Journal of Circuits, Systems and Computers

Fourier transform points out that a certain function that meets certain conditions can be decomposed into a linear combination of trigonometric functions (sine or cosine functions). The functions of high-order cumulant include suppressing the Gaussian noise, eliminating independent signal components, and identifying the phase coupling phenomenon of the signal. To prove this hypothesis, this study constructs a cosine signal with the phase coupling phenomenon based on Fourier transform theory which substitutes it into the third-order cumulant expression and performs detailed reasoning. The constructed signal is extended to the complex signal domain and the same conclusion is obtained. The number of coupled signals is expanded from three to a higher value. The results of the study give definite mathematical and physical meaning to the bispectral peaks. The collected mechanical vibration signals are given to demonstrate this conclusion. The demonstrated characteristics of high-order cumulants have made them widely used in many fields.

This work presents a tunable receiver analog front-end (AFE) circuit, capable of achieving data rates between 14 Gb/s and 28 Gb/s. The circuit is implemented using the TSMC 65 nm CMOS technology. The circuit incorporates a continuous-time linear equalizer (CTLE), a transimpedance amplifier (TIA) and an output buffer. A $g_m$-boosted technique is employed within the CTLE for bandwidth and gain enhancement, while the TIA leverages a super source follower (SSF) structure to combat parasitic output node issues. By adjusting the tunable structures in the circuit, the proposed AFE can achieve variable DC gain at 14 GHz. Simulation results reveal that the AFE can compensate for the channel loss at the 14 GHz Nyquist frequency, delivering eye diagram spreads of approximately 0.85 UI and 0.78 UI at 14 Gb/s and 28 Gb/s, respectively. The complete post-simulation layout area is only 0.0026mm². Operating with a power consumption of 10.5mW, the AFE exhibits an exceptional FoM of 11.8 pJ/bit/dB.

The low-light image enhancement task aims to improve the visibility of information in the dark to obtain more data and utilize it, while also improving the visual quality of the image. In this paper, we propose a dual cross-attention multi-stage embedding network (DCMENet) for fast and accurate enhancement of low-light images into high-quality images with high visibility. The problem that enhanced images tend to have more noise in them, which affects the image quality, is improved by introducing an attention mechanism in the encoder–decoder structure. In addition, the encoder–decoder can focus most of its attention on the dark areas of the image and better attend to the detailed features in the image that are obscured by the dark areas. In particular, the poor performance of the Transformer when the dataset size is small is solved by fusing the CNN-Attention and Transformer in the encoder. Considering the purpose of the low-light image enhancement task, we raise the importance of recovering image detail information to the same level as reconstructing the lighting. For features such as texture details in images, cascade extraction using spatial attention and pixel attention can reduce the model complexity while the performance is also improved. Finally, the global features obtained by the encoder–decoder are fused into the shallow feature extraction structure to reconstruct the illumination while guiding the network for the focused extraction of information in the dark. The proposed DCMENet achieves the best results in both objective quality assessment and subjective evaluation, while for the computer vision tasks working in low-light environments as well, the enhanced images using the DCMENet proposed in this paper show the best performance.

Deep neural network has made surprising achievements in natural language processing, image pattern classification recognition, and other domains in the last few years. It is still tough to apply to hardware-constrained or mobile equipment because of the huge number of parameters, high storage as well as computing costs. In this paper, a new sparse iteration neural network architecture is proposed. First, the pruning method is used to compress the model size and make the network sparse. Then the architecture is iterated on the sparse network model, and the network performance is improved without adding additional parameters. Finally, the hybrid deep learning model was carried out on CV tasks and NLP tasks on ANN, CNN, and Transformer. Compared with the sparse network architecture, we finally found that the accuracy of the MINST, CIFAR10, PASCAL VOC 2012, and SQuAD datasets is improved by 0.47%, 0.64%, 3.75%, and 15.06%, respectively.

