Accurately estimating the actual capacity of the battery is crucial for stable battery operation and user safety. This paper combined feature extraction through differential thermal voltammetry analysis and long short-term memory for accurate state of health estimation. First, differential thermal voltammetry curves according to battery degradation were extracted for various cathode materials. Then, health indicators are collected from the differential thermal voltammetry curve for deep learning-based state of health estimation. In particular, to improve the state of health estimation performance, the integral value of the differential thermal voltammetry curve through the specific voltage ranges was additionally introduced along with the peak and valley. Second, the correlation between the extracted health indicators and capacity was analyzed using Pearson correlation analysis. Finally, a framework was developed to estimate the state of health of the battery using high-quality health indicators as inputs to the long short-term memory model. The state of health estimation performance of the proposed algorithm, which reflected the integral value of the differential thermal voltammetry curve, was compared with the case that did not reflect. From the result, the mean absolute error decreased by 11.6% and the root mean square error by 10.01% for the two battery data sets.

Fly-buck converter is a multi-output converter with the structure of a synchronous buck converter structure on the primary side and a flyback converter structure on the secondary side, and can be utilized in various applications due to its many advantages. In terms of control, the primary side of the fly-buck converter has the same structure as a synchronous buck converter, allowing the constant-on-time (COT) control to be applied to the fly-buck converter. However, due to the inherent energy transfer principle, the primary-side output voltage regulation of COT controlled fly-buck converters may be poor, which can deteriorate the overall converter performance. Therefore, the primary output capacitor must be carefully designed to improve the voltage regulation characteristics. In this paper, a theoretical analysis of the output voltage regulation in COT controlled fly-buck converter is conducted, and based on this, a design guideline for the primary output capacitor considering the output voltage regulation is presented. The validity of the analysis and design guidelines was verified using a 5 W prototype of the COT controlled fly-buck converter for telecommunication auxiliary power supply.

Analysis, Design, and Implementation of a Soft-Switched Active-Clamped Forward Converter with a Current-Doubler Rectifier Active-clamp circuit;Current-doubler rectifier;Forward converter;Zero-voltage switching; This study examines the zero-voltage switching (ZVS) operation of an active-clamped forward converter (ACFC) with a current-doubler rectifier (CDR). The ZVS condition can be obtained with a much smaller leakage inductance compared to that of a conventional ACFC. Due to the significantly reduced leakage inductance, the design is optimized and the circulating loss is reduced. The operation of the ACFC with a CDR is analyzed, and a detailed ZVS analysis is conducted on the basis of a steady-state analysis. From the results, a design consideration for ZVS improvement is presented. Loss analyses of the converters shows that enhanced soft-switching contributes to an efficiency improvement under light-load condition. Experimental results from a 100-W (5-V/20-A) prototype verify that the ACFC with a CDR can attain ZVS across an extended load range of loads and achieve a higher efficiency than conventional ACFCs.

This paper proposes a new current balancing method for alternating current (AC) - light emitting diodes (LED). The proposed method uses the bi-directional current mirror circuit with negative feedback that consists of two reference currents, n+1 NPN transistors, n+1 PNP transistors, and 2(n+1) resistors for n pairs of anti-parallel LED strings. It can balance the LED currents in multiple AC-LED strings regardless of the polarity of the AC source while reducing the differences in LED currents caused by the early effect. In addition, the proposed circuit can achieve LED current balancing even in the variations of parameters among the transistors due to the negative feedback operation. Simulations and experiments were performed to verify the effectiveness of the proposed circuit. The balancing operation of the proposed circuit was simulated at low (50/60 Hz) and high (60 kHz) frequencies. In addition, open/short protection, power factor, and efficiency of an LED driver using the proposed circuit were analyzed and compared with a basic mirror circuit. For the experiments, a prototype was built and tested for 5W LED lighting consisting of three pairs of anti-parallel LED strings. The maximum LED current difference $\Delta i_{LED} $ was reduced from 45 mA to 4 mA by the balancing operation of the proposed circuit.

While deep learning‐based approaches for state of health (SOH) estimation in lithium‐ion batteries have been actively studied, most models face deployment constraints in on‐device applications due to their high complexity and large number of parameters. Although previous studies have introduced knowledge distillation (KD) for model compression, single‐teacher architectures exhibit limited performance improvement due to insufficient knowledge diversity. To resolve this issue, this study proposes a multi‐teacher knowledge distillation (MTKD) framework to simultaneously achieve efficient SOH estimation and model compression. From raw charging data, a total of 18 health indicators (HIs) were obtained from diverse perspectives, including temporal information, statistical features, equivalent circuit model (ECM) parameters, and incremental calculation. Key features were selected through Pearson correlation analysis and the maximal information coefficient (MIC), and were utilized as inputs for the deep learning models. Subsequently, large‐scale teacher models based on deep neural network (DNN), long short‐term memory (LSTM), and one‐dimensional convolution neural network (1D CNN) architectures were trained to capture various degradation characteristics, including nonlinear relationships, temporal dependencies, and local patterns. The lightweight student model was then trained using soft targets obtained from the teacher models along with ground truth labels. Experimental results demonstrate that the student model trained with the proposed MTKD achieved a 45.98% reduction in root mean square error (RMSE) and a 15.72% improvement in coefficient of determination ( R ² ) compared to single‐teacher KD (STKD). This study successfully extends KD research beyond traditional computer vision and image processing domains, demonstrating practical applicability in battery data‐driven applications.