Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Kritphon Phanrattanachai has a total of 31 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

    • 18 November 2026
    • Published by MDPI AG in Preprints.org
      • https://doi.org/10.20944/preprints202611.1922.v1
    Open Access

    A Marchand Balun Based on Centered Symmetric Inductive Compensation for High-Impedance Transformation

    by Jitjark Nualkham, Chaiya Chomchalao, Kritphon Phanrattanachai, Somchat Sonasang, Niwat Angkawisittpan, Ravee Phromloungsri

    This paper presents a Marchand balun with centered symmetric inductive compensation for high-impedance transformation. Identical inductive discontinuities, modeled as series inductances, sit at the midpoint of the coupled-line conductors of both sections to offset the unequal even- and odd-mode phase velocities while preserving symmetry. Modal ABCD matrices and a nodal reduction to S-parameters give a closed-form compensation inductance that sets the output balance at the design frequency, and a differential-evolution optimization of the network model maps the balanced bandwidth against the coupling level. At the coupling of the fabricated design, the model predicts a simultaneous fractional bandwidth of 26.22% with the closed-form inductance and 37.78% with an optimized inductance of about half that value, against 6.67% without compensation. A 50-Ω to 300-Ω differential balun was fabricated on an AD260A substrate with a conventional counterpart, and both were compared at their best-match frequencies. The proposed balun reaches a measured \( |S_{11}| \) of -17.80 dB at 2.08 GHz, with an amplitude imbalance of 0.05 dB and a phase error of 0.40°, against -10.20 dB, 0.15 dB and 7.50° at 2.10 GHz for the conventional balun. Its measured 10-dB return-loss bandwidth of 7.80 % is narrower than the model prediction, which neglects loss and discontinuities.

  • Open Access

    Cyber-Resilient Dynamic EV Hosting-Capacity Control Against False Data Injection Attacks in Smart Distribution Systems

    by Pantree Khompittaya, Kritphon Phanrattanachai, Prakasit Prabpal, Somchat Sonasang, Niwat Angkawisittpan, Thananan Chooseang, Yanapat Songthong, Decha Wongpatsa

    Dynamic hosting-capacity control lets electric vehicle (EV) charging stations use the voltage headroom of distribution feeders. Because charging power is set from voltage measurements, a false data injection attack (FDIA) that inflates the reported voltage makes the controller over-allocate power and drives the real voltage below its limit. This paper models the closed loop of measurement, estimation, control and feeder for a 0.6 MW station on the IEEE 33-bus system and proposes a physics-informed defense: a consistency test between the controller’s voltage and a reconstruction built only from protected substation measurements, the station’s revenue meter and load forecasts. For attacks confined to the field-meter set and chosen independently of the measurement noise, invariance to the attack allocation follows by design because the trusted reconstruction has zero gain from those field meters. Under the nominal 10% load pseudo-measurement uncertainty and the protected-meter noise levels used in the case study, the layered detector identifies 70–95% of stealthy 0.0075 p.u. attacks, against 34–66% for conventional residual-based detectors; over 30 simulated days, it cuts daily under-voltage time from about 180 min to 2–8 min for injections of 0.01 p.u. or more. This detection resolution degrades as forecast or protected-anchor uncertainty increases. A supervised neural classifier matches it only on attacks it was trained on and is fully evaded by a model-aware attacker. A detector-aware attacker (A4) evades the PI alarm on 96–98% of attacked steps; resilience in that case comes primarily from the adaptive safety margin rather than from detection itself, at a cost of 5.5–7.2% of EV energy. Because the trusted-reconstruction error has unbounded Gaussian tails, the resulting safety bound is probabilistic rather than absolute. Model bias must be removed by offset re-estimation; noisier protected meters reduce detection power. The reported evidence is simulation-based and should be interpreted as proof-of-concept validation pending hardware-in-the-loop and field-data evaluation.

  • Open Access

    Reliability of Machine Learning for Soil-Nutrient Estimation from Long-Term IoT Sensing: A Temporal-Validation and SHAP-Based Study Under Tropical Conditions

    by Kritphon Phanrattanachai, Decha Wongpatsa, Niwat Angkawisittpan, Somchat Sonasang

    Low-cost IoT soil sensors are attractive for precision agriculture because they can record field conditions continuously and at low cost. However, when machine learning models are trained on dense time-series data, their reported accuracy can be misleading if their temporal structure is ignored. This study examined Random Forest prediction of sensor-derived nitrogen (N), phosphorus (P), and potassium (K) readings using 63,769 observations collected over 170 days from two LoRa-based soil-sensing nodes in a tropical field site in Thailand. Soil moisture, temperature, electrical conductivity (EC), and pH were used as inputs; the outputs were the N, P, and K values reported by the composite sensor, not laboratory-confirmed nutrient concentrations. Model evaluation compared repeated random split (RS) validation with chronological split (CS) validation. RS gave high apparent performance (R2=0.890–0.901), but CS failed for all targets (R2=−0.50, −8.06, and −1.31 for N, P, and K, respectively). Jensen–Shannon divergence values of 0.070–0.256 indicated measurable train–test distributional drift. SHAP analysis further showed that EC and temperature were the main contributors on average, while humidity contributed least in most validation–target combinations. The findings suggest that random-split validation can greatly overstate the usefulness of ML models for future IoT soil-sensor data. Time-aware validation, periodic laboratory calibration, and cautious interpretation are therefore essential before sensor-derived NPK predictions are used for agronomic decisions.

