Journal of Circuits, Systems and Computers

With the continuous development of encryption technology, the share of encrypted traffic in the network is increasing, which brings great challenges to the traditional methods of rule-based traffic identification. Deep learning is becoming an inspiring methodology to solve the problem. Previous studies have confirmed that time characteristics play an important role in Tor traffic classification. We find that there is a similarity of time characteristics among different programs. This paper proposes an end-to-end classification framework: the temporal sequence of data fluctuation network (TSDFN). It first extracts the temporal sequence of data fluctuation in the original flow and then uses the GRU network to learn the hidden temporal features. Experiments on public data sets validate the effectiveness of our proposal over other methods.

Existing Bayesian network (BN) structure learning algorithms based on dynamic programming have high computational complexity and are difficult to apply to large-scale networks. Therefore, this paper proposes a Dynamic Programming BN structure learning algorithm based on Mutual Information, the MIDP (Dynamic Programming Based on Mutual Information) algorithm. The algorithm uses mutual information to build the maximum spanning tree and $M$-order matrix, and introduces a penalty coefficient d based on the matrix-based node removal strategy, so as to reduce the number of scoring calculations and time consumption of the algorithm. Simulation results show that, compared with DP, SMDP and MEDP algorithms, the MIDP algorithm can reduce the calculation times and time consumption of algorithm scores while maintaining the accuracy of the algorithm when selecting the appropriate $d$ value.

This paper demonstrates a low-phase noise and high FoM_T differential CMOS LC-VCO working at 1.29–1.61GHz. An RC filter tail current source is used in this design to reduce the second-harmonic oscillation phase noise. The VCO is implemented in 180nm process technology. The post-layout simulation results indicate that this proposed VCO operates faithfully in the 1.29–1.61GHz band with a frequency tuning range (FTR) of 320MHz. By using inversion mode N-MOSFET varactor, the designed VCO obtains 22.06% FTR. The phase noise of the VCO is $-$130.21dBc/Hz at 1MHz offset frequency. It consumes 4.01mW average power from 1.3V supply. The cell area occupied by the design is $534\times 489{\mu \mathrm{\text{m}}}^2$. Figure-of-merit considering FTR (FoM${}_{\text{T}})$ is $-$194.14dBc/Hz at 1MHz offset frequency, phase noise and average power consumption. The proposed VCO can be used for low-phase noise GPS and satellite navigation.

This paper proposes a deduplication-based mixed page strategy, called DM-Pages, to improve the energy efficiency of disk storage systems and the cache performance. DM-Pages strives to improve the cache hit rate, simultaneously reduce the number of power-state transitions of a disk storage system and keep it in a low-power state as long as possible. DM-Pages deploys three innovative techniques. It adjusts the size of huge pages dynamically to maximize the cache hit rate. Meantime, content-aware deduplication is applied to eliminate redundant cache pages, targeting to enhance cache utilization. In addition, DM-Pages incorporates a mixed page mechanism into disk energy management to save energy. Our experimental results show that DM-Pages significantly improves the energy efficiency of disk storage systems and achieves a high cache hit rate. In particular, compared with traditional disk storage systems, DM-Pages conserves 6.59 times energy and improves the cache hit rate up to 3.36 times.

A three bit digitally controlled differential delay circuit is investigated in this work. The circuit is coupled to a normal high current power delivery network (PDN), which is capable of coming up from sleep mode (0A) to 10 or 30A or else 50A in 10ns. Simulated in a 0.09-$\mu$m CMOS process and power supply voltage ($V_{\mathrm{\text{dd}}}$) of 1.1V, the proposed circuit in post-layout can operate from 435.5ps to 152.2ps. The corresponding input vector (CBA) during this period switches from 000 to 111. Further, there are also intermediate states as the delay varies between the said ranges. These values of delay being dependent on the input vector has been noted to reduce monotonically. In addition to that, the post-layout delay sustains a shift of 4.0 and 6.5ps for NN and SS process corner, respectively, following a $\pm 10^{\circ }$C change in temperature. But an abrupt high current $I(t)$ pumped into the chip causes the voltage ($V_{\mathrm{\text{p}}}$) near the device to drop due to PDN. It also fluctuates with the natural frequency of PDN. The noise so developed induces jitter in the output swing and causes more delay between input and output in comparison to the circuit designed with zero noise in power supply. The jitter achieved for a current ramp $I(t)$ of 0–50A in 10ns for the two vector CBA happens to be around 5.0 and 4.4ps, respectively. The current ramp also generates power supply noise of 0.468V occurring close to the silicon chip. This, in turn, corresponds to a $V_p$ of 0.642V near the die, appearing to be much less than the required $V_{\mathrm{\text{dd}}}$ of 1.1V. Therefore, the delay is found to be affected significantly due to the sudden current ramp. It is also noted that a constant DC voltage of 0.642V has quite different effect on delay and jitter than a fluctuating AC noise having the first droop same as that DC voltage.

