Abstract：To address the complex dynamic security threats confronting cloud-native education platforms and the limitations of traditional defense mechanisms—notably weak environmental awareness, lack of intelligent decision-making, inadequate generalization capabilities, and privacy-compliance conflicts—this paper proposes a dynamic security defense system based on a mixture-of-experts model. The system implements a four-layer collaborative protection framework:a spatio-temporal graph neural network models traffic spatio-temporal features;a multimodal convolutional neural network fuses containerized heterogeneous data;a federated learning component facilitates privacy-preserving knowledge sharing;a large language model generates executable defense strategies.Core innovations include:designing a differentiable gating network (utilizing Top-2 sparse activation) to dynamically route attack features to optimal expert models;establishing a threat intensity index-driven elastic resource scheduling mechanism for Kubernetes. This research provides an elastically scalable security paradigm for cloud-native education platforms. The methodology offers both theoretical value and practical significance for building adaptive security architectures, laying the groundwork for the development of intelligent proactive defense systems.

Abstract：Brain-computer interfaces (BCIs) provide users with a control channel that does not rely on peripheral nerves or muscles, making them widely used in the rehabilitation of patients with neuromuscular diseases. Motor imagery (MI)-based BCIs use brain signals generated during the imagination of motor tasks to enable control without physical movement. To improve the classification performance of MI-BCIs, this study proposes a method that combines Pearson correlation coefficient-based channel selection (PCCS) and filter bank regularized common spatial pattern (FBRCSP). First, PCCS is used to select relevant channels from EEG signals, retaining key information and removing redundancy. Then, FBRCSP is employed to optimize spatial patterns and extract discriminative features. The proposed method is validated on the BCI Competition III IVa and IV Dataset I datasets. Experimental results demonstrate that the PCCS-FBRCSP method outperforms traditional approaches in terms of classification accuracy and stability, particularly in cross-individual experiments, exhibiting stronger robustness. Additionally, the method effectively reduces computational complexity and holds great potential for practical applications.

Abstract：Reinforcement learning evaluation metrics, serving as core tools for measuring the performance of agents and guiding algorithm optimization, face key challenges such as the singularity of metrics, environmental dependence, and the lack of interpretability in practical applications. This paper systematically analyzes the classification framework of existing evaluation metrics, proposes a multi-dimensional metric system based on performance, learning process, strategy, robustness, and efficiency, and explores its applicability and limitations in different task scenarios (such as sparse reward and high-dimensional state space). The study indicates that traditional metrics are prone to overlooking the requirements of safety, efficiency, and alignment with human preferences in complex environments, and there is a need to design evaluation methods that integrate multiple objectives in combination with the characteristics of tasks. For future research, this paper suggests focusing on directions such as multi-objective Pareto optimization, reward modeling based on human feedback, and the quantification of exploration efficiency in sparse reward environments, so as to enhance the comprehensiveness and interpretability of evaluations. By combining theoretical analysis with practical cases, this paper provides methodological support for the standardization of the reinforcement learning evaluation system and its adaptation across different fields, thus promoting its efficient implementation in complex scenarios.

Abstract：In order to solve the problem of serious electromagnetic interference of inductive power supply chip and large output ripple of capacitive power supply chip, a capacitive power supply chip is designed. The output ripple voltage is lower than 1 mV under high frequency condition with a non-inductance structure, which is close to the inductance power chip. The output voltage of the chip can be adjusted in the range of -16 V to -3 V, and the operating frequency can be changed between 90 kHz and 1 MHz. The circuit is designed and implemented based on 250 nm process. The simulation results show that under the conditions of input voltage of 16 V, output current of 100 mA and operating frequency of 1 MHz, the output ripple voltage is 0.6 mV and output voltage is -14.84 V, and the efficiency reaches 87.1%. When the operating frequency is reduced to 90 kHz, the output ripple voltage is 10 mV, the output voltage is -14.83 V, and the efficiency is increased to 91.9%. The experimental data show that the chip has good adaptability and can meet the needs of various application scenarios.

Abstract：Aiming at the problem of potential overload and short-circuit issues in low-dropout linear regulators(LDO), the article designed a foldback current limit circuit in a 180 nm BCD process. By incorporating a reference establishment signal, the proposed design effectively prevents latch-up issues commonly encountered in conventional foldback current limit circuits. Simulation results demonstrate that when reaching current limit thresholds, the output current reduces to 200 mA and the output voltage decreases by approximately 400 mV before initiating foldback operation, with both LDO output current and voltage subsequently decreasing. The findings indicate that this circuit successfully achieves current limit protection while significantly reducing system power consumption under overcurrent conditions.

