In this paper, we propose a reconfigurable antenna using graphene which can work in terahertz band. The antenna use folded slot structure and coplanar waveguide to achieve impedance matching easily. By adjusting the conductivity of the four graphene sheets added between the slots, we can change the resonant frequencies of the antenna from 230.5 GHz to 270.5 GHz. In addition, a metasurface consisted of square closed-ring resonators (CRR) is under the antenna used to increase the gain of antenna. Because of the reflection of metasurface, the gain of antenna in two work mode increases 2.48 dBi and 3.55 dBi respectively.
With the transformation of the computing power supply pattern in the digital economy era, the new network infrastructure with computing power as the core has become an important driving force to realize the sharing of computing power resources and support the digital economy transformation. In the computing power network, multiple heterogeneous user terminals access the network frequently in various ways to obtain computing power services anytime and anywhere, which increases the openness and dynamics of the network. Hence, the computing power network will face more severe security challenges. However, the traditional network-based security defense pattern usually statically supplements security protection components for specific security issues, which cannot actively adapt to user needs to adjust defense strategies flexibly, which is difficult to deal with security risks in computing-network integration scenarios. Therefore, facing the security requirements of the new computing power network, this paper regards security as the internal attribute of the network and proposes a multi-dimensional collaborative autonomous defense paradigm based on the smart computing integration networks, which combines the design of “three layers” and “three domains” of the network. In the “three layers”, this paper defines the security inherent service at the generalized service layer, adapts the security function at the mapping adaptation layer, and executes the security strategy at the fusion component layer. In the “three domains”, the resource adaptation is guided by the entity domain, the security service process is driven by the knowledge domain, and the specific security technologies are implemented by the sense control domain. It constructs a basic management and control process that integrates “detection”, “trace”, and “defense”, in which security policies and technologies can be flexibly adjusted according to scenario scalability and business security. Finally, the proposed paradigm is verified through simulation experiments, and the results prove the effectiveness of the proposed paradigm and also provide a reference for further research and application of smart computing fusion security in the future.
In this article, a circularly polarized leaky-wave antenna with beam-scanning from backward to forward is proposed based on the slotline structure. Each leaky-wave unit cell consists of a slotline and four short-ended slots loaded on the two strips. To realize the property of backward-to-forward beam scanning, the length of each short-ended slot is about half of the guided wavelength at the broadside frequency. By designing the location of the short-ended slots as well as the angle between the short-ended slot and the propagation direction of the guided wave, the circularly polarized ability is obtained. As a validation of the proposed circularly polarized leaky-wave antenna, a prototype with broadside frequency of 10 GHz is designed and fabricated. Both the simulated results and the measured results show that this antenna exhibits good properties of circular polarization and beam scanning. The measured results suggest that this antenna exhibits left-handed circular polarization in the upper-half space above the antenna plane, and scans from -27° to +25° with gain variation of 7.0 to 10.9 dBic as the frequency varies from 9 GHz to 11 GHz. Symmetrically, this antenna features right-handed circular polarization in the lower-half space below the antenna plane, with beam scanning from -153° through ±180° to +155°and gain variation of 8.4~12.3 dBic. Compared with the reported antennas, the proposed circularly polarized leaky-wave antenna features a simple configuration. Meanwhile, the proposed design based on the short-ended slot achieves a low transmission coefficient in the working band.
Auto tongue color classification is an important research topic in the study of TCM (Traditional Chinese Medicine) objectification. Affected by various factors such as doctor’s experience and illumination conditions, there often exist errors in the manually annotated labels, that is, noisy labels. Noisy labels will cause the model not to converge in the training process and the generalization ability will be poor. Therefore, in this paper, a TCM tongue color classification method is proposed by progressively correcting noisy samples. First, according to the characteristics of the tongue color classification, a global-local feature fusion method is proposed, which is embedded in the ResNet18 backbone network, constructing a tongue color classification network. The ensemble learning paradigm is adopted to improve the reliability and stability of the classification model. Next, for the classification network training problem under noisy samples, a sample attention mechanism and a re-labeling mechanism are proposed. During the training process, different weights are assigned to clean samples and noisy samples, and the noisy samples are gradually adjusted. Finally, the network model is optimized and trained with the Boostrapping loss function to suppress the impact of noisy samples on the classification performance. The experimental results on two tongue color classification datasets SIPL-A and SIPL-B show that, the proposed method can effectively correct noisy labels, thereby, significantly improving the tongue color classification accuracy. Compared with the existing image classification methods under noisy samples, the proposed method can achieve a higher classification accuracy, reaching 94.6% and 93.65%, respectively.