The lightning current dispersion process through the steel bar inside the concrete pile foundation of the transmission line tower is impeded by the high concrete resistivity. In order to improve the current dispersion performance of the pile foundation, this paper proposes a method to place flexible graphite grounding electrode along the shaft wall of the tower foundation and verified its feasibility. Both single-pile and four-pile foundations are applied to build vertical grounding models of transmission line tower with three different grounding structures, respectively, including flexible graphite grounding line, flexible graphite grounding fabric and combined graphite electrode. The effect of three grounding structures on the current dispersion and resistance reduction is studied by finite-element method. Simulation results indicate that by applying the above three grounding structures on the vertical pile foundation can significantly improve the current density distribution of the excavated pile foundation and reduce the grounding resistance. When the grounding resistance is $200\mathrm{\Omega }\cdot \mathrm{\text{m}}$, the resistance reduction efficiencies of the above three methods can reach 48.56%, 53.86% and 54.53%, respectively. Considering the limited construction area, this paper compares the foundation that is placed combined graphite grounding electrode with square-ray grounding grid, and it can be seen that by placing vertical grounding electrode can save the land area of 295.99m². This paper can provide reference for the design of excavation pile foundation grounding structure under limited construction area.

Traditional large-scale process manufacturing is gradually transformed into customized discrete manufacturing with the fierce global competition. Production planning has an important impact on improving manufacturing efficiency in the ever-changing from the view of engineering management. However, many nonprocessing-related factors in the flexible manufacturing system make it different between the formulation and implementation of the production plan. We established a multi-target optimization model based on the scheduling data of a discrete manufacturing company. In order to optimize the local effect of the scheduling model, we proposed an improved genetic algorithm with local search (GALS). The results of the experiments show that GALS is far superior to the current genetic algorithm scheduling in terms of the number and quality of scheduling solutions. Compared with the current scheduling strategy of the enterprise, the scheduling strategy given by GALS achieved an average improvement of 29.61% in minimizing completion time, achieved 44.8% in minimizing transportation time, and achieved 44.64% in machine load balancing.

Agent-based modeling is a promising approach for developing simulation tools for natural hazards in different areas, such as during urban search and rescue (USAR) operations. The previous studies aimed to develop a dynamic agent-based simulation model in post-earthquake emergency response operations combining rescuers and victims using geospatial information systems and multi-agent systems (GISs and MASs, respectively). We use multi-agent system engineering (MaSE) to construct this model and cluster points are proposed to assemble rescuers and victims. We also proposed an approach for dynamic task allocation and for establishing collaboration among agents based on contract net protocol (CNP) and the $k$-means clustering method. Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) and expert evaluation is used to sort the priorities of cluster points. The experimental background is set in the central area of Chengdu, Sichuan Province, China. 200 victims with different demands and 30 rescuers with the same rescue capability are randomly distributed in this area. We use different victim speeds to simulate different degrees of seismic damage. Compared with normal search and rescue, the ERSCIED can make some use of the capabilities of the victims for the improvement of emergency effectiveness when the victims and rescuers have a similar speed (1:1) or the victims are not much slower than the rescuers (1:2). When the victims are much slower than the rescuers (1:10), the rescuers have to wait for a long time at the cluster points and it will lead to a waste of resources. The parameters of rescuers and victims can be changed to meet different emergencies and the ERSCIED can provide effective auxiliary decision-making information for decision-makers.

Online crowdfunding, an innovative model based on “$\mathrm{\text{Internet}}+\mathrm{\text{Finance}}$”, is a hot spot for financing via Internet. Crowdfunding based on blockchain is an emerging economic phenomenon and becomes one of the most advanced risk financing strategies. However, crowdfunding transactions face security threats due to identity leaks, quantum attacks and the untraceable nature of blind signatures, which facilitate criminal activity. Different from the previous works, which ignored the importance of traceability, in this paper, we establish a blockchain-empowered secure crowdfunding architecture and propose an anti-quantum partially blind signature algorithm based on the verifiable identity of both sides. Specially, for one thing, the private key decided by user identity is generated by lattice-based sample matrix, and the privacy of user identity can be ensured and traced by the rejection sampling theorem. For another thing, we design an improved krill herd algorithm (IKHA) to increase the credit factor of fundraisers for dealing with project investment issues. The simulation evaluates the correctness and effectiveness of our theoretical analyses. Compared with the current popular schemes, the proposed IKH algorithm has a higher convergence speed and can optimize investment efficiency.

Existing super-resolution methods convert high-resolution images into low-resolution images, and use the synthesized images as input to train the model. However, it is difficult for synthetic low-resolution images to reflect the characteristics of real low-resolution images, resulting in poor model performance in practical applications. To address this problem, we propose a recurrent super-resolution framework, which consists of a degradation model and a reconstruction model. The degradation model degenerates the real high-resolution image into a more real low-resolution image, which is used as the input of the super-resolution reconstruction network, and then uses the reconstruction model to reconstruct the low-resolution image, and calculates the error with the original image. The generated high-resolution image is input into the degradation model again for degradation processing, forming a symmetrical and cyclic network structure, so that the super-resolution model has a better effect when reconstructing the real low-scoring image. In addition, the spatial attention mechanism is introduced into the generator network, which expands the receptive field of the convolution kernel, better extracts long-distance image features and improves the texture details of super-resolution images, which is consistent with the global.

In modern power system, along with the developments of the data collecting technologies, the intensive and high-dimensional load data collection can be achieved. Therefore, to deeply reveal the patterns and behaviors hidden in the load dataset using load classification is of great significance for improving the service quality and the user experience of the power system. However, inevitable issues, for example the data missing and class imbalance are frequently reported in the present load dataset, which deteriorates the performance of the classification algorithms. Also, due to the special features, for example the time series, periodicity, and fluctuation of the load data, the traditional data classification algorithms also encounter performance defects. Therefore, this paper presents a data augmentation based enhanced temporal convolutional network (TCN) algorithm in enabling load classification. In the data augmentation phase, first an LRTC-TSVD algorithm is presented to implement the missing data completion. Second, a WGAN based class balancing approach is further presented to solve the class imbalance issue. Then, in the enhanced TCN phase, a WeightNorm, exponential linear unit (ELU) activation function, residual connection, and bidirectional feature fusion techniques based improved TCN (ITCN) algorithm is presented to carry out the accurate load data classification. Combining the data augmentation and the enhanced TCN phases, the ITCN algorithm is finally conducted. Based on the benchmark load datasets, the performances of the presented ITCN are evaluated. The experimental results report that the presented data augmentations can improve the quality of the dataset, moreover the classification algorithm is able to achieve the satisfied classification accuracy.

Image registration of multimodal remote sensing images plays a vital role in remote sensing image analysis. However, there are significant nonlinear intensity differences between multimodal remote sensing image pairs, making it difficult for most traditional image registration algorithms to meet the registration requirements. In this paper, we propose a novel edge descriptor utilizing edge information, which has not only affine invariance but is also insensitive to nonlinear intensity differences. Moreover, we utilize the proposed descriptor to design a multimodal image registration algorithm. We use several different multimodal image pairs to evaluate the proposed algorithm. The experimental results show that the proposed algorithm holds a stable performance and can still achieve accurate spatial alignment even with the huge nonlinear intensity differences.

Feature detection is one of the basic research topics in the fields of image processing and corner that has been widely applied on vehicle detection, UAV image matching, camera calibration and so on. Particularly, a fast corner detector is of much benefit to many real-time tasks. A novel discrete curvature estimation method for corner detection based on the ratio of center distances of symmetric contour (RCDSC) is proposed in this paper. Benefiting from calculating the Euclidean distance twice only to estimate the discrete curvature at each point on a contour, RCDSC is much fast compared with other corner detectors. In addition, by choosing a relatively large radius of region of support and employing relative distance instead of absolute distance for constructing corner response function, RCDSC is also much robust to noises and local variations of contour. Extensive experiments exhibit the effectiveness and the efficiency of RCDSC in terms of average repeatability (AR), accuracy (ACU) and localization error (LE).

Following the experimental realization of memristors, researchers have focused on memristor-based circuits. Chaotic circuits can be implemented easily using a memristor due to its nonvolatile and nonlinear behavior. This study presents a memristor-based four-dimensional (4D) chaotic oscillator with a line equilibria. A memristor having quadratic memductance was utilized to implement the proposed chaotic oscillator. The 4D chaotic oscillator with quartic nonlinearity was designed as a result of the quadratic memductance. In terms of communication security, random number generation and image and audio encryption, systems with quartic nonlinearity or that are higher-dimensional are better than systems that are lower-dimensional or possess quadratic/cubic nonlinearity. The performance of the proposed chaotic circuit was investigated according to properties such as phase portraits, Jacobian matrices, equilibrium points, Lyapunov exponents and bifurcation analyses. Furthermore, the proposed system is multistable and its solutions tend to appear as twin attractors when initial conditions approach their equilibria. The Lyapunov-based nonlinear controller was constructed for controlling the proposed system having a line equilibria. The effect of the initial conditions on the controlling indicators was also studied. In conclusion, by using discrete circuit elements, the proposed circuit was constructed, and its experimental results demonstrated a good agreement with the simulation results.

This paper presents a new circuit structure of a multi-mode active filter, using only 16 transistors and 2 grounded capacitors in carbon nanotube field-effect transistor (CNFET) technology which is biased at $\pm 0.5$V supply. The proposed multi-input, single-output (MISO) filter has the capability of working as low-pass, band-pass, high-pass, band-stop and all-pass filters in all the operating modes (voltage, current, transconductance and transresistance). In addition, the quality factor ($Q$) parameter can be tuned electronically independent of the center frequency (${\omega }_0$). The HSPICE simulation results show that the proposed filter consumes only 971$\mu$W at 1GHz center frequency, assuming $C_1=C_2=0.1$pF and the nanotube pitch parameter of ($S=20$ nm) and the chiral vector of (22,0), while the number of nanotubes is considered as $N=16$ for all the transistors. Moreover, the main circuit performances such as the center frequency and the power consumption of the circuit vary by 5.4% and 11.6%, respectively, for the standard temperature variation.

The increasingly complicated and versatile convolutional neural networks (CNNs) models bring challenges to hardware acceleration in terms of performance, energy efficiency and flexibility. This paper proposes a reconfigurable neural accelerator (RNA) for flexible and efficient CNN acceleration. To provide hardware flexibility, RNA employs dynamically reconfigurable computing framework to rapidly configure data path between processing elements (PE) at run-time, as well as an interlaced data access mechanism for multi-bank RAM. To achieve high energy efficiency, three optimization mechanisms, including image row broadcasting dataflow (IRBD), tile-by-tile computing (TTC), and zero detection technology (ZDT), are dedicatedly designed for RNA to exploit data reuse and decrease memory bandwidth requirement, which is the key to improving performance and saving power consumption. To save hardware overhead, an online dynamic adaptive data truncation (DADT) mechanism is designed to compensate accuracy loss of multiplication results so that the computational precision in RNA can be reduced from 16-bit to 8-bit, which contributes to reducing the area of data path. The RNA architecture is implemented on Xilinx XC7Z100 FPGA and works at 250MHz. Experimental results show that the performance of running LeNet, AlexNet and VGG are 500 GOPS, 598 GOPS and 660 GOPS, respectively. Compared to previous FPGA-based designs, RNA achieves $1.5\times -4.3\times$ performance speedup and $7.6\times -8.4\times$ improvements on energy efficiency.

Recently, massive deep learning-based image dehazing methods have sprung up. These methods can effectively remove most of the haze and obtain far better results than the traditional methods. With the removal of the haze, however, edge details of the image are also lost, which is usually more noticeable in the gradient space. This paper proposes a gradient guided dual-branch network (GGDB-Net) for image dehazing. Specifically, we explore the hazy image gradient map to guide our model to focus on the hazy regions and edge restoration. We implement two parallel branches with a comprehensive loss function, which collaborate to dehaze and repair the lost edges in haze images. Moreover, the gradient-guided approach can potentially be applied to existing learning-based image dehazing models to boost their performance. Experimental results indicate that our results have good visual perceptions and are comparable to state-of-the-art methods in quantitative metrics.

Automatic classification of texts is a well-known topic in natural language processing (NLP), and it involves categorizing unstructured texts into specific groups. Classification algorithms for a given problem are searched for, as well as the appropriate features to use as input. These models have been used for analysis of data with a large number of sparse dimensions. In medical field, categorization of medical texts is a difficult task because these types of texts contain several terminologies that define medical ideas and words. Medical data also lack proper sentence structure and do not adhere to the standard rules of natural language grammar. The main problem in this classification task is to choose an appropriate representation as input. To solve the representation problem, we propose a new extension model for medical text personification. Our model starts with the preprocessing of medical text. Then, we use external terminology resources to expand the text. After that, we use a combination of different text vector representations. To speed up the process, we try to increase the accuracy while selecting the optimal classification parameters. In this topic, we use an adaptive particle swarm optimization (PSO) algorithm to select the best parameters of a machine learning model. The PubMed, Hallmarks and AIM dataset were used to test our model. Our categorization text model outperforms the competition and provides much higher retrieval accuracy than previous models.