Engineering

Abstract: The defining workload of 6G is expected to be machine-native intelligence rather than human-driven data transfer, yet no study has measured the arrival process a deployed autonomous agent generates on a real radio link. This letter contributes the first empirical characterization of real agent traffic from a deployed platform over a real cellular link (extending prior controlled testbed measurements to a live wireless edge client), observed via application-layer telemetry, which captures the coupled generation–transmission process without packet capture. We propose a two-level framework—withintask (Level 1) versus cross-task pooled (Level 2)—and show that the pooled heavy tail is quantitatively explained by log-variance additivity, dominated by within-task log-variance (∼ 83%) with a smaller cross-task heterogeneity contribution (∼ 17%). Maximum-likelihood fitting with KS and AIC selects log-normal inter-arrival times and decisively rejects Poisson (\(p=7.9\times10^{-123}\)). The findings yield quantitative, p99-based uplink capacity-planning rules for 6G edge clients.

Abstract: Space-based quantum key distribution (QKD) has emerged as the leading approach for establishing information-theoretically secure communication links beyond the reach of terrestrial fiber networks. Over the past decade, missions ranging from the 635 kg Micius satellite to 3U CubeSats such as SpooQy-1 have demonstrated satellite-to-ground QKD, entanglement distribution, and quantum teleportation across distances exceeding 1,200 km. However, existing reviews of these missions focus primarily on quantum protocols and link layer physics, leaving the system level engineering trade offs particularly those governing platform selection, SWaP budgets, and subsystem integration largely unexamined. This paper presents a systems engineering review of space based quantum communication missions developed between 2016 and 2025. Each mission is analyzed through a unified framework encompassing link budget decomposition, payload miniaturization trends, and Acquisition, Tracking, and Pointing (ATP) architectures. A comparative SWaP analysis reveals that payload mass has decreased by approximately one order of magnitude from ~230 kg (Micius) to ~23 kg (Jinan-1) while reported quantum bit error rates have remained below 5%. The analysis identifies three critical gaps: (i) the absence of formal Multidisciplinary Design Optimization (MDO) frameworks in quantum satellite design, (ii) the lack of standardized performance metrics for cross mission comparison, and (iii) the disconnect between constellation level network studies and subsystem level hardware constraints. Based on these findings, system-level design strategies and a technology roadmap are proposed to support the transition from single-link demonstrations to scalable, constellation-ready quantum communication infrastructures.

Abstract: Telecom taught the world how to run heterogeneous generations in parallel for decades (1G–5G and beyond). This paper formalizes that analogy for artificial intelligence (AI), where rule‑based, statistical/ML, deep learning, foundation models, and agentic systems will coexist for the foreseeable future. We make five contributions: (1) a formalized mapping between coexistence phenomena in mobile networks and AI multi‑version ecosystems; (2) a layered architecture for version‑aware AI orchestration with capability negotiation and sandboxed execution; (3) two analytical risk models quantifying interoperability and security exposure under coexistence; (4) a structured threat analysis with attack paths unique to cross‑model and cross‑agent amplification; and (5) a quantitative evaluation protocol encompassing semantic drift, safety policy violation rates, and change‑failure risk during phased upgrades. We detail engineering and governance controls that mirror proven telecom practices (multi‑RAT, phased migration, telemetry‑driven operations) which can help multi variant AI systems.