The theoretical model of cyclotron electron to radiate vortex microwave photons is crucial for the technology of quantum state vortex electromagnetic wave. This paper is the second part of “The Vortex Electron and Vortex Microwave Photon” series, which establishes the theoretical model related to the “Vortex Microwave Photon Radiation”. The radiation by energy level transition of electrons can generate a single microwave photon carrying IOAM (Intrinsic Orbital Angular Momentum). Aiming to clarify this radiation mechanism, the probabilities of Landau energy level transition in non-relativistic and relativistic effects are deduced. Because of the linear relationship between the Landau energy level and the IOAM in the non-relativistic effect, the electron can only radiate the plane wave photons regardless of the initial state of the transition. It is opposed to the relativistic effect, where the microwave photon with rich IOAM modes values can be obtained. This paper also points out that in practical engineering, some cyclotron devices can be used as radiation sources. For mode selection, according to the characteristic of the frequency distinction between different IOAM modes, the iris-embedded waveguide filters can be used for frequency selection. At the same time, the microwave photons of specific IOAM modes can be selected. This paper concludes with a discussion of the quantum and statistical state vortex electromagnetic waves which reveals the corresponding pro and con in wireless communications.
The long magnetic lens is an important part of the dilation framing camera. It can generate a strong magnetic field in the drift region when stimulated by large current, thus improving the spatial resolution of the camera. Due to the joule heat effect, the long magnetic lens cannot work in large current environment for a long time. In this study, the current pulse generator applied to the long magnetic lens is designed based on the series and parallel Marx structure. The effects of charging voltage, energy-storage capacitance, load, circuit series and parallel stages on the output current pulse are studied. While the charging voltage and energy-storage capacitance increases, the pulse peak current becomes larger. The same is true when the number of series stages is increased or the load decreases. When the number of parallel stages increases, the peak current increases at first and then decreases. In addition, the pulse width is larger by increasing the energy-storage capacitance, the load, and the number of parallel stages. When the number of series stages increases, the pulse width decreases. A long magnetic lens model is established to simulate the magnetic field intensity distribution on the axis. The larger the inner diameter of the long magnetic lens, the weaker the magnetic induction intensity. The greater the excitation current, the stronger the magnetic induction intensity. A pulse with peak current of 457 A and full width at half maximum of 6.76 ms is generated by Marx pulse circuit with four stages in series and two in parallel. The current pulse excites the long magnetic lens, and the spatial resolution of 53.72 μm is obtained.
Static random-access memory (SRAM) physical unclonable function (PUF) makes use of the process deviation in the manufacturing process of transistors with identical parameter design, which generates the key response that cannot be cloned for each chip. Due to the randomness of SRAM-PUF internal error distribution, key reconstruction requires the use of error correction codes, and the area of error correction circuits is positively related to its error correction capability. In order to reduce the error distribution of SRAM-PUF and reduce the area of error correction circuits,this paper proposes a data remanence preselection algorithm through the research on characteristics of SRAM data remanence, screening SRAM cells, improving the stability of PUF response, and screening SRAM blocks using block optimization algorithm, reduce the dispersion of the response, which generates SRAM PUF response in a shorter time and resource consumption. Experimental results show that 256 bits SRAM-PUF response has 99.8% stability and bit error rate under different temperatures (-40 ℃~80 ℃) and ±10% voltage fluctuations. Compared with the general temporary majority voting (TMV) algorithm, the stability is improved by 1.7% and the error rate is reduced by times, compared with 1 000 times of TMV, linear reduction of time complexity from O(2 000n) to O(900n). After 72 hours of aging testing, the stability of the SRAM-PUF pre-selected using the data remanence algorithm only decreased by 0.2%.