Abstract：System on Chip (SoC) chips will perform SM2 signature and verification on product firmware to confirm the legitimacy of the product firmware. However, in the existing technologies, product firmware has not conducted identity authentication on the chips. Due to the inability to determine whether the chips are legitimate, the security of the product firmware and data cannot be protected. Based on the SM2 algorithm, this paper implements a bidirectional identity authentication technology for chip identification of product firmware and firmware identification of chips in the SoC security chip import process. The scheme is highly secure and highly implementable and has universal practical reference significance for the SoC security chip to conduct bidirectional identity authentication of boot programs and product firmware. The SM2 module designed in this paper for embedded SoC security chips has the advantages of small chip area, low circuit design difficulty, simple CPU operation, and suitability for embedded SoC security chips.

Abstract：As a core power management component in modern electronic systems, the startup inrush current of Low Dropout linear regulators (LDO) poses significant threats to system reliability. This paper focuses on the MSK5101 LDO, addressing the input inrush current spike phenomenon observed after implementing an RC delay circuit at the enable terminal. A suppression method based on dynamic optimization of feedback networks is proposed. Through establishing a small-signal model and conducting frequency-domain analysis, the influence patterns of feedback network parameters on system response speed and stability are revealed. The proposed approach achieves equivalent soft-start functionality by increasing feedback resistance and adjusting compensation capacitance. Experimental results demonstrate that the optimized configuration reduces inrush current peak from 2.7 A to 0.87 A (a reduction of 67.8%), while extending output voltage settling time from 2.9 ms to 12.3 ms, significantly enhancing the stability and reliability of aerospace power systems.

Abstract：A pre-variable gain amplifier for high precision temperature measuring system is designed in 0.18 μm process, compared with the traditional preamplifier, and the influence of the feedback network leakage on the front-end sampling circuit is avoided by using the opening loop structure. Through the current source of the power voltage, the current is provided for the cross-guide amplifier circuit, which produces a transconductance that does not contain the process parameters, and then the gain unrelated to voltage supply, temperature, and process parameters is obtained by the operation amplifier. By changing the load of the whole difference cross-amplifier amplifier, 10 and 40 times variable gain are implemented. The simulation results show that when the amplification gain is 10 times, the maximum gain variation with temperature is 0.19 dB, the maximum gain variation with voltage supply is 0.15 dB, the maximum gain variation with process is 0.12 dB. When the amplification gain is 40 times, the maximum gain variation with temperature is 0.16 dB, the maximum gain variation with voltage supply is 0.23 dB, the maximum gain variation with process is 0.13 dB.

Abstract：With the continuous evolution of cyberattack technologies, Advanced Persistent Threats (APT) targeting enterprises and organizations have become increasingly prevalent. The success of APT attacks largely depends on the execution quality during the post-exploitation phase, where attackers use sophisticated adversarial techniques to maintain persistent control and exfiltrate data. This paper focuses on adversarial techniques in the post-exploitation phase, discussing technologies such as Bootkit, COM Interface Abuse, BYOVD, and VEIL7, and introduces a new adversarial framework. Through experiments, the paper demonstrates its bypass capabilities against multiple antivirus tools, as well as its advantages over existing adversarial tools. Additionally, corresponding defense strategies for this framework are proposed to advance research on countermeasure techniques and optimize defense mechanisms.

Abstract：With the development of the information battlefield, the application of Link16 data link in satellites is becoming an irreversible trend. Space-ground fusion is an important direction to achieve more efficient and real-time information exchange, but in practical applications, the challenges brought by Doppler effect and signal delay cannot be ignored. This paper proposes a location-equivalent Link16 data link terminal low-orbit satellite access design scheme. By calculating the delay value and Doppler value at the delay center and carrying out signal pre-compensation, the Doppler caused by signal transmission between the satellite and the earth is eliminated and the time reference of satellite-borne Link16 is aligned. The Link16 data link terminal low-orbit satellite access design is realized, and the simulation shows that the scheme is feasible and can support the application of data link earth and earth integration.

