This study investigates deployment-dependent link quality of LoRa telemetry for volcanic environmental monitoring through controlled antenna-configuration testing, multi-site field measurements, repeated-site observations across campaigns, and out-of-deployment validation at Kawah Putih, West Java, Indonesia. The analysis includes 677 observations from a controlled antenna experiment and 17,792 operational telemetry records collected during three field campaigns. The two datasets were analyzed separately and together represent 18,469 measured observations. In the controlled experiment, the Omni-to-Omni configuration achieved the highest median RSSI of −62 dBm. Clear differences were observed among antenna configurations, with large effect sizes for both RSSI (ε² = 0.873) and SNR (ε² = 0.786). At monitoring areas revisited in different campaigns, median RSSI changed by more than 40 dB, indicating substantial deployment-dependent variation even within the same canonical monitoring areas. Campaign identity was associated with the largest increase in explanatory performance, with ΔR² = 0.543 for RSSI and ΔR² = 0.512 for SNR. Although the explanatory models accounted for approximately 65% of the variability within the observed deployments, this performance did not translate into equally strong prediction under unseen field conditions. Among the fitted predictive models, the lowest mean leave-one-campaign-out RMSE was 18.98 dB for RSSI and 6.19 dB for SNR, while the corresponding out-of-sample R² values remained negative. These findings show that LoRa link quality can vary substantially with deployment conditions and that signal-level performance observed during one field campaign should not be assumed to remain consistent in subsequent deployments. Cross-configuration, multi-site, and cross-campaign evaluation is therefore important for assessing the transferability of LoRa links under field conditions.
The integration of tokenized real-world assets into blockchain-based financial systems is progressing more rapidly than the development of compliance tools. Predominantly, decentralized finance protocols implement compliance controls at the application or smart-contract layer. This approach presents a challenge: participants who engage directly with the underlying protocol can circumvent these controls, resulting in inconsistent enforcement among different actors. Additionally, the rules are static, making it difficult to maintain their relevance as jurisdictional requirements evolve. Although AI models can interpret changing regulations, there is currently no mechanism that links this reasoning to enforcement that is both deterministic and immune to circumvention at the protocol level. To address this issue, we propose the AI-Assisted Runtime Compliance Architecture. This architecture integrates rules into the core of the protocol and connects them to an AI compliance model, rather than layering them on top. It comprises four collaborative layers. The intelligence layer, trained on legal parameters across various jurisdictions and updated through a retrieval pipeline, generates explainable determinations. The verification layer employs zero-knowledge proofs to ensure participant data privacy. The runtime enforcement layer, positioned below the application layer, is deterministic by consensus and impervious to bypass. The asset and governance layer manages the assets, connects with external regulatory data, and looks after system updates. This architecture makes sure that only the final compliance check results are stored on the blockchain. When evaluated against current failure modes, this architecture addresses all five identified problem dimensions. A preliminary reference implementation and automated test suite, covering all four layers and each identified threat class, confirm that the architecture’s core mechanisms behave as designed, establishing an initial empirical foundation for the larger-scale, multi-jurisdictional validation identified as future work.
Digital transformation (DX) in Higher Education Institutions (HEIs) is considered both important and complex. Yet, its main features are not clearly defined and are seldom viewed through the lens of excellence. This study introduces a new conceptual framework for the main DX dimensions in HEIs and explores how these dimensions align with the Baldrige Education Framework. The study used a two-phase approach. First, a Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guideline systematic literature review (SLR) identified 16 eligible studies for qualitative analysis. After a Critical Appraisal Skills Programme (CASP) check, 10 studies were retained for qualitative analysis. Using NVivo and BERTopic semantic grouping, we identified 8 main dimensions: Digital Learning, Digital Strategy, Stakeholder Engagement, Digital Readiness, Transformation Frameworks, Digital Leadership, Pedagogical Capacity, and Performance Evaluation. Second, these dimensions were compared and aligned with the Baldrige Education Framework. Our findings show that these dimensions are closely connected to the Baldrige Education Framework criteria, except for Digital Readiness, which serves as a cross-cutting enabling dimension rather than a discrete category. This study provides a provisional framework to help with research, institutional assessment, and planning for transformation across HEIs.
The rapid adoption of large language models (LLMs) in cybersecurity has generated a growing and diverse body of research spanning threat detection, vulnerability analysis, malware investigation, incident response, and security automation. Despite this growth, existing reviews largely organise the literature by application domain and few relate research activity to operational threat models. This study addresses this gap through a systematic mapping study of 246 peer-reviewed publications published between 2023 and 2026. Following PRISMA guidelines, the literature is analysed through the MITRE ATT&CK framework, providing a structured mapping of LLM-based cybersecurity research to adversarial tactics, techniques, and mitigations. The findings indicate an uneven distribution of research effort across the cybersecurity lifecycle, with substantial concentration in selected ATT&CK areas and limited coverage of several operationally important tactics. The analysis also highlights recurring challenges related to data availability, robustness, scalability, generalisation, and reproducibility. Beyond synthesising current research activity, the study provides an ATT&CK-grounded perspective on the field that supports the identification of research gaps, facilitates comparison across application domains, and establishes a foundation for future investigations of LLM-enabled cybersecurity systems.
This paper presents a hierarchical mode-based stability analysis of a 50-W Ku-band GaN monolithic microwave integrated circuit (MMIC) high-power amplifier and its experimental validation. The amplifier combines two four-stage, 1:2:4:8 corporate-combined unit amplifiers through balanced Lange couplers. Each eight-device output stage has seven independent odd modes whose oscillations are invisible to external measurements and to conventional two-port stability metrics. From the combining hierarchy, a complete set of eight mode-selective excitations is constructed and analyzed by common-denominator pole-zero identification at all sixteen gate and drain ports with a per-port observability factor ρ. Under even-mode excitation the right-half-plane poles are only weakly observed, a consequence of layout-induced asymmetry that leaves the instability inaccessible to symmetric observation. Comparing ρ across the hierarchical modes resolves it into the pair- and group-level odd modes, whose virtual grounds are confirmed as effective stabilization locations. Resistors placed there eliminate the unstable poles in all eight modes. The method thus reduces the stabilization of a large parallel-combined stage to a systematic sequence of detection, mode classification, and virtual-ground placement. It is validated on a fabricated 0.25-μm GaN-on-SiC amplifier delivering an average of 50.3 W (40.1–55.6 W) with a small-signal gain of 26–33 dB and a large-signal gain of 22.0–23.45 dB over 15.25–16.5 GHz.
