RSMA streams under perfect successive interference cancellation (pSIC) using Meijer- $G$ functions. For the imperfect SIC (ipSIC) case, a Gauss-Laguerre quadrature approximate expression is developed. Additionally, tight Jensen upper bounds are derived to provide compact design guidelines at moderateto-high signal-to-noise ratio (SNR). Under pSIC, asymptotic high- SNR analysis reveals that the EC for common and private streams is saturated at deterministic ceilings that depend on the power allocation coefficients. The ipSIC ceiling lies strictly below the pSIC ceiling and retains a residual dependence on the interference statistics. Monte Carlo simulations confirm all analytical results. As a comparison with Non-Orthogonal Multiple Access (NOMA), the findings show that RSMA achieves a consistent EC gain across low-to-medium SNR regimes, with more enhancement under ipSIC.

Cognitive radio networks (CRNs) enable unlicensed users to opportunistically exploit spectrum holes while protecting licensed primary users (PUs). In multi-user CRNs supporting bandwidth-intensive services, meeting high-rate demands often requires channel aggregation, where multiple idle licensed channels are used in parallel. However, under simultaneous uncertain PU returns and proactive jamming attacks, aggregating more channels is not always beneficial; while it increases the achievable aggregate rate, it also raises the risk of forced termination because successful delivery requires all selected channels to remain PU-idle and jammer-free throughout the entire control-plus-data interval. This creates a fundamental tradeoff between rate, reliability, and interruption risk. In this paper, we study the joint optimization of channel-set selection and packet-size adaptation under quality-of-service (QoS) and reliability constraints. We model the residual PU-idle time using an idle-age-dependent, distribution-agnostic survival function and develop an analogous residual jammer-free model for proactive channel-biased jamming. Based on these models, we derive a packet-success expression that explicitly accounts for control overhead and interruptions caused by both PU activity and jamming. The resulting decision problem is formulated as an infinite-horizon discounted Markov decision process (MDP) with state-dependent feasible actions under reliability, SNR-feasibility, and aggregate-rate constraints. To solve it online, we propose a jamming- and PU-aware deep reinforcement learning framework that combines action masking, a Deep Q-Network (DQN), convergence-guided exploration, and candidate-action reduction to efficiently handle the combinatorial action space. Simulation results show consistent gains over staged, fixed-packet, and fragmentation-based baselines under heterogeneous, non-memoryless PU and jamming dynamics.

Higher education is experiencing a significant shift towards Industry 5.0. The present episode is defined by an emphasis on sustainability and human-centered global networks. The modern supply chain has transformed into a multi-tiered ecosystem that is more intricate and unstable. Students rarely get the opportunity to apply their knowledge in supply chain management (SCM) courses. Immersive learning may be augmented by digital twins (DT); yet, Modern classroom systems sometimes face issues of latency and concurrency. This paper proposes a three-tier distributed architecture based on Edge, Fog, and Cloud layers. The architecture supports a scalable learning environment for SCM simulations. The approach is consistent with Experiential Learning Theory (ELT). It enables students to engage with the digital twin and witness the outcomes of their selections instantly. The system supports up to thirty concurrent users through distributed processing. This enables both Concrete Experience and Active Experimentation during classroom simulations. Simulation data are stored in the Cloud layer, which allows students to review and reflect on the outcomes of their decisions. The proposed approach positions digital twins as learning tools rather than only technical systems. It also supports the goals of Industry 5.0, where technology is designed around human learning and decision-making. A functional model to teach the SCM course in a virtual business environment using DT technology is proposed by this paper. Spesifically, we develop a design-oriented architectural framework for classroom-scale distributed SCM learning rather than as an empirically validated approach, and its main contribution lies in the explicit treatment of synchronization, concurrency, and layer-specific processing responsibilities.

Indoor autonomous drones rely on robust wireless connectivity to operate in GPS-denied environments; however, indoor radio-frequency (RF) propagation is strongly affected by multipath, blockage, and device instability. These effects become particularly critical for hovering unmanned aerial vehicles (UAVs), known as drones, where vibration and orientation variations can significantly alter the wireless channel. This paper presents a measurement-based characterization of large-scale indoor RF propagation, explicitly comparing stationary ground-to-ground links with hovering UAV-to-ground links under identical indoor conditions. Using a purpose-built ESP32-based measurement platform operating at 2.4 GHz, extensive RSSI and RSSI-referenced signal-to-noise ratio (SNR) proxy measurements are collected across a realistic office environment. Based on spatially averaged measurements, log-distance path-loss models are derived for both scenarios. Unlike prior indoor channel modelling studies that consider ground-to-ground communications, this work quantifies the effect of hovering-induced vibration and drift as well as antenna orientation on large-scale indoor propagation, showing that hovering UAV-to-ground links exhibit a substantially higher path-loss exponent (2.46) than stationary ground-to-ground links (1.50), along with increased shadowing variability (approximately 6.9 dB versus 6.1 dB). In some indoor locations, UAV hovering introduces RSSI and SNR proxy degradations of up to 20 dB relative to the stationary baseline. These findings demonstrate that conventional indoor propagation models can significantly underestimate attenuation and variability in indoor UAV deployments, highlighting the need to explicitly account for hovering-induced effects when designing and planning indoor UAV communication systems.

Decentralized vehicle-to-everything (V2X) communication enables direct device-to-device transmissions that can offload cellular infrastructure and reduce end-to-end latency. However, the benefits of full-duplex (FD) non-orthogonal multiple access (NOMA) are highly sensitive to residual self-interference (SI), fast channel fluctuations, and heterogeneous quality-of-service (QoS) requirements. In practice, fixed duplexing, fixed successive interference cancellation (SIC) ordering, and static power allocation can lead to unreliable decoding and inefficient operation across different SNR and SI regimes. This paper proposes an enhanced adaptive hybrid-duplex NOMA framework for decentralized V2X clusters that improves spectral efficiency while preserving QoS robustness under realistic fading and residual SI. The proposed framework integrates: 1) channel-disparity aware user scheduling and pairing to exploit near-far diversity; 2) adaptive SIC decoding-order selection per channel realization to mitigate decoding failures; 3) continuous time-splitting (partial-FD) control that smoothly trades FD multiplexing gains against residual SI instead of relying on a binary FD/HD switch; and 4) outage-aware robust power allocation that enforces minimum-rate constraints through deterministic safety margins. Extensive Monte-Carlo evaluations under Rayleigh (urban/crowded) and Rician (suburban/LoS) channels demonstrate that the proposed strategy consistently improves throughput across SNR and SI conditions while maintaining competitive Jain fairness (>0.85) and practical computational complexity of O(UlogU) for greedy pairing.

Global Navigation Satellite Systems (GNSS) are increasingly vulnerable to environmental disruptions, signal interference, and cyber threats in vehicular networks. To address these challenges, this paper introduces the GNSS Signal Resiliency Metric (GSRM), a novel framework that integrates signal availability with stochastic modeling to evaluate robustness under uncertainty. Leveraging Ornstein–Uhlenbeck (OU) process parameters estimated from real GNSS measurements in the UrbanLoco dataset, GSRM quantifies performance and recovery behavior across diverse environments. Results reveal strong agreement between OU modeling and experimental availability data, validating the framework's applicability. Distinct resiliency profiles emerge: open-sky conditions achieve rapid resiliency saturation (composite score 0.88), suburban environments show moderate gains, while urban canyons exhibit delayed recovery and lower resiliency (composite score 0.62) due to multipath and non-line-of-sight effects. A nonlinear relationship between availability and resiliency underscores the importance of continuous signal access. By integrating availability, accuracy, continuity, and integrity, GSRM provides a reproducible, data-driven, and scalable tool for assessing GNSS robustness, supporting adaptive receiver design, redundancy planning, and coexistence strategies in next-generation vehicular networks.

Cognitive radio allows secondary users to exploit idle licensed spectrum while protecting primary users (PUs), but transmissions fail if a PU returns or jamming occurs. Thus, packet size and channel assignment must be jointly optimized to balance throughput and interruption risk. Unlike prior work that assumes memoryless exponential PU activity and jammer-free intervals, in practice spectrum-occupancy measurements often deviate from the exponential distribution and may exhibit heavy-tailed behavior, making channel age informative. This paper develops a jamming-aware, age-aware contention-based optimization framework with success-probability and signal-to-noise ratio (SNR) constraints. It also derives exact success expressions and a closed-form packet-duration approximation. The main novelty is the coupled channel/packet decision under non-memoryless PU/jammer interruption, exact age-conditioned probability-of-success (PoS), and closed-form packet sizing with exact discrete projection. This provides a lightweight continuous packet-duration candidate, maps it to the practical discrete packet alphabet, and admits it only after exact PoS/SNR rechecking. Simulations show improved throughput over fixed-packet and memoryless methods under realistic PU and jamming distributions.

Millimeter-wave (mmWave) communications offer large bandwidth for high data rates, yet suffer from severe path loss and blockage, which makes highly directional beam alignment essential and training overhead a critical bottleneck. Reconfigurable intelligent surfaces (RISs), implemented as nearly passive programmable metasurfaces, can reshape the radio environment to create strong reflected paths and enhance mmWave coverage via controllable beamforming, but practical operation requires fast and reliable beam training over a large cascaded search space. To that end, coded beam training (CBT) was proposed to reduce training overhead by mapping hierarchical beam search to an error-control coding structure; however, existing RIS-CBT approaches mainly rely on hard decisions and fixed pilot allocation, which limits robustness under stringent overhead budgets and varying received signal-to-noise ratio (SNR) conditions. To deal with this challenge, this paper proposes a reliability-aware RIS beam training method that combines soft-information decoding with confidence-driven pilot reallocation. Specifically, we derive stage-wise soft metrics from the 2 & times; 2 pilot measurements and apply a soft maximum a posteriori (MAP) decoder to exploit measurement reliability rather than hard slicing. When decoding confidence is low, the remaining pilot budget is adaptively reallocated to repeat the most uncertain stages, improving overall decision reliability without increasing the nominal codebook depth. Simulation results for offgrid, RIS-assisted mmWave channels demonstrate that the proposed method achieves a higher success probability and improved achievable rate compared with binary hierarchical training and hard-decision CBT baselines across a wide range of SNRs and training overheads, and it approaches the oracle upper bound corresponding to perfect channel state information (CSI) with substantially reduced training overhead, while maintaining practical computational complexity

Indoor flying networks (IFNs) can provide flexible monitoring support in crowded indoor environments, but practical deployment must jointly account for user mobility, limited UAV energy, motion boundaries, and reliable UAV-to-access-point (AP) communication. This paper studies energy- and service-reliability-constrained three-dimensional (3D) UAV deployment in a hybrid LiFi-WiFi IFN. The optimization objective is to maximize the number of mobile users monitored by each UAV, while energy, boundary, speed, and AP-connectivity requirements are imposed as operational constraints. To solve the resulting sequential decision problem, we use a coverage-driven deep reinforcement learning (DRL) framework whose main state keeps the UAV position, residual energy, and current coverage. The paper also implements an aggregate density-aware state variant based on coarse cell-level occupancy, density centroid, and density-centroid displacement to quantify the benefit of explicit mobility-context information without requiring exact user localization. The UAV operates in a continuous physical space, yet the DQL controller selects actions from a discrete set of local 3D motion primitives. This discretization is motivated by the limited onboard computational resources of the indoor UAV and the need to make rapid local positioning decisions. Collision avoidance is provided structurally by non-overlapping sub-region assignment, while boundary and wireless-service feasibility are enforced through action screening and reward penalties. The proposed policy is evaluated against several practical schemes (random selection, centroid-based selection, greedy heuristics, particle swarm optimization, and genetic algorithms), as well as against an upper-bound benchmark based on finite-grid integer linear programming (ILP). The results show that the proposed DRL method substantially outperforms practical baselines in terms of the average number of monitored users, while still achieving 93.7% of the performance of the finite-grid ILP benchmark. The evaluation also reports multi-seed robustness, QoS-violation rates, reward-threshold sensitivity, scalability, UAV-speed effects, and LiFi/WiFi service-reliability behavior.

Asymmetric information is considered a central challenge that significantly affects market efficiency. Its negative impacts, particularly in principal–agent relationships, result in adverse selection and moral hazard. Adverse selection occurs before contracting and is the result of hidden information. On the other hand, moral hazard arises afterward due to the agent’s hidden actions (behavior change of the agent) that shift the risk to the principal. In this paper, we focus on mitigating moral hazard by reducing information asymmetry through utilizing the recent advances of artificial intelligence (AI) methods. Specifically, we introduce an AI-assisted contract-theoretic framework, referred to as LLM-MH, in which the principal receives multiple noisy, unverifiable signals and employs a large language model (LLM) to infer the credibility of the agent’s narratives. We formulate the multi-period moral hazard problem, derive optimal contracts under different informational environments, and incorporate prospect theory to capture bounded rationality in effort choice. The proposed LLM-MH framework, as a soft verification method, enables the principal to use an LLM to evaluate the agent’s multi-signal narratives (such as reports or logs) to construct credibility ratings. By jointly considering the outcomes and the LLM-generated credibility ratings, the proposed framework allows the principal to design incentive-compatible contracts that can substantially mitigate the impact of moral hazard on the principal’s utility. We conduct a set of simulation experiments to investigate the effectiveness of the proposed LLM-based framework. Compared to reference models, the results reveal that the proposed LLM-MH framework effectively manages moral hazard by reducing information asymmetry, learning optimal agent effort, enhancing principal-agent utilities, and preventing strategic manipulation. The proposed LLM-MH framework provides one of the first practical and formal foundations for integrating LLM into contract theory, representing modern organizational environments characterized by complex textual communication.

Indoor wireless systems are increasingly constrained by radio-frequency congestion, motivating hybrid WiFi/LiFi architectures that exploit the complementary strengths of radio-frequency and optical wireless access. A key design challenge is how to complement an existing WiFi deployment with the fewest LiFi access points (APs) while jointly satisfying communication and lighting requirements. This paper studies LiFi AP placement in indoor hybrid WiFi/LiFi networks under data-rate, coverage, illuminance, and inter-AP overlap constraints. The placement problem is formulated as a mixed-integer nonlinear programming (MINLP) problem whose combinatorial structure makes exact solution computationally challenging. Accordingly, we propose a finite-termination deployment heuristic with polynomial grid-evaluation complexity to obtain a scalable solution. The proposed coverage-aware algorithm combines a Mat & eacute;rn hard-core point process (MHCP), grid-based feasibility evaluation, and iterative bisection over the LiFi AP count. MHCP generates spatially separated candidate AP layouts, reducing clustering and redundant overlap, while grid-based relaxation converts continuous rate, coverage, and illuminance constraints into tractable feasibility checks. The bisection stage then identifies the smallest AP count that satisfies the hybrid WiFi/LiFi deployment constraints. Simulation results evaluate the proposed method through small-room benchmark studies and large-room hybrid WiFi/LiFi deployments. In a tractable small-room case, comparison with a grid-relaxed exact integer linear programming (ILP) benchmark shows that the proposed method achieves near-optimal AP counts. Large-room results show that the proposed method satisfies the target quality-of-service, lighting, and overlap constraints while requiring fewer LiFi APs than particle swarm optimization (PSO)-based placement, geometry-based greedy coverage, regular-grid placement, and uniform-random placement. The results also demonstrate the practical value of hybrid WiFi/LiFi integration for indoor coverage and connectivity.

Beamspace massive MIMO with a discrete lens array can significantly reduce the number of RF chains in terahertz systems. However, effective beam selection in such systems requires an accurate beamspace channel estimate, which is difficult to obtain under user mobility because the channel varies rapidly and frequent retraining is prohibitive in wideband THz systems. A priori-aided (PA) tracking mitigates pilot overhead by predicting the dominant beam window of the compressible beamspace channel, but it still has two main limitations. First, it relies on an on-grid angle approximation. Second, it only estimates the channel over a small beam window and forces the remaining coefficients to zero, which leads to a truncation error floor at high SNR even with perfect recovery of the retained coefficients. In this paper, we propose a parametric PA tracking method that overcomes these issues within the PA tracking framework. The key idea is to exploit the known Dirichlet-type pattern of the line-of-sight beamspace channel across beams. This allows us to represent the channel using only two continuous parameters: the spatial direction and a complex gain. In each time slot, we first predict a coarse direction using estimates from previous slots. We then construct a small set of beams around the predicted main lobe. The continuous direction is refined through a one-dimensional search, where, for each trial direction, the complex gain is obtained in closed-form using least squares (with an MAP/ridge regularization for numerical stability). Based on the estimated direction-gain pair, we reconstruct the full beamspace channel without hard truncation and incorporate lightweight reliability checks to suppress rare wrong-lobe updates at low SNR. The proposed method preserves the low pilot overhead of conventional PA tracking while improving estimation accuracy. For a given pilot budget, it achieves a lower normalized mean squared error (NMSE) by eliminating both grid mismatch and truncation errors, and it can operate reliably even with fewer pilots than the PA window size under high-mobility conditions (e.g., Q < V). Simulation results demonstrate significant gains in NMSE and achievable sum-rate compared with conventional orthogonal matching pursuit (OMP)-based estimation and standard PA tracking, as well as Bayesian PA baselines based on Kalman filtering and unscented Kalman filtering, over a wide range of SNRs and pilot budgets; in particular, the proposed method removes the high-SNR NMSE floor of PA and achieves sum-rates close to perfect-CSI beam selection. Finally, we quantify the wideband array-gain loss of phase-only (lens-like) beamforming relative to delay-phase (TTD) beamforming to contextualize the adopted DLA architecture.

The increasing demands for high spectral efficiency, low energy consumption, flexible deployment, and stringent reliability in beyond 5G/6G systems motivate integrating unmanned aerial vehicles (UAVs) with cognitive radio (CR) and multi-antenna (MIMO) technologies. In CR-MIMO UAV networks, CR improves spectrum efficiency by allowing secondary UAVs to opportunistically exploit underutilized licensed spectrum while protecting primary users (PUs). Furthermore, MIMO technology increases spectral and energy efficiency by using spatial multiplexing, diversity, and array/beamforming gains. Due to the UAVs’ limited battery capacity, a key challenge in enabling efficient CR MIMO UAV networking is to maximize the number of served UAVs while minimizing the required transmit power under a set of quality-of-service, power, and spectrum access constraints. To address this, we propose a reliability-aware, batch-based framework for power allocation and channel assignment in CR-MIMO UAV networks. Unlike traditional sequential methods, this batching paradigm assigns power/channels to multiple UAVs simultaneously, resulting in more power-efficient, concurrent UAV transmissions. Specifically, the joint power allocation and channel assignment problem for multiple contending UAVs is formulated as a mixed-integer nonlinear program, which is known to be NP-hard. For a scalable solution, we introduce a two-stage, polynomial-time, batch-based framework that decouples power allocation from channel assignment. First, the framework formulates and solves a convex per-antenna power minimization problem for each UAV-channel pair, enforcing rate, reliability, and power budget constraints, which leads to a closed-form per-antenna power solution. Based on the computed powers, the second stage performs batch-based channel assignment to minimize required transmit power under exclusive-assignment and maximum-matching constraints. This is achieved by formulating and solving a totally unimodular binary linear program that corresponds to a minimum-weight maximum matching problem, which can be solved optimally using the Hopcroft-Karp algorithm. The polynomial-time complexity of the proposed algorithm is established through analytical computational analysis. Simulations in realistic indoor scenarios demonstrate that the proposed approach consistently satisfies the imposed constraints under varying PU traffic, serves more UAVs with higher success probability, and reduces the total transmit power compared to baseline methods with comparable computational complexity for practical network conditions.