This paper proposes a dynamic metamaterial structural polarization converter based on bulk Dirac semimetal (BDS). Different from the traditional polarization converter, the polarization conversion ratio and asymmetric transmissions of this polarization converter can be adjusted by Fermi level. The simulation results show that when the Fermi level of BDS are adjusted at 90 meV, the polarization conversion ratio is greater than 90% in the frequency range of 1.085~1.872 THz (Bandwidth is 0.787 THz). The polarization conversion ratio is peaked at 98% at the 1.159 THz. The asymmetric transmissions of this polarization converter is greater than 60% in the big frequency range of 1.229~1.831 THz. BDS can be controlled by Fermi energy, that explained the control mechanism of polarization conversion rate and the asymmetric transmission. The polarization rotation angle, ellipticity angle, surface current and electric field distribution are studied to clarify the physical mechanism of polarization deflection. By applying gate voltage to control the Fermi level of BDS, the dynamic tuning of polarization conversion rate and asymmetric transmission of the polarization converter is realized. The tunable asymmetric transmission characteristics of this polarization converter provide a new idea for the manufacture of multiplexers, THz diodes and other equipment.
In order to improve the accuracy of time-varying channel estimation in generalized frequency division multiplexing (GFDM) systems, a joint iterative channel estimation and symbol detection algorithm for GFDM systems using sparse Bayesian learning is proposed. Specifically, we use a GFDM multi-response signal model with non-interfering pilot insertion. Under the sparse Bayesian learning framework, we combine the expectation-maximization (EM) algorithm and the Kalman filter and smoothing algorithm to realize the maximum likelihood estimation of the block time-varying channel. Consequently, GFDM symbols are detected based on the estimated channel state information (CSI), and the accuracy of the channel estimation and symbol detection is progressively improved through the iterative processing of the channel estimation and symbol detection. Simulation results demonstrate that the proposed algorithm can achieve better bit error rate (BER) performance close to that under perfect CSI conditions, and it has the advantages of fast convergence speed and high robustness to Doppler frequency shift.
In order to solve the problems of high dynamic topology, large data transmission, multi-signaling cost of low-orbit satellite network, and to meet the growing demand of global coverage and massive number of mobile users, the software defined satellite network architecture and virtual topology model of high orbit and low orbit satellite collaboration are built on the basis of quality of services (QoS) of different traffic in operational performance index systems. A priority assignment function considering both traffic arrival rate and wait time is designed, and a joint inter-satellite routing algorithm based on traffic-aware and traffic scheduling is proposed. The algorithm includes a low-complexity intra-orbit priority forwarding routing algorithm for high QoS requirement services and a link weighted routing algorithm for low QoS requirement services. Simulation results show that the performance of the proposed algorithm is significantly better than that of the traditional algorithm in average end-to-end delay, packet loss rate and throughput.
Diversified communication mission requires more and more antennas to be integrated into the same system. Adjacent antennas inevitably influence each other, leading to a decline of the mutual isolation of the antennas’ ports. This paper proposes a planar/wideband coupling suppression structure (PWCSS) resonated in the high frequency band to reduce the radar scattering cross section (RCS) of the low band element and absorb energy from the low band element in the high frequency band and improve the mutual isolation of the two antenna elements’ ports in the high band. Compared to the previous approach, the PWCSS proposed in this paper has the following advantages. First, its usage will not bring extra gain loss for lower band element, and will not affect the radiation pattern of both the lower and higher band elements. Second, it does not require lamination process, which means low cost and high integration. Third, it contains open circuit microstrips with various lengths and positions, which can effectively extend the bandwidth of coupling suppression. The author simulated, manufactured and tested the antenna. Results show that the average/maximum mutual isolation in the high frequency band can be enhanced by 16.6 dB and 30 dB when the PWCSS is applied. This coupling suppression method can be very useful to base station, detection, radar and other multi-antenna systems, and has a good application prospect.