  • Open Access

    Nondestructive Quality Evaluation of Nakhon Phanom Lychee (NP.1) Using Parallel Coupled Line Microstrip via Isolation Loss Parameter Analysis

    by Decha Wongpatsa, Kritphon Phanrattanachai, Siriwat Ninlawat, Phatcharanat Saeng-On, Niwat Angkawisittpan, Somchat Sonasang, Ravee Phomloungsri

    This study investigates the isolated-port response of a four-port parallel coupled-line microstrip structure as a nondestructive microwave indicator for storage-related changes in Nakhon Phanom Lychee (NP.1). Unlike conventional one-port or two-port microwave sensors that mainly rely on reflection, forward transmission, resonant-frequency shift, or quality factor, the present approach monitors the nominally isolated port, where the measured signal is affected by perturbation of the coupled-line field balance. Ten fruits were stored under refrigerated conditions and evaluated at 7-day intervals over three storage periods. Each fruit was measured ten times during each period at room temperature using a two-port vector network analyzer over 0.3–0.8 GHz. In the adopted connection, the measured VNA S21 corresponded to the sensor isolated-port parameter S31. The mean resonant frequency decreased from 487 MHz in Week 1 to 478 MHz in Weeks 2 and 3. The S31 magnitude at the resonant dip became progressively less negative, changing from −59.59 dB to −53.95 dB and −48.27 dB, corresponding to a total change of 11.32 dB. Although the resonant frequency did not distinguish Weeks 2 and 3, the isolation magnitude remained different between these periods. The measured response is consistent with storage-related changes in the effective electromagnetic properties of the fruit. These preliminary results support the feasibility of using isolated-port magnitude as an additional microwave feature for nondestructive lychee monitoring. Further validation using direct physicochemical, dielectric, and metrological measurements is required before quantitative quality assessment can be established.

  • Open Access

    Interpretable Machine Learning for PM2.5 Forecasting Using a Long-Term IoT Environmental Dataset in Tropical Thailand

    by Kritphon Phanrattanachai, Worachai Srisamoodkham, Niwat Angkawisittpan, Somchat Sonasang

    A field record from the SCG/Wheather001 IoT station in Thailand was used to test short-gap PM2.5 prediction. The file contained 191,489 observations collected from 10 August 2024 to 25 May 2026. After removing non-physical sensor readings and keeping only consecutive records with acceptable time gaps, 182,242 records remained for modelling. Random Forest, XGBoost, and LightGBM were compared using a chronological 80:20 train–test split. XGBoost gave the lowest test error in this station record, with R2=0.824, RMSE = 6.031 μg m−3, and MAE = 3.845 μg m−3. SHAP analysis showed that the next PM2.5 value was driven mainly by the current PM2.5 reading. PM10, the second PM2.5 channel, the second PM10 channel, and PM1 also contributed to the prediction. Weather and time variables had smaller effects for this short-horizon task, but they were still useful for checking whether the model response was physically reasonable. For this station file, the screened IoT data were sufficient for a short-gap PM2.5 nowcast. The result should be read as a local forecasting test rather than as a general model for all monitoring sites. SHAP helped check whether the forecast was driven mainly by recent particle loading, which is useful before such output is shown on a local dashboard.

  • Open Access

    Cyber-Resilient Dynamic EV Hosting Capacity Control Against False Data Injection Attacks in Smart Distribution Systems

    by Pantree Khompittaya, Kritphon Phanrattanachai, Prakasit Prabpal, Somchat Sonasang, Niwat Angkawisittpan, Thananan Chooseang, Yanapat Songthong, Decha Wongpatsa

    Dynamic hosting capacity control lets electric vehicle (EV) charging stations use the voltage headroom of distribution feeders. Because charging power is set from voltage measurements, a false data injection attack (FDIA) that inflates the reported voltage makes the controller over-allocate power and drives the real voltage below its limit. This paper models the closed loop of measurement, estimation, control and feeder for a 0.6 MW station on the IEEE 33-bus system and proposes a physics-informed defense: a consistency test between the controller’s voltage and a reconstruction built only from protected substation measurements, the station’s revenue meter and load forecasts. For attacks independent of the measurement noise, its detection probability depends only on the injected shift. With the chi-square bad-data detector at a 1% false-alarm rate, it detects 70–95% of stealthy 0.0075 p.u. attacks, against 34–66% for residual-based detectors, and over 30 simulated days cuts daily under-voltage time from about 180 min to 2–8 min for injections of 0.01 p.u. or more. A supervised neural classifier matches it only on attacks it was trained on and is fully evaded by a model-aware attacker. An adaptive safety margin counters detector-aware attackers for 5.5–7.2% of EV energy. Model bias must be removed by offset re-estimation; noisier protected meters reduce detection power.

  • Open Access

    Ransomware and artificial intelligence: a comprehensive systematic review of reviews

    by Therdpong Daengsi, Phisit Pornpongtechavanich, Paradorn Boonpoor, Kathawut Wattanachukul, Korn Puangnak, Kritphon Phanrattanachai, Pongpisit Wuttidittachotti, Paramate Horkaew

    This study provides a comprehensive synthesis of artificial intelligence (AI), especially machine learning (ML) and deep learning (DL) in ransomware defense. Using a “review of reviews” methodology based on the PRISMA, this paper gathers insights on how AI is transforming ransomware detection, prevention, and mitigation strategies in the past five years (2020-2024). The findings highlight the effectiveness of hybrid models, which combine multiple analysis techniques such as code inspection (static analysis) and behavior monitoring during execution (dynamic analysis). The study also explores anomaly detection and early warning mechanisms before encryption that tackle ransomware’s growing complexity. It also examines key challenges in ransomware defense, such as techniques designed to deceive AI driven detection, and the lack of strong and diverse datasets. It highlights AI’s role in early detection and real-time response systems, enhancing scalability and resilience. With the systematic review of reviews approach, the contributions of this study are systematically consolidating research insights from multiple review articles, identifying effective AI models, and bridging theory with practice to foster collaboration among academia, industry, and policymakers. Future research directions are anticipated and practical recommendations for cybersecurity practitioners are provided. Finally, it presents a roadmap for advancing AI-driven countermeasures, for the protection of key systems and infrastructures against evolving ransomware threats.


  • Open Access

    Analyzing 5G performance: investigating altitude-induced variations

    by Therdpong Daengsi, Pakkasit Sriamorntrakul, Surachai Chatchalermpun, Kritphon Phanrattanachai

    Since the launch of fifth generation (5G) services in Thailand in 2020, there have been continuous improvements in 5G coverage. Currently, 5G coverage extends to most areas throughout the country. However, coverage issues persist not only in rural areas but also in high-rise buildings in urban areas. Consequently, a study was conducted within such buildings. This paper assesses the performance of 5G at different altitude test points. The chosen location for the field tests was a high-rise building within a crowded public hospital, which receives numerous patients every weekday, in the major urban area of Bangkok. Two smartphones from the same manufacturer, both supporting 5G technology and equipped with the Speedtest application, were employed as tools for this study. Tests were carried out on the third and twenty-fourth floors of the high-rise building for data collection. The primary finding of this study reveals that download speeds exhibited a significant decrease with increasing altitude of the test points, as evidenced by statistical analysis (p-values0.001). This implies an issue with altitude-induced variations, indicating a need for the improvement of indoor 5G coverage in high-rise buildings.


  • Open Access

    Comparison of 5G performance post-merger between two network operators using field tests in urban areas

    by Surachai Chatchalermpun, Therdpong Daengsi, Pakkasit Sriamorntrakul, Kritphon Phanrattanachai

    In late Q1/2023, DTAC and TRUE officially completed their merger. Consequently, this study was initiated to ascertain whether their respective fifth-generation (5G) networks had been seamlessly integrated several months following the merger. The investigation involved conducting drive tests along two predefined routes within the urban areas of Bangkok, employing the G-NetTrack pro tool for testing and data collection. Additionally, stationary tests were conducted in two crowded places using an application called Speedtest. Subsequently, an array of quality of service (QoS) metrics, including reference signal received power (RSRP), reference signal received quality (RSRQ), signal to noise ratio (SNR), download (DL), upload (UL) speeds and latency, were meticulously analyzed and presented. The findings of this study unveiled that, despite the successful completion of the DTAC and TRUE merger from a business standpoint, the technical integration of their respective 5G networks had not been finalized, although there were no significant differences between DTAC and TRUE for DL (p-value=0.542) and UL (p-value=0.090). Notably, significant differences were found between DTAC and TRUE for four metrics, including RSRP, RSRQ, SNR and latency (p-values0.05). Remarkably, roaming functionalities were still operational between the two networks.


  • Open Access

    Analyzing 5G Performance: Investigating Altitude-Induced Variations

    by Therdpong Daengsi, Pakkasit Sriamorntrakul, Surachai Chatchalermpun, Kritphon Phanrattanachai

    Since the launch of 5G services in Thailand in 2020, there have been continuous improvements in 5G coverage. Currently, 5G coverage extends to most areas throughout the country. However, coverage issues persist not only in rural areas but also in high-rise buildings in urban areas. Consequently, a study was conducted within such buildings. This paper assesses the performance of 5G at different altitude test points. The chosen location for the field tests was a high-rise building within a crowded public hospital, which receives numerous patients every weekday, in the major urban area of Bangkok. Two smartphones from the same manufacturer, both supporting 5G technology and equipped with the Speedtest application, were employed as tools for this study. Tests were carried out on the third and twenty-fourth floors of the high-rise building for data collection. The primary finding of this study reveals that download speeds exhibited a significant decrease with increasing altitude of the test points, as evidenced by statistical analysis (pvalues < 0.001). This implies an issue with altitude-induced variations, indicating a need for the improvement of indoor 5G coverage in high-rise buildings.