Cognitive radio (CR) networks enable unlicensed users to opportunistically access licensed spectrum while protecting primary users (PUs). Under multi-user settings, batch admission can further improve access efficiency as the controller can jointly observe several contending CR users and coordinate channel reservations instead of serving users sequentially. While such an admission policy can offer several benefits, it becomes challenging to realize under proactive jamming and in the presence of uncertain PU returns. This is because a CR transmission can be prematurely terminated whenever it is interrupted by unpredictable PU activity or jamming attacks during the control-and-data interval. Under these conditions, reliable and efficient CR access requires a joint decision over user admission, channel assignment, and packet size while satisfying rate, reliability, and exclusivity constraints. Meanwhile, most existing anti-jamming CR designs are sequential in nature, whereas they mainly optimize channel selection under simplified memoryless activity models. However, the limited batch-based schemes generally do not capture channel assignment with packet-size adaptation and age-aware opportunity information. This paper proposes a learning-based age-aware admission-control batch-based scheduler for CR networks under proactive jamming and in the presence of uncertain PU returns. In particular, the proposed scheduler learns a continuous safe transmission time on each available channel from the observed PU idle age, jammer-free age, and short channel histories. The learned safe time is then combined with the user-specific rate to estimate the expected goodput of each admissible packet size on every feasible user–channel edge. It then assigns each edge the goodput weight associated with its best packet decision and solves a weighted bipartite matching problem to obtain a joint user–channel–packet assignment. Simulation results reveal that the proposed batch-based scheduler consistently improves throughput–reliability trade-off compared to a set of baselines, including sequential and batch-based baselines.

Intelligent reflecting surfaces (IRSs) and hybrid orthogonal frequency-division multiple access/non-orthogonal multiple access (OFDMA–NOMA) can improve connectivity and spectral efficiency in dense cognitive Internet of Things (IoT) networks, but their joint control is challenging under dynamic channel availability and stringent quality-of-service requirements. This paper considers an opportunistic overlay downlink in which a secondary base station jointly determines NOMA power allocation and IRS phase shifts over the idle subchannels. The resulting nonconvex problem includes four hard constraints, the transmit-power budget, idle-channel access, NOMA power ordering, and IRS unit-modulus operation, and two soft probabilistic targets for outage and delay. We develop a feasibility-preserving twin delayed deep deterministic policy gradient method (FP-TD3). Raw actor outputs are mapped through an idle-masked power projection, a monotone NOMA power split, and a unit-modulus phase parameterization, thereby satisfying the four hard constraints for every executed and target action. Outage and delay are incorporated through normalized violation slacks. A heuristic-seeded residual phase representation, projected phase refinement, and candidate-selection shield further improve the per-block reward while preserving hard feasibility. The numerical benchmark assumes perfect instantaneous CSI: before selecting each action, the SBS knows the current direct SBS–SU, SBS–IRS, and IRS–SU channel realizations. Simulations over common channel and availability realizations compare FP-TD3 with penalty-only and projection-based deep reinforcement learning methods, metaheuristics, and non-learning baselines. FP-TD3 achieves the highest sum rate and energy efficiency among the evaluated methods, improves the sum rate by up to 36.99% over TD3+Proj (Joint), maintains zero hard-constraint violations by construction, and requires less than 1 ms per allocation on the tested CPU.

Cognitive radio (CR) enables opportunistic spectrum access while protecting primary users (PUs). This opportunistic transmission is uncertain due to time-varying PU activity (unpredictable PU returns) and heterogeneous channel quality. PU idle times differ across channels, in which some have long mean availability, others only short opportunities, and PU returns vary in severity. In multi-user CR networks (CRNs), contending CR users must coordinate over a control channel to reserve an idle PU channel and select a packet size that maximizes throughput while reducing PU-return-induced forced terminations. Most existing designs optimize channel assignment or packet sizing in isolation, assume memoryless PU idle times, and often ignore control contention/reservation overhead that directly reduces the residual transmission time. This paper investigates joint channel assignment and packet-size control under unknown PU availability while explicitly accounting for realistic control overhead time. We model the Residual PU-idle Time with Idle Age using a distribution-agnostic residual-life survival function and incorporate control overhead into the packet probability of success (PoS). We formulate a maximum-profit, discounted-return problem that maximizes long-term network goodput subject to a required PoS, a received signal-to-noise (SNR) threshold, and exclusive channel occupancy. Due to the uncertainty in PU activity and the long-term nature of the optimization problem, which cannot be solved using conventional optimization methods, we model it as a constrained Markov decision process (CMDP). To solve the resulting CMDP, we propose an on-policy State-Action-Reward-State-Action (SARSA) reinforcement learning (RL) algorithm with convergence-based exploration and a quantized/pruned action-space mechanism that filters infeasible channel–packet-size pairs to reduce online search complexity. The proposed RL-based algorithm learns perchannel PU dynamics online and jointly selects the throughput-maximizing channel and packet size. Our method is a hybrid model-assisted/model-free design, in which online survival/quantile estimates are used only for PoS-based feasibility screening, while SARSA learns the long-term policy directly from interaction without requiring a parametric PU model. Simulations under heterogeneous PU availability/severity show up to 32% throughput improvement, with fewer retransmission attempts than reference schemes.

Indoor wireless systems are increasingly constrained by radio-frequency congestion, motivating hybrid WiFi/LiFi architectures that exploit the complementary strengths of radio-frequency and optical wireless access. A key design challenge is how to complement an existingWiFi deployment with the fewest LiFi access points (APs) while jointly satisfying communication and lighting requirements. This paper studies LiFi AP placement in indoor hybrid WiFi/LiFi networks under data-rate, coverage, illuminance, and inter-AP overlap constraints. The placement problem is formulated as a mixed-integer nonlinear programming (MINLP) problem whose combinatorial structure makes exact solution computationally challenging. Accordingly, we propose a finite-termination deployment heuristic with polynomial grid-evaluation complexity to obtain a scalable solution. The proposed coverage-aware algorithm combines a Matérn hard-core point process (MHCP), grid-based feasibility evaluation, and iterative bisection over the LiFi AP count. MHCP generates spatially separated candidate AP layouts, reducing clustering and redundant overlap, while grid-based relaxation converts continuous rate, coverage, and illuminance constraints into tractable feasibility checks. The bisection stage then identifies the smallest AP count that satisfies the hybrid WiFi/LiFi deployment constraints. Simulation results evaluate the proposed method through small-room benchmark studies and large-room hybrid WiFi/LiFi deployments. In a tractable small-room case, comparison with a grid-relaxed exact integer linear programming (ILP) benchmark shows that the proposed method achieves near-optimal AP counts. Large-room results show that the proposed method satisfies the target quality-of-service, lighting, and overlap constraints while requiring fewer LiFi APs than particle swarm optimization (PSO)-based placement, geometry-based greedy coverage, regular-grid placement, and uniform-random placement. The results also demonstrate the practical value of hybridWiFi/LiFi integration for indoor coverage and connectivity.

Commercial lending markets are affected by asymmetric information, in which borrowers possess private knowledge about financial condition, operational resilience, and risk exposure, while lenders observe incomplete and noisy signals. This imbalance can increase adverse selection, moral hazard, mispriced credit, and the exclusion of creditworthy borrowers. This paper proposes a Quantum-Enhanced Credit Intelligence Framework (QECIF), a hybrid quantum-classical architecture that integrates quantum machine learning (QML), the Quantum Approximate Optimization Algorithm (QAOA), quantum amplitude-estimation-inspired risk analytics, and quantum-secure data exchange. The paper formalizes asymmetric information through a Bayesian latent-type model, defines the major mathematical symbols, explains the architecture layer by layer, and provides an end-to-end workflow for credit assessment and loan portfolio optimization. The contribution is architectural and conceptual, i.e., QECIF does not assume immediate quantum superiority, but identifies credit-risk subproblems where quantum methods can be evaluated under noisy intermediate-scale quantum constraints and future fault-tolerant settings.