With the increasing demand for high-speed, ultra-compact data signal management, photonics integrated circuits (PIC) have attracted extensive attention in advanced photonic signal processing. Programmable photonic chips are gradually becoming favored by developers for their inherent generic and reconfigurable features. Based on United Microelectronics Center Co.,Ltd (CUMEC) silicon-on-insulator (SOI) integrated photonics platform, we have designed and fabricated a 3×3 hexagon-shape programmable photonic chip containing 9 hexagon units. The optical switch's extinction ratio of up to 30 dB was achieved by the “Iteration Scanning Method”, and there is a 21% chip size reduction through “Flatting” layout optimization. The typical functions of this programmable chip are further verified by tuning the input/output coupling coefficients of tunable basic unit (TBU). This allows for a Mach-Zehnder interferometer (MZI) or a micro-ring resonator (ORR) with a tunable extinction ratio, resonance wavelength and free spectra range to be realized, and the corresponding extinction ratio can reach 42.3 dB and 28 dB, respectively. Other functions, such as an optical delay line with a wide tunable range, as well as routing between any input and output ports, are also successfully realized with this programmable photonic chip. This is the largest scale hexagon-shape programmable chip which could be integrated with active devices. This chip offers a wide range of potential applications, in the areas of microwave photonics signal processing, bio-chemical sensing, quantum information and optical computing.
To realize the orthogonal matching pursuit (OMP) algorithm on a miniaturized and low-cost hardware platform, for calculation of the least square method in the OMP algorithm, this paper constructs a deterministic perception matrix and proposes a low-complexity, low-resource weighted QR decomposition OMP (WQR-OMP) algorithm hardware architecture, and the WQR-OMP SOC system is built on the ZYNQ 7020 chip. The WQR-OMP algorithm is that after the QR decomposition of the sensing matrix according to the distribution characteristics of the elements in the triangular matrix , the elements on the main diagonal are retained through the weighting operation, which returns other elements to zero to obtain the diagonal matrix , and then approximately computes the solution for the sparse vector. The experimental results show that compared with the hardware architecture of OMP algorithm based on QR decomposition OMP (QR-OMP) and Batch-OMP algorithm, the WQR-OMP algorithm has lower computational complexity and fewer storage resources. The reconstruction time of the WQR-OMP SOC system is about 400 ms for 256×256 resolution images at a compression rate of 0.25, which is 6.3 times faster than the ARM processor does. Compared with other existing researchers, this system further improves the reconstruction speed with less consumption of Block RAM storage resources and is suitable for hardware platforms with limited storage resources.
In this paper, single-layer Graphene, single-layer VS2 and single-layer BN were synthesized into Graphene/VS2/BN van der Waals heterostructure by van der Waals interaction, and the feasibility of using it as anode electrode material in li-ion batterys (LIBs) was studied by combining it with different amounts of lithium. Graphene/VS2/BN van der Waals three-layer heterostructure has a formation energy of eV/Å2, which has strong stability and can be synthesized theoretically. At the same time, the in-plane stiffness of Graphene/VS2/BN van der waals heterostructure is also calculated, and the Young’s modulus (Y) is 886.88 N/m, which is higher than that of single-layer VS2 (82.5 N/m), and it has good mechanical properties. The adsorption energy ( ) of Li adsorbed on the surface and interface of Graphene/VS2/BN van der Waals three-layer heterostructure is much larger than that of the corresponding monolayer, which indicates that it has good adsorption performance for Li. When Li migrates at different surfaces and interfaces of Graphene/VS2/BN van der Waals three-layer heterostructure, the diffusion barrier is very small (0.3~0.6 eV), which is extremely beneficial to the battery rate performance. Our research shows that the Graphene/VS2/BN van der Waals three-layer heterostructure has a broad prospect in anode electrode materials of LIBs.
The mode division multiplexing (MDM) system based on the few mode fiber (FMF) can realize the parallel transmission of different independent channels by using the orthogonal modes in a single fiber core, which can effectively improve the system capacity and solve the “capacity crunch” of the single mode fiber (SMF) system. In this paper, we realize the transmission of the 32 Gbaud four-mode dual-polarization 16-ary quadrature amplitude modulation (16QAM) signals in 80 wavelength channels over 1000 km six-mode graded-index fiber by using jointly the MDM, wavelength division multiplexing (WDM) and polarization division multiplexing (PDM) techniques. In addition, we use the hybrid time-frequency domain data-assisