Science China Information Sciences最新文献

This work presents an adaptive clock optimization scheme for TX to alleviate the timing constraints for the retimer. Using the PR and inverse-PR-based phase detector, the optimal clock phase is selected for retiming with only 8 UI convergence time. By adopting the proposed technique, we realize a 1–56 Gb/s DAC-DSP-based TX in 28-nm CMOS. Measurement results show that the rising edge of retiming clock is located in the center of data when the phase adjustment completed. The total TX consumes 164 mWat 56-Gb/s PAM4 signaling with 97.8% RLM in 0.25 mm² area. Therefore, the proposed retiming clock optimization scheme is a promising scheme for high-speed TX.

A 10-GS/s 8-bit 2× time interleaved hybrid ADC with λ/4 reference T-Line sharing technique has been demonstrated. Meanwhile, an on-chip 5th HILO with a center frequency of 20 GHz is embedded to generate the low-jitter propagation signal for T-Lines. The proposed sharing reference T-Line technique and serpentine routing enable a significant improvement on silicon-area efficiency. The ADC achieves a measured 3-dB effective bandwidth close to 6.7 GHz and an ENOB exceeding 5.5. The hybrid ADC chip fabricated in 0.9-V 28-nm CMOS achieves a 10-GS/s sampling frequency with a 31.22-mW power consumption. At Nyquist input, the SNDR is 38.11 dB, while at over-Nyquist 7.89 GHz input, the SNDR is 34.2 dB.

CSP is one of the most widely used signal processing approaches in EEG-based MI classification; however, the CSP optimization objective is not completely consistent with the final classification objective, and hence it does not necessarily lead to the best classification performance. This study has proposed a retraining framework, which retrains a neural network with the same forward computational process and initial parameters as the CSP-based traditional model, and further optimizes it on the labeled training data using gradient descent. Experiments on four MI datasets demonstrated that retraining improved traditional models’ classification performance and outperformed several popular deep neural network models, especially when the amount of labeled training data was very small. Our work demonstrates the advantage of integrating knowledge from traditional models and from the training data in EEG-based BCIs.

Tuning ferroelectricity of Hf_0.5Zr_0.5O₂ is crucial for facilitating its practical applications in various fields, including in-memory and neuromorphic computing. Previous studies have revealed that the electrodes have a significant influence on ferroelectricity, and changing electrode materials can realize different but discrete ferroelectric polarization values. Here, we introduce an alloy-electrode method, in order to achieve gradual and accurate modulation of ferroelectric polarization, especially useful for matching the polarization charges at the interface of ferroelectric insulators and ferroelectric semiconductors. Au and W electrodes are chosen as baselines for realizing weak and strong ferroelectric polarization, where the intermediate states can be achieved by adjusting the ratio of metals in the Au-W alloy. To demonstrate the generality of this approach, the Cu-W alloy electrode is also realized for tuning ferroelectric polarization. The effect of alloy electrodes on device leakage current, endurance, and retention is evaluated. In addition, the temperature stability of ferroelectric capacitors is tested, where limited changes in both remnant polarization and coercive voltages are observed, showing the great potential of the ferroelectric hafnium oxide. Such gradual modulation of ferroelectric polarization could facilitate the application of Hf_0.5Zr_0.5O₂ in in-memory and neuromorphic computing.

This study presents a column-shared hTDC with pixel-to-pixel coincidence detection and compact analog pulse counters. The column-shared architecture allows the hTDC to have no large area consumption caused by ADC and memories in pixels. Thanks to the pixel-to-pixel coincidence detection, only one SPAD is needed in each pixel. The application of the analog counter greatly reduces the area occupied by the counter. The simulation results show that the proposed hTDC can effectively reduce the pixel area while ensuring accuracy under high background light conditions, so it is suitable for outdoor applications.

In this paper, an interval estimation scheme is developed for delayed switched positive systems (DSPS) with mode-dependent average dwell time switching. A lossless zonotopic estimation approach is proposed for the delayed intersection zonotope with the positive generator matrix. First, considering the existence of asynchronism between the system mode and the correction matrix mode, the mismatched intersection zonotope is constructed for DSPS to verify the consistency between the system model and outputs. Then, by utilizing the introduced radius definitions, the ℓ_∞ performance is addressed to optimize the size of delayed intersection zonotopes. Subsequently, we present a joint-design approach of switching signals and the mode-dependent correction matrix by constructing positive generator matrix-based delayed radius functions. Furthermore, guaranteed nonnegative state bounds are derived for the considered DSPS based on the proposed lossless zonotopic estimation criteria. Finally, detailed simulations are conducted to validate the feasibility and superiority of the developed methods.

With the rapid development of network technology and the automation process for 5G, cyber-attacks have become increasingly complex and threatening. In response to these threats, researchers have developed various network intrusion detection systems (NIDS) to monitor network traffic. However, the incessant emergence of new attack techniques and the lack of system interpretability pose challenges to improving the detection performance of NIDS. To address these issues, this paper proposes a hybrid explainable neural network-based framework that improves both the interpretability of our model and the performance in detecting new attacks through the innovative application of the explainable artificial intelligence (XAI) method. We effectively introduce the Shapley additive explanations (SHAP) method to explain a light gradient boosting machine (LightGBM) model. Additionally, we propose an autoencoder long-term short-term memory (AE-LSTM) network to reconstruct SHAP values previously generated. Furthermore, we define a threshold based on reconstruction errors observed during the training phase. Any network flow that surpasses the specified threshold is classified as an attack flow. This approach enhances the framework’s ability to accurately identify attacks. We achieve an accuracy of 92.65%, a recall of 95.26%, a precision of 92.57%, and an F1-score of 93.90% on the dataset NSL-KDD. Experimental results demonstrate that our approach generates detection performance on par with state-of-the-art methods.

Artificial intelligence technology is widely used in the field of wireless sensor networks (WSN). Due to its inexplicability, the interference factors in the process of WSN object localization cannot be effectively eliminated. In this paper, an explainable-AI-based two-stage solution is proposed for WSN object localization. In this solution, mobile transceivers are used to enlarge the positioning range and eliminate the blind area for object localization. The motion parameters of transceivers are considered to be unavailable, and the localization problem is highly nonlinear with respect to the unknown parameters. To address this, an explainable AI model is proposed to solve the localization problem. Since the relationship among the variables is difficult to fully include in the first-stage traditional model, we develop a two-stage explainable AI solution for this localization problem. The two-stage solution is actually a comprehensive consideration of the relationship between variables. The solution can continue to use the constraints unused in the first-stage during the second-stage, thereby improving the performance of the solution. Therefore, the two-stage solution has stronger robustness compared to the closed-form solution. Experimental results show that the performance of both the two-stage solution and the traditional solution will be affected by numerical changes in unknown parameters. However, the two-stage solution performs better than the traditional solution, especially with a small number of mobile transceivers and sensors or in the presence of high noise. Furthermore, we have also verified the feasibility of the proposed explainable-AI-based two-stage solution.

Game theory studies the mathematical models for self-interested individuals. Nash equilibrium is arguably the most central solution in game theory. While finding the Nash equilibrium in general is known as polynomial parity arguments on directed graphs (PPAD)-complete, learning in games provides an alternative to approximate Nash equilibrium, which iteratively updates the player’s strategy through interactions with other players. Rules and models have been developed for learning in games, such as fictitious play and no-regret learning. Particularly, with recent advances in online learning and deep reinforcement learning, techniques from these fields greatly boost the breakthroughs in learning in games from theory to application. As a result, we have witnessed many superhuman game AI systems. The techniques used in these systems evolve from conventional search and learning to purely reinforcement learning (RL)-style learning methods, gradually getting rid of the domain knowledge. In this article, we systematically review the above techniques, discuss the trend of basic learning rules towards a unified framework, and recap applications in large games. Finally, we discuss some future directions and make the prospect of future game AI systems. We hope this article will give some insights into designing novel approaches.

This study explores a new navigation method using multi-path solar panel-reflected solar oscillations. Considering the solar panels of BeiDou-3 M1–M24 and GPS satellites as examples, the simulations show that the mean position error of FY-1 using solar panel-reflected solar oscillations is only 20.61 m in 30 days. Compared with the existing autonomous navigation methods for the Earth satellites, the newly proposed method has two advantages. (1) It has the highest navigation accuracy. (2) It does not require any additional accurate geomagnetic map, gravity gradient map, or refraction model. While the proposed method requires at least two atomic frequency discriminators to obtain the measurements and its accuracy is affected by the geometric relationship between the Earth satellite, reflected satellite, and Sun, which are the inherent drawbacks of the method. It is notable that the influence of the relativistic effects on the measurement accuracy needs further research.