The fast scenario transfer for portrait styles has been a universal concern in the area of image processing. Conventionally, it was realized mostly by manual manipulation based on expertise experience, which costs a large amount of human labor. To fit the growing business volume in the era of Internet 4.0, it is supposed to investigate reliable autonomous workflow that generates portrait-style results via intelligent algorithms. Therefore, this paper proposes a fast scenario transfer method for portrait styles through collaborative awareness of convolutional neural network (CNN) and generative learning (GAL). On the one hand, a CNN structure is designed to make feature extraction towards the initial images, which can be used for following image generation. On the other hand, the GAL is utilized to generate new images with post-transferred scenario styles, so that the expected scenario style transfer results can be output by generating new images through long-term iterations. Especially, the loss function is designed by simultaneously considering local and global characteristics. Finally, the analysis and experimental results show that the correction process can be completed by injecting the extracted input face features into the feature maps of corresponding sizes in the generation process using a pre-trained feature extractor. Thus, the face conversion efficiency can be improved.

Denial of Service (DoS)-Oriented cyber attack has been a major threat for physical security in many kinds of network media, including the Internet of Vehicles (IoV). This paper focuses on the scenario of IoV, and proposes a machine learning-based fast detection method for adversarial neural network-based fast detection method for DoS-oriented cyber attacks. First, by analyzing the implementation principles and attack characteristics of three attack types, three aspects of statistical features are extracted: maximum matching packet growth rate, source address entropy value, and flow table similarity. Then, they are used as the input features to establish an adversarial machine learning-based DoS cyber attack detection method. On this basis, the field features of six stream rules are extracted, and two DoS cyber attack detection methods via machine learning are formulated. The proposals are able to detect the low-rate DoS-based cyber attacks against the data layer. The experimental results show that the proposed DoS attack detection method based on machine learning can effectively detect three DoS attacks under IoV, and these two algorithms have higher detection rates when compared with other algorithms.

Accurately predicting the maximum temperature is essential for studying human comfort, ecological environment development and social progress. However, traditional prediction methods are inefficient and inaccurate when dealing with large volumes of meteorological data. To tackle these challenges, this paper introduces an integrated approach, the ARIMA–LSTM–XGBoost model, which combines the strengths of autoregressive integrated moving average (ARIMA), long short-term memory network (LSTM) and eXtreme Gradient Boosting (XGBoost) to predict the maximum temperature. The proposed model enhances the prediction accuracy and convergence rate through techniques like MAPE reciprocal weight (MAPE-RW) and Schedule Sampling. Additionally, the model selects the best performing model using the early stopping method. This paper compares and analyzes the prediction results of the ARIMA, LSTM, XGBoost and ARIMA–LSTM–XGBoost models. The experimental results indicate that the ARIMA–LSTM–XGBoost model proposed in this paper achieves superior prediction accuracy, performance, and confidence. The ARIMA–LSTM–XGBoost model shows a Root-Mean-Squared Error (RMSE) of 1.381, significantly outperforming the ARIMA model (3.828), LSTM model (3.360) and XGBoost model (1.422). The coefficient of determination ($R^2)$ is 0.977, surpassing the values of 0.905 for the ARIMA model, 0.922 for the LSTM model and 0.910 for the XGBoost model. The ARIMA–LSTM–XGBoost model also exhibits a higher confidence level compared to the individual models.

In this paper, we propose a cascade AdaBoost neural network (CANN) based on concepts and construct of AdaBoost neurons and cascade structure. Compared with AdaBoost, CANN can represent complex relationships between features. In CANN, representation learning is performed through AdaBoost, and the method of random selection features is utilized to encourage the diversity of AdaBoost neurons. Through the cascade structure, CANN has the context structure for complex feature representation. At the same time, in order to avoid the problem of feature disappearance, shortcut connection is used to add the previous information to the later nodes. Furthermore, particle swarm optimization (PSO) algorithm is utilized to optimize the structure of CANN, it can obtain the number of iterations to achieve better performance. Two types of CANN are proposed based — binary-classification CANN (BCANN) or multi-classification CANN (MCANN). The performance of CANN is evaluated with two kinds of data sets: machine learning data sets and atrial fibrillation data set. A comparative analysis illustrates that the proposed CANN leads to better performance than the models reported in the literature.

A Transimpedance amplifier (TIA)-based optical receiver in active voltage-current feedback topology is designed for optical interconnects at 0.18 $\mu \mathrm{\text{m}}$ CMOS technology. The proposed inductorless TIA incorporates a gain boosting amplifier in its path to achieve high gain and bandwidth. This gain boosting amplifier is based on a regulated gain cascode (RGC) amplifier. The RGC stage raises the $g_m$ of the circuit, reducing the input resistance of the input stage; as a result, the bandwidth extends. To further amplify the gain, a common source (CS) stage is used at the output without much effect on the system’s overall bandwidth. An analytical model of the proposed TIA is developed and later simulated at room temperature in the Cadence Virtuoso analog design environment. The pre-layout and post-layout simulation results of the proposed TIA have been shown to validate its performance. It achieves a transimpedance gain of 62 $dB\mathrm{\Omega }$ while consuming power of 14.5 mW from a 1.8 V supply voltage. The bandwidth is extended to 8.2 GHz from 1.83 GHz with an input photodiode capacitance of 300 fF. The current spectral density of input-referred noise is 31 pA/$\sqrt{}Hz$.

A Sigma-Delta modulator (SDM) realized with a fully differential third-order single-loop cascaded integrator feedforward (CIFF) architecture is proposed. A pair of low-frequency chopper switches are nested outside the chopper amplifier to further reduce the residual offset voltage. To reduce the power consumption and ensure linearity, a high-speed dynamic comparator is also used to implement a one-bit quantizer. The proposed architecture and the corresponding functionality are first simulated in MATLAB Simulink at the behavioral level. The results show that the designed modulator has an SNDR of 124.9dB corresponding to an ENOB of 20.46 bits at a clock frequency of 256kHz and 312.5Hz input with a differential-mode voltage of 700mV sinusoidal waveform. Based on SMIC 180nm/1.8V standard CMOS process on the Cadence platform, the subcircuit-level simulation is also performed, while the result shows that the proposed modulator can effectively achieve 115.52dB SNDR, 18.90-bit ENOB, and 8.40mW power consumption, which correspond to FoM${}_{\mathrm{\text{W}}}$ and FoM${}_{\mathrm{\text{schreier}}}$ of 0.067pJ/step and 163.27dB, respectively. The proposed modulator shows a significant advantage to be applied for high-precision analog-to-digital conversion applications such as high-quality equipment for audio, ECG and EEG signal sensing.

The fetal electrocardiogram (FECG) signal carries several key pieces of information that could help clinicians make accurate decisions during labor. In order to make effective fetal monitoring, the detection of FECG signals is more important. However, the detection and classification of FECG signals are complicated tasks due to inappropriate learning ability. Thus, the proposed study aims to develop an efficient deep learning model for offering prenatal treatment to arrhythmic fetus by classifying the FECG signals. The input FECG signals are pre-processed in the initial stage using the Least Extended Mean Square Algorithm (LEXSA) to eliminate the awful noises. These pre-processed signals are fed to the input of the Ensemble Wavelet Empirical Component (EnSem_WEC)-based feature extraction stage, where the respective features are attained by applying Empirical Mode Decomposition (EMD), Independent Component Analysis (ICA) and Wavelet Transform (WT). The extracted features select the essential feature subset to avoid the high dimensionality issues through Enhanced Moth Flame Optimization (EMoFo) algorithm. Finally, the FECG signals are classified using the selected features by proposing a new Dense Attention-based Stacked Bi-directional Long Short Term Memory network model (DA_SBi-LSTM). The model is optimized by tuning the hyperparameter using the Pelican Optimization Algorithm (POA). The simulation analysis proves that the proposed deep learning model attained higher classification performance in terms of accuracy of 98.93%, precision of 97.5%, recall of 97.5%, and F1-score of 97.8%. As compared with other existing methods, the proposed techniques attained superior classification outcomes.

Nowadays, artificial intelligence-assisted automatic technological analyses of sports matches have received a more general demand. Among them, the effective recognition of sports actions from videos or images acts as the most fundamental issue to be solved. As sports gestures are characterized by remarkable specialty and instantaneity, it is required to develop specific effective recognition algorithms for this purpose. As a result, this work takes the case of Taekwondo sports as an example, and introduces deep vision learning to develop a specific intelligent recognition method for dynamic sports gestures. After sampling the key frames from dynamic videos, vital joint points of limbs are first obtained as the basic bottom features. Then, gesture expertise of this domain is formed to construct a multiview feature extraction work flow to better analyze complex characteristics of each gesture. Finally, the extracted feature representation is input into a recurrent neural network structure to output the discriminative results. At last, experiments are conducted for assessment, whose results show that the proposal can reach an accuracy of 95.6% and can be well suitable for the investigated problem scenario.

The credit risk assessment acts as an important part in daily affairs for financial institutions. But in the era of big data, the growing business volume makes it an urgent demand to develop digital ways of credit risk assessment. Currently, the machine learning is universally employed to establish various data-driven models for this purpose. However, machine learning models generally suffer from limited ability of feature representation and robustness, and cannot deal with more complex financial security scenarios. To deal with this issue, this work introduces ensemble learning to construct a stronger credit risk prediction model via integration of several basic machine learning models. Thus, an ensemble learning-enhanced smart prediction model for financial credit risk is proposed in this paper. Three classification-based machine learning models (support vector machine, artificial neural network and radial basis function) are selected as the basic classifiers, and “voting” strategy is utilized to integrate them into a novel strong classifier. A real-world financial credit dataset released by a Chinese commercial bank was selected as the experimental scenario. The obtained results show that the proposal has better prediction accuracy compared with basic machine learning models without ensemble learning.

Aiming at the problems about the incompleteness and low accuracy in classifying image feature extraction in existing image object detection methods. First, an improved image object detection method based on faster Region-Convolutional Neural Network (RCNN) is proposed, and Region of Interest (ROI) align is used instead of RoI pooling to reduce the error in the pooling process. Second, it adds the layer of convolution before adding network full-connection layer, thus reducing its parameters, enhance the performance of the classifier and avoid over-fitting. Finally, the combination of softmax and central loss functions is for training network-based model to increase the differences between categories and reduce the changes within categories, and thus the model with network could have more targeted and diversified features. Research results found that in total, method average accuracy is 91.04%, indicating higher than the accuracy of other methods and has good image target detection ability.

A Facial Expression Recognition (FER) method based on Conv3D-ConvLSTM-SEnet in an online education environment is proposed to address the issue of low accuracy in current classroom expression recognition (FER) methods. Firstly, ConvLSTM is used to integrate the local feature extraction ability of CNN and the temporal modeling ability of LSTM, and based on this, the rich spatial features of the image are characterized. Then, by introducing Depth Separated Convolution (DSC) to change the number of output channels and separate each channel, the Feature Maps (F-M) are sequentially concatenated to obtain a multi-channel output F-M. Finally, based on ConvLSTM and SEnet modules, a Conv3D ConvLSTM-based FER model was proposed by redistributing the extracted abstract features and basic texture features, achieving high accuracy classroom FER. The proposed classroom FER method and the other five methods were compared and analyzed through simulation experiments using the CK$+$, FER2013 and JAFFE datasets, respectively. The results indicate that the proposed method achieves the highest accuracy, precision, recall, and F1 score, with improvements of at least 1.85%, 2.41%, 1.18% and 2.05%, respectively, compared to the other five methods on the FER2013 dataset. The proposed method can perform FER on students in online classroom teaching, detect their emotions in real-time, and help teachers make better course teaching adjustments based on this.

The 21st century is an era of rapid development of high-tech industry. Through continuous collection and sharing of data, IoT technology connected by things has penetrated into every aspect of human life. Modern organizations find that they increasingly rely on knowledge and information sharing and interconnection to enhance their innovation and development capabilities. Therefore, this paper studies the strategic decision-making of knowledge sharing among employees on the IoT platform drawing on evolutionary game theory, this paper constructs a three-party game model composed of “organization-knowledge sharers-knowledge seekers” from the perspective of stakeholders, and discusses the strategic choice of organization and employee behavior under the dynamic decision mechanism and when the game reaches equilibrium and stability. Moreover, this paper uses MATLAB 2016a to simulate the model. The results show that under the premise of higher organizational rewards, synergistic benefits, high-shared extra income, knowledge-sharing preferences, incentive preferences of the organization, the system is easier to reach the ideal state. With the smaller cost of sharing and incentive, organization and employees are more willing to be motivators and sharers, while the incentive degree of organization should be controlled within a reasonable range. This paper can provide specific theoretical and practical guidance for the practice of organizational knowledge management.

The technology for path planning of independent mobile robots is mature, but multi-robot path planning for unified scheduling and allocation is much more complex than single-robot path planning. This requires consideration of collision problems between robots, general optimal path problems, etc. This paper proposes the use of the BSO algorithm for unified scheduling and allocation of multiple robots to improve the efficiency of task execution. The BSO algorithm is a new type of intelligent optimization algorithm that uses clustering ideas to search for local optimal solutions and obtains global optimal solutions by comparing local optimal solutions. It also uses mutation ideas to increase the diversity of the algorithm and avoid becoming trapped in local optimal solutions. Using the GA/SA algorithm and the proposed BSO algorithm for computer simulation comparison, we obtained the optimal path planning for the three robots under unified scheduling. The total distance of the optimal path obtained by the BSO algorithm was 27.36% and 25.31% shorter than those of the GA and SA algorithms, respectively. To further test the performance of the BSO algorithm, we conducted additional experiments on the unified scheduling of multiple robots. The experimental results show that the proposed BSO algorithm can significantly improve the efficiency. The multi-robot under unified scheduling performs point-to-point path planning without collisions, and they can traverse all task target points in the shortest path without repetition. This algorithm is suitable for multi-robot tasks in large-scale environments.

Battery Management System (BMS) functions to monitor individual cell in a battery pack and its crucial task is to maintain stability throughout the battery pack. The BMS is responsible for maintaining the safety of the battery as well as not to harm the user or environment. The parameters that are to be monitored in a battery are Voltage, Current and Temperature. With the collected data, BMS carefully monitors the charging–discharging behavior of the battery particularly in the Lithium-ion (Li-ion) batteries in which charging and discharging behavior are completely different. This paper proposes a real-time IOT connected deep learning algorithm for estimation of State-of-Charge (SoC) of Li-ion batteries. This paper provides unique objectives and congruence between model-based conventional methods and state-of-the-art deep learning algorithm, specifically Feed Forward Neural Network (FNN) which is nonRecurrent. This paper also highlights the advantages of Internet-of-Things (IoT) connected deep learning algorithm for estimation of State-of-Charge of Li-ion batteries in Hybrid Electric Vehicles (HEVs) and Electric Vehicles (EVs). The major advantage of the proposed method is that the Artificial Intelligence (AI)-based techniques aim to bring the estimation error less than 2% at a low cost and time without the model of the battery, at par with conventional method of Extended Kalman Filter (EKF) which is the best ever practical estimation theory. Another advantage of the proposed method is that in an abnormal condition (i.e., Unsafe Temperature) the IF This Then That (IFTTT) IoT mobile application interfaced with BMS through ThingSpeak cloud, sends a notification alert to the battery expert or to the user prior to an emergency. Finally, the real-time data of the battery parameters are collected through ThingSpeak cloud platform for future research and analysis.