Abstract: Web tracking and browser fingerprinting are typically invisible to users and are difficult to characterize reliably from a single website observation. This study evaluates privacy- and security-relevant indicators across a curated frame of 947 Ecuador-associated domains and introduces a completion-aware automation and data-engineering pipeline that uses Blacklight as an external measurement instrument. The developed application normalizes and deduplicates domains, controls a headless Chromium browser through Playwright, initiates scans through the Blacklight web interface, waits until the platform signals scan completion or a maximum timeout is reached, and, only after completion, captures and preserves the original JSON download. For each artifact, the application records acquisition metadata and errors, computes a SHA-256 integrity hash, extracts instrument-reported variables, validates summary-versus-detail consistency, and joins a manually established sector/subsector taxonomy while preserving raw, derived, and classification layers separately. Three measurements were conducted on 25 June 2026, 25 July 2026, and 25 August 2026. Successful JSON export rates were 58.0%, 55.8%, and 59.0%; the strict quality-controlled longitudinal panel contained 381 domains across 19 sectors. Within that panel, tracker prevalence was 15.0%-15.5%, fingerprinting 37.5%-38.1%, cookies 34.1%-34.4%, third-party domains 74.3%-74.8%, Content Security Policy (CSP) presence 22.3%-22.6%, and instrument-defined critical-category fingerprinting 7.3%-8.1%. None of the six binary conditions changed significantly after Holm correction (all adjusted p >= 0.812), with three-round exact agreement of 96.6%-99.5% and Fleiss’ kappa of 0.927-0.986. Count reliability was also high (ICC(A,1) = 0.897-0.993). Persistent domain-item pairs accounted for 89.2% of tracker-domain, 89.1% of tracker-company, 91.3% of fingerprinting-technique, and 95.5% of third-party-domain detections. Under Ecuador’s Organic Law on Personal Data Protection (LOPDP), these results are interpreted as persistent compliance-verification signals rather than proof of unlawful processing. The combined framework strengthens reproducibility, audit prioritization, and evidentiary traceability while preserving the distinction between technical detection and legal conclusion.

Abstract: Underwater visible light communication (UVLC) offers a high-speed alternative to acoustic and radio-frequency underwater links, but its performance is highly sensitive to channel impairments such as bubble-induced scattering. This paper presents an RGB wavelength-division-multiplexed UVLC system simulated in OptiSystem across a five-point turbidity gradient from clear to bubble-induced conditions, evaluated using Q-factor, bit error rate, bandwidth, and eye diagram metrics. An eleven-feature machine learning framework extracted from eye diagrams is classified using an Extreme Gradient Boosting model under a stratified, group-aware protocol, achieving 98.75% test accuracy and 96.18% (±2.17%) repeated cross-validation accuracy, statistically validated against support vector machine and Random Forest classifiers via a paired t-test. Physical-layer results reveal a monotonic wavelength-dependent degradation that becomes markedly non-uniform across the turbidity gradient, and an independently computed Mie scattering analysis shows bubble-induced scattering converging toward a wavelength-insensitive regime at physically realistic bubble radii, consistent with the classifier’s confusion pattern. A classification-triggered adaptive power strategy, validated through direct re-simulation, improves bit error rate by four to nine orders of magnitude under bubble-induced conditions, while Shannon capacity analysis indicates multi-gigabit-per-second headroom even under degraded conditions. These results demonstrate a physically grounded, validated framework integrating simulation, classification, and adaptive control for intelligent underwater optical communication.

Abstract: In recent years, the Digital Twin paradigm has emerged as a key enabler for the monitoring, analysis, and optimization of complex systems through the continuous interaction between physical entities and their digital counterparts. In the context of communication networks, the application of this paradigm has led to the concept of Network Digital Twins (NDTs), which aim to provide accurate and continuously synchronized representations of networks throughout their entire operational lifecycle. However, despite the growing number of architectural proposals and conceptual frameworks, the practical deployment of fully operational Network Digital Twins remains a significant challenge. In particular, existing approaches often assume the availability of structured, timely, and homogeneous network data, overlooking the inherent complexity of acquiring, normalizing, and maintaining such information across heterogeneous network infrastructures. This article addresses this challenge by presenting DANA, a lifecycle-aware system specifically designed for network data acquisition in Network Digital Twin environments. Rather than proposing a new digital twin architecture, the contribution focuses on the systematic collection, normalization, and dissemination of network data required for both the initial creation and the continuous operation of Network Digital Twins. DANA follows a device-agnostic design, enabling interaction with heterogeneous network devices and systems, and explicitly distinguishes between data acquisition for initial modeling and for runtime monitoring and synchronization. The system adopts a modular architecture and an event-driven publish/subscribe communication model to support scalable and bidirectional information exchange between the physical and digital domains. The proposed solution is experimentally validated through realistic network scenarios based on network emulation and container-based orchestration platforms, demonstrating its applicability to practical and operational Network Digital Twin deployments.