Abstract：The energy consumption of IT equipment directly affects the power consumption of industrial data centers, and predicting the energy consumption of IT equipment is of great significance for optimizing energy management and resource planning. However, due to the non-linear and non-stationary nature of IT energy consumption data, the prediction accuracy is low. In this regard, by combining the advantages of convolutional neural network (CNN), bidirectional long short-term memory network (BiLSTM), and attention mechanism, local features of IT equipment energy consumption and deep key information in the data are extracted separately. Based on the self tested IT equipment energy consumption dataset, an energy consumption prediction model based on CNN-BiLSTM-Attention is constructed. The R2, MAE, and RMSE of this model are 0.905 3, 0.050 4, and 0.067 3, respectively. Compared with existing LSTM, BiLSTM and CNN-BiLSTM models, this model has improved to varying degrees, indicating that this model can be applied to accurate prediction of IT equipment energy consumption in industrial data centers.

Abstract：The ZUC algorithm is a commercial sequence cipher algorithm independently developed in China, which has been applied in complex scenarios such as real-time server computation and big data processing. The high-speed implementation of ZUC has important practical significance for its application promotion. Based on this, the ZUC algorithm was implemented on the FPGA hardware platform to meet the high-performance requirements of the ZUC applicable environment. By optimizing core operations such as modular multiplication and modular addition, and adopting a streamlined structure design, the ZUC algorithm was realized. The experimental results show that the data throughput of the ZUC algorithm core can reach 10.4 Gb/s. Compared with existing research results, it reduces the delay of critical paths, improves the operating frequency of the algorithm, and achieves a good balance between throughput and hardware resource consumption, providing a new solution for the high-performance implementation of the ZUC algorithm.

Abstract：Network boundary recognition is an important step in network topology measurement. Currently, regional network boundary recognition technology is mainly based on sampling detection of regional networks and boundary screening using IP address based geolocation technology. This method has problems such as incomplete network topology acquisition and limited boundary verification methods. In response to this issue, this article proposes a high coverage detection strategy for regional network boundary recognition and a regional boundary IP recognition technology based on multi-source information fusion. By designing the deployment location of detection points and target coverage strategy to measure the topological boundaries of the region, the probability of discovering boundaries is improved. Multiple sources of information such as WHOIS, latency, and traceroute are obtained to supplement the source of boundary verification data. A machine learning based verification model is established to improve the credibility and coverage of regional network boundary recognition. This article applies this method to the network boundary recognition and verification work in region A, and verifies the effectiveness of this method.

Abstract：The integration and debugging of the radio-frequency system of synchrotron radiation accelerator cannot be separated from the radio-frequency cavity, but its cost is expensive, the processing cycle is long, and there is a risk of damaging the cavity when using the actual radio-frequency cavity to test the low-level radio-frequency controller (LLRF); In addition, the center frequency of the actual radio-frequency cavity is fixed, and only LLRF controllers operating at the corresponding frequency can be tested. To solve the above problems, this paper designs a simulated radio-frequency cavity device whose center frequency can follow the low level and can also mix white noise, used to test the performance of the low-level controller. The text elaborates in detail on the design concept and implementation method of the simulated radio-frequency cavity, and by linking and testing the designed simulated cavity with the LLRF, it proves that the simulated cavity can test the LLRF operating at different frequencies, and test its amplitude and phase stability as well as its closed-loop control capability. The simulated cavity designed in this paper has great practical significance for the integration and commissioning of the radio-frequency system, which can greatly improve the testing efficiency of the low-level controller, and at the same time effectively avoid the risk of damaging the actual radio-frequency cavity for testing the LLRF.

Abstract：The up-down converter module is the key component in microwave circuits, and its performance directly affects the overall communication system. In this paper, design of an X-band up-down converter module based on image rejection converter is presented. By analyzing and calculating the main indicators of the entire machine, the down-conversion adopts a one-time frequency conversion scheme, and the up-conversion adopts a twice-time frequency conversion scheme. By using a fixed local frequency for multiple frequency conversions while meeting the requirements of the image-rejection, the up-conversion has a high RF/LO isolation. The test results show that the image-rejection of the down-conversion is 29 dBc and the RF/LO isolation of the up-conversion is 71 dBc. The module indicators are in good compliance and meet the usage requirement, which has reference value for the design of similar miniaturized up-down converter.