Abstract: Reliable ship-to-shore backhaul is essential for maritime Internet of Things (IoT) data delivery, yet cellular and satellite connectivity varies by route, operating phase, and qualification criterion. This study analyses 606,625 georeferenced monitoring records collected during a nine-month, 12-voyage campaign aboard a 1,800-TEU (twenty-foot equivalent unit) container ship operating between Korea and Southeast Asia, with the cellular–GEO analysis limited to the active VSAT service period. During sailing, cellular attachment was reported for 70.3% of valid cellular-state records, but only 28.0% satisfied the adopted RAT-specific received-power criteria, with leg-level qualified fractions ranging from 77.1% (Incheon–Busan) to 17.3% (Shanghai–Ho Chi Minh). Within the joint-analysis window, either the cellular criterion was satisfied or a valid GEO probe response was observed in 99.40% of records, although the residual gap reached 3.11% on Laem Chabang–Ho Chi Minh. Under retrospective cellular-first allocation, raising the candidate VSAT SNR threshold from 6 to 7 dB increased the store-and-forward share from 6.70% to 25.73% without reducing the median RTT of the retained GEO records, and the longest buffered interval grew from 2.02 to 9.86 h. These results show that route segment, operating phase, state definition, and threshold selection materially influence link allocation and buffering implications; live traffic steering and application-level availability were not evaluated.

Abstract: In this study, we present the architecture, implementation, and optimization of an intelligent substation system that integrates private 5G networks, multi-access edge computing (MEC), and AI-powered sensing. The deployment of wireless networks in high-voltage substations involves unique challenges owing to severe electromagnetic interference (EMI) and multipath fading caused by dense metal structures. To address these issues, we propose a hierarchical edge-cloud orchestration architecture combined with an adaptive mobility management (AMM) scheme. The AMM scheme dynamically optimizes handover parameters—specifically time-to-trigger (TTT) and hysteresis—to mitigate radio link failures during robotic inspections. We deployed the proposed system across two operational substations and evaluated its performance under real-world conditions. The results indicate that the proposed optimization stabilizes the end-to-end latency below 50 ms in 95% of cases and achieves a handover success rate of 100% even in Non-Line-of-Sight (NLOS) regions. Anomaly detection models hosted on the MEC achieved accuracy of 94% across the partial discharge, thermal, and visual domains. Experimental analysis demonstrates that the proposed architecture significantly outperforms the conventional static configurations in terms of uplink throughput reliability and connection stability, thereby presenting a practical blueprint for modernizing substation automation.

Abstract: Unmanned Aerial Vehicles (UAVs) are revolutionizing power grid inspection, yet ensuring reliable, high-quality image transmission in complex environments remains a significant challenge for autonomous operations. Traditional path planning algorithms often neglect communication constraints, risking mission failure in areas with poor connectivity—a critical concern for future UAV swarm deployments where inter-agent coordination depends on sustained link quality. This paper proposes a novel communication-aware path planning framework to address this gap. We formulate the problem as a multi-objective optimization that simultaneously considers safety, complete inspection coverage, path length, and communication quality. The core of our solution is an Improved Grey Wolf Optimizer (IGWO), which incorporates a nonlinear convergence factor, a genetic mutation strategy, and a modified leader update mechanism to enhance global search capability and avoid local optima. To evaluate the proposed framework, we adopt a dual-scenario validation strategy: in the mountainous setting, we employ a synthetic conical-peak terrain model with the standardized 3rd Generation Partnership Project (3GPP) Rural Macrocell (RMa) channel model; in the urban setting, we use real-world building footprints from OpenStreetMap (Hangzhou, China) with deterministic ray-tracing propagation modeling. Comprehensive simulations in both scenarios demonstrate that IGWO achieves a 50% success rate in dense urban areas, significantly outperforming Particle Swarm Optimization (PSO) at 10% and Genetic Algorithm (GA) at 28%. Crucially, communication-aware paths reduce average outage duration by over 65% while often yielding shorter paths. This work confirms that explicit communication modeling is indispensable for reliable UAV inspection and that the proposed IGWO offers a robust and scalable solution for future autonomous and cooperative power grid maintenance systems.

Abstract: Short-term fluctuations in reference signal received power (RSRP) under high-mobility conditions can repeatedly reset the time-to-trigger (TTT) timer and cause spatial dispersion of Event A3 handover triggers in high-speed railway 5G-R systems. To address this problem, this study develops a physics-driven, Doppler-aware RSRP input-calibration framework that preserves the standardized handover margin (HOM), TTT, Event A3 semantics, and Radio Resource Control (RRC) procedure. The relative residual Doppler shift is derived from train kinematics and base-station geometry and estimated online using a scalar Kalman filter. The resulting estimate is mapped to an orthogonal frequency-division multiplexing frequency-offset-induced equivalent RSRP loss subject to a physical upper bound. Geometry-based detrending, speed-adaptive causal smoothing of the residual component, and a low-weight bounded position-aided correction are subsequently applied to reconstruct the serving- and candidate-cell RSRP inputs. Event-driven simulations covering viaduct and mountainous railway scenarios, together with ablation, Doppler-sensitivity, speed-adaptability, and observation-error analyses, are conducted to evaluate the proposed method. At a train speed of 350 km/h, the method reduces the root-mean-square error of the inter-cell RSRP difference from 2.3449 to 1.9053 dB and decreases the mean handover trigger-location error from 54.79 to 42.35 m, while maintaining a handover success rate of 99.49%, comparable to the 99.29% achieved by conventional Event A3. The results further demonstrate that, under a carrier frequency of 2.1 GHz and a subcarrier spacing of 30 kHz, the overall performance improvement is governed primarily by geometric detrending, speed-adaptive residual smoothing, and bounded position correction, whereas Doppler-loss compensation provides a small but physically consistent correction of the deterministic measurement bias. The proposed framework therefore improves RSRP input quality and handover trigger-location consistency without modifying the standardized handover decision logic.

Abstract: Frequency-hopping spread spectrum (FHSS) information and communication systems are widely used to enhance the resilience of wireless networks to interference. The effectiveness of traditional algorithms is significantly reduced in the presence of active and intelligent jammers. The aim of this manuscript is to develop an adaptive FHSS algorithm based on statistical learning of frequency-channel quality and to analyze its effectiveness compared with existing algorithms. A software simulator has been developed that implements six FHSS algorithms, five jamming models, and statistical evaluation using the Monte Carlo method based on the bit error rate, error vector magnitude, and normalized throughput metrics. The software-implemented adaptive algorithm delivered the best results among all those studied, achieving the highest overall performance metric. Compared with a system without FHSS, 64.23 % reduction in the bit error rate, 22.76 % increase in throughput and 59.32 % reduction in the error vector magnitude were achieved. The results obtained confirm the feasibility of using adaptive statistical frequency channel selection to improve the noise immunity of FHSS systems. Promising areas for further research include the modeling of intelligent jammers and the experimental verification of the implemented algorithm on computing platforms with limited resources.

Abstract: Sixth-generation (6G) networks will be AI-native: machine-learning models will decide how radio resources are allocated, where functions are placed, which control loops run, and how traffic is routed. Recent orchestration frameworks show that sharing common AI functions and xApps across services yields large resource and revenue gains, yet these gains are computed as if security were free. In reality, every shared function must also satisfy authentication, trust, confidentiality, attestation, and auditability requirements—and adding these requirements changes the structure of the underlying placement and sharing optimisation problems. This article presents a cross-layer framework for making AI-native 6G radio access networks secure, privacy-preserving, and revenue-aware at the same time. The framework is built from nine interlocking pillars organised in three layers: (i) an O-RAN orchestration substrate covering functional-split-aware placement, conflict-aware RIC control-loop scheduling, and joint radio–compute slicing; (ii) a confidential intelligence layer covering trust-weighted robust federated learning, attested confidential edge inference with hybrid post-quantum sessions, and security-aware xApp sharing; and (iii) network-wide assurance covering security-aware revenue-optimal orchestration, cross-domain zero-trust slice access, and an end-to-end resilience and formal-verification capstone. Each pillar is presented with a research question, a precise problem formulation with objective and constraints, a candidate solution strategy expressed as pseudocode, and evaluation methodology. We further show how the nine pillars interlock through four vertical threads—placement–sharing, trust, post-quantum protection, and audit—so that, taken together, they answer a single central question: can security constraints be made first-class terms of revenue-optimal 6G orchestration while the resulting system remains scalable, provable, and practical?

Abstract: Rapid, reliable, and energy-efficient data collection is essential for disaster response, where terrestrial communication networks are often disrupted or unavailable. Unmanned Aerial Vehicles (UAVs) provide a flexible solution for collecting critical sensing data, but their operation is constrained by limited onboard energy, stochastic wireless conditions, complex three-dimensional environments, and stringent latency requirements. This paper proposes a structured multi-UAV framework for disaster-response data collection that separates the optimisation process into spatial, temporal, and safety layers. In the spatial layer, three-dimensional Travelling Salesman Problem with Neighbourhoods (3D-TSPN) modelling allows UAVs to collect sensor data by entering sensing regions rather than visiting exact sensor coordinates. An Age of Information (AoI)-aware Genetic Algorithm (GA) is then used to optimise the sensor visiting order, while RRT-Connect verifies obstacle-aware feasible paths in the 3D environment. In the temporal layer, a Lyapunov-based control mechanism makes binary computation decisions between local processing and offloading to a single MEC node, balancing delay, energy consumption, AoI, and queue stability. In the safety layer, continuous-time conflict detection and temporal offset resolution are used to monitor inter-UAV separation and obstacle-related risks. The proposed framework is evaluated through simulation under wireless, mobility, computation, and obstacle constraints. The results show that the framework achieves a 100% data collection success rate, zero obstacle collisions, zero inter-UAV collisions, and no dropped tasks. Compared with planning-oriented and MEC-oriented baselines, the proposed framework improves information freshness, average delay, processing delay, energy efficiency, and system cost, while maintaining reliable multi-UAV coordination in a complex 3D environment. In addition, the layered design clarifies the role of each component: 3D-TSPN improves spatial flexibility, AoI-aware GA enhances route ordering, RRT-Connect supports obstacle-aware path feasibility, Lyapunov-based binary offloading stabilises computation decisions, and safety monitoring improves operational reliability. These findings demonstrate that a simplified layer-based integration of mobility planning, computation control, and safety monitoring can provide an effective solution for multi-UAV disaster-response data collection in complex 3D environments.

Abstract: Joint scheduling and resource allocation are investigated in heterogeneous queuing systems—where multiple queues have distinct priorities and share a single output link—with the aim of maximizing throughput utility under per-queue delay constraints. The primary challenge arises from stochastic packet arrivals and the time-varying capacity of the shared link. To address this challenge, conventional heuristic methods and unconstrained learning-based methods exhibit inherent limitations, particularly when handling bursty traffic with stringent delay requirements. The former lack real-time reactivity to system dynamics; the latter, which encode delay constraints into the reward, may sacrifice deadline compliance for higher cumulative rewards. To this end, this paper proposes a constrained soft actor-critic (CSAC) approach with two key designs. First, we decouple the delay constraint of the queue with a stringent delay requirement from the reward, treating it as a standalone violation budget. Second, we introduce a two-stage mapping mechanism that transforms continuous policy logits into integer transmission quotas. We compare the proposed approach with an unconstrained soft actor-critic (SAC) baseline and several heuristic baselines under heterogeneous traffic conditions. The experimental setup comprises one Markov-modulated Poisson process (MMPP) highest-priority queue with a maximum allowable queuing delay violation rate of 1%, along with two independent Poisson process lower-priority queues with distinct priorities, each with a best-effort queuing delay requirement. Results demonstrate that the proposed approach reduces the highest-priority queue’s delay violation rate to 0.05%±0.09%, compared to 7.87%±5.98% for unconstrained SAC and 18.38% and 34.28% for the two heuristic baselines, while maintaining comparable throughput, lower-priority delays, and overflow packet counts.

Abstract: 6G antenna and intelligent reflecting surface (IRS) performance is bottlenecked by substrate material properties. This paper uses the IRS Tri-Constraint (360° phase range, sub-1 dB insertion loss, thermal stability from −20°C to +60°C) to compare liquid crystal polymer (LCP), graphene-on-quartz, and PVDF at 140 GHz. A transparent link budget translates each material's properties into a capacity-based and a coverage-based subscriber count. Graphene ranks highest on both measures, narrowly ahead of LCP, with PVDF trailing substantially. A 200,000-trial Monte Carlo analysis shows the ranking is robust (99.3% joint) under idealised assumptions but far less robust (3.0%) when literature-reported fabricated-device insertion losses are substituted, indicating the ranking currently outpaces demonstrated hardware.

Abstract: A recurrent challenge in scaling ultra-wideband (UWB) motion-capture systems is interference management whenmanyranging transactions coexist in time and space. To address this, we study a two-layer localization architecture that separates field-level player localization from local on-body pose tracking, allowing the two tasks to operate with different communication regimes and spatial-reuse policies. A stochastic-geometry framework is used to map sport-dependent parameters, including player density, field size, tag count, anchor count, update rates, and ranging airtime, to reliability and update-rate tradeoffs. The analytical model is parametrized based on controlled experiments that characterize ranging success under temporal overlap, player distance, and variable-delay scheduling, which we use to design a proximity-aware local coordination strategy. We apply our proposed approach to soccer, volleyball, and ice hockey as representative use cases. Our results show that proximity-aware coordination can provide a scalable and lightweight interference management mechanism. Coordination is activated only where local player clustering creates strong interference, while spatially separated players continue to share resources without coordination. For the highest density scenario tested, this increases the local-layer ranging success from below 50% without coordination to over 80% in fourand eight-player congestion clusters, while avoiding network-wide coordination overhead.

Abstract: Pneumonia detection in chest X-rays is commonly treated as a binary classification task, although the radiographic burden of disease varies substantially across patients. This study investigates whether the expert bounding-box annotations available in the RSNA 2018 Pneumonia Detection dataset can be reused to derive an interpretable severity grading framework based on lesion extent. For each pneumonia-positive image, total disease burden was computed as the cumulative area of all annotated opacity boxes, and a balanced three-tier split was generated, defining Severity 1 (low/mild), Severity 2 (moderate), and Severity 3 (severe). The resulting severity groups contained 1999, 2006, and 2007 images, respectively. To validate the usefulness of this annotation-driven severity formulation, we trained Vision Transformer models in both a classical binary setting and a severity-specialized setup. The classical binary baseline reached a maximum validation accuracy of 95.05%, while the specialized models achieved 94.98% for Severity 1, 97.88% for Severity 2, and 99.08% for Severity 3. These results indicate that radiographic burden derived from bounding-box extent supports highly separable severity-aware subproblems, particularly for severe pneumonia, while lower-burden categories remain more difficult. The proposed framework offers a transparent and reproducible way to transform RSNA 2018 lesion annotations into ordinal severity labels. Beyond its original use as a detection and localization benchmark, the dataset can therefore also support severity-aware classification, radiographic burden analysis, and future explainable AI studies in chest X-ray interpretation.