Abstract：With the development of a globalized business environment, the widespread adoption of remote work, and potential inconveniences caused by unexpected events, the demand for online meetings has been increasing. To address this need, this paper designs and implements a multi-functional, cascaded omnidirectional conference microphone. The device integrates a System-on-Chip (SoC), touch sensor, and MEMS microphone, combined with Power over Ethernet (PoE) technology, to form a hardware platform that supports audio acquisition, data transmission, and multi-device cascading. In terms of algorithms, the system incorporates echo cancellation and reverberation suppression techniques to enhance audio quality. Experimental results show that a single device can effectively capture sound within a 7-meter radius, supporting up to four cascaded devices for stable operation in large spaces. Additionally, the speaker's frequency response remains within ±60 dB, while the echo cancellation efficiency exceeds 98% and the reverberation suppression ratio reaches over 94%, meeting the design requirements for conference microphones.

Abstract：This study addresses the need for efficient detection of physiological parameters of Betula luminifera seedling leaf in forestry field applications, by developing a handheld device based on near-infrared spectroscopy. The device integrates a micro near-infrared spectral sensor, an Android smartphone, and a cloud server, enabling data acquisition and analysis via Bluetooth and mobile networks. A prediction model for leaf water content and chlorophyll content was constructed using Partial Least Squares Regression (PLSR), and Multivariate Scatter Correction (MSC) was applied to improve the stability of the data. Validation results show that the water content model has an R2 of 0.923 and an RMSE of 0.044, while the chlorophyll model has an R2 of 0.824 and an RMSE of 2.569, indicating high prediction accuracy and reliability. This study provides a convenient, non-destructive detection method with broad application prospects.

Abstract：Current RF power supplies widely used in plasma magnetron sputtering coating systems suffer from insufficient power control precision, limited adjustable power ranges, and significant output fluctuations caused by inaccurate power calculations. To address these challenges, this study proposes a radio frequency (RF) power calculation and control system for plasma coating applications, utilizing a digital down-conversion (DDC)-based RF signal generator. By directly calculating the output power through voltage-current (VI) sensor-coupled transmission line data and dynamically adjusting the signal amplitude via a controller, the system achieves precise power regulation. Simulation results demonstrate rapid and accurate power computation with high-precision control. Prototype testing confirms that the proposed method significantly reduces power output errors and enhances coating film quality on substrates, thereby providing new insights for advancing plasma magnetron sputtering technology.

Abstract：Providing accurate flow information in industrial production plays an important role in improving system performance. The thermal flow sensor based on MEMS technology reflects the flow change in the pipeline by collecting the temperature difference change, which has the characteristics of small volume, high precision and sensitive micro-flow section. In this paper, the thermal distribution model of thermal flow sensor is introduced, and the coupling simulation of electric field, flow field and thermal field of flow sensor in pipeline is carried out by COMSOL Multiphysics software to determine the key parameters of sensor chip. The flow sensor chip is manufactured by MEMS processing technology, and a calibration test platform is built to package and test the sensor chip. The measuring range of the sensor reaches 300 SLM, and the accuracy level of digital output reaches 1.0 and that of analog output reaches 1.5.

Abstract：Tomographic Synthetic Aperture Radar (TomoSAR) is an advanced 3D remote sensing technology capable of enhancing imaging quality through the incorporation of prior models. Existing methods typically use explicit regularization constraints in the reconstruction process. However, non-analytic priors face challenges in TomoSAR imaging. This paper combines Fast Iterative Shrinkage-Thresholding Algorithm (FISTA) with Plug-and-Play (PnP) framework, embedding a BM4D-based non-analytic prior during iterations. Our proposed method enables feature enhancement in TomoSAR imaging without designing explicit regularization operators. Experimental results show that the proposed framework offers fast convergence. The BM4D prior introduced via this framework achieves excellent performance in TomoSAR imaging of targets with complex structure features. Our proposed framework provides a universal solution for the application of non-analytic prior models in tomographic imaging.

Abstract：In the actual indoor positioning business scenarios, there exist the problems of uneven distribution of positioning errors and excessive local positioning errors. To address these issues, a system framework for the evaluation of indoor positioning error distribution and local positioning repair has been designed through research. This framework employs the methods of spatial triangulation and contour interpolation to conduct a quantitative evaluation and visual display of the distribution of positioning errors. It repairs the local positioning accuracy of the areas with excessive errors by correcting the local signal fingerprint maps. Through experimental tests, the framework has successfully achieved the quantitative monitoring and visualization of the spatial distribution of indoor positioning errors. The local positioning accuracy has been effectively improved through local positioning repair, verifying the effectiveness of the proposed framework. It provides a feasible solution for the fine-grained local error monitoring and repair in indoor positioning services.