Abstract: Next-generation wireless systems are becoming more complex, and the need for intelligent mobility management mechanisms that can ensure service continuity and efficiently utilize network resources has been growing. Frequent handovers, unequal distribution of traffic, variable network conditions, and multiple radio access technologies are some of the challenges that traditional mobility control strategies are likely to face in 5G and future 6G networks with dense deployments of small cells, heterogeneous architectures, and highly mobile users. These constraints frequently lead to sub-optimal user experience, higher overhead in signalling and inefficient use of resources. The introduction of new tools through the advancements of artificial intelligence (AI), specifically machine learning and deep learning techniques have created new opportunities for "predictive" and "adaptive" mobility optimization. The use of data-driven decision-making can enable AI-based solutions to predict user movements, fine-tune the execution of handover and dynamically allocate radio resources to enhance network performance. This survey This paper presents a comprehensive survey of AI-enabled handover strategies for mobility load management 5G, Beyond-5G (B5G), and upcoming 6G networks. This study provides a review of current studies, classifies the framework approaches of mobility management based on AI technologies, and identifies their architecture, learning and optimization goals. Moreover, the survey assesses the performance of intelligent handover schemes to solve the critical issues like load balancing, interference mitigation, connection reliability and quality of service maintenance. Key performance indicators related to mobility robustness, resource efficiency and service continuity are compared between the conventional mobility management methods and the AI based ones. Last but not least, the paper outlines unsolved problems, new trends and potential areas of research that will guide the evolution of autonomous mobility management solutions for the future of wireless communication networks.

Abstract: Large-scale network operations require engineers to work across heterogeneous tools and dashboards, which can lengthen Mean Time to Repair (MTTR) and affect availability when it delays incident resolution. We present an agentic ChatOps system in which a supervisor orchestrates large language model (LLM) agents that route natural-language intents to specialized workers issuing planned tool calls across operations systems, with retrieval-augmented grounding in a network source of truth. We deploy the approach in a major global network—a deployment that has since grown to fourteen specialized workers plus dedicated expert and autonomous agents—and detail seven representative use cases, including closed-loop link-flapping remediation that validates candidate changes in a digital twin before committing version-controlled configuration. Across six recurring tasks, the integrated ChatOps automation system was associated with per-event handling times lower by roughly 10× to 2400×, and the link-flap cycle was shortened from about 30 to 3 minutes. Over a representative 90-day window it handled roughly 7,400 production interactions, with positive feedback on most rated responses; an offline benchmark of 200 questions scored by an LLM-as-a-judge yielded mean relevance and context relevance of 0.85 and 0.79. The results support modeled availability improvement when saved handling time reduces incident MTTR, in a deployment designed for confidentiality and guarded by validation pre-checks.

Abstract: This paper presents a robust and efficient Convolutional Neural Network (CNN)-based channel estimator for fifth-generation (5G) Orthogonal Frequency Division Multiplexing (OFDM) systems. While conventional Least Squares (LS) and Minimum Mean Square Error (MMSE) estimators degrade significantly in high-mobility, non-linear, and frequency-selective fading environments, the proposed framework treats the time-frequency resource grid as a spatial image, enabling implicit learning of complex fading dynamics without explicit statistical modeling. The model is trained on 105 synthetic channel realizations spanning Rayleigh, Rician (K = 5 dB), AWGN, CDL-A, CDL-B, and TDL-A channel profiles and validated through rigorous MATLAB simulations. Key quantitative results demonstrate: (i) a 7-fold BER reduction over MMSE at 20 dB SNR on CDL- A (1.0 × 10⁻³ vs. 7.0 × 10⁻³); (ii) a 3–5 dB NMSE improvement across the full 0–30 dB SNR range; (iii) robust performance under Doppler spreads up to 300 km/h with less than 0.5 dB BER penalty; (iv) a 50% reduction in pilot overhead while maintaining superior MSE performance; and (v) spectral efficiency within 0.35 bits/s/Hz of the perfect-CSI Shannon bound. With a measured inference latency of 0.8 ms and a lightweight design of 2.3 × 10⁶ parameters, the proposed CNN-CE is validated as a practically deployable and resource-efficient technology for 5G and beyond-5G (B5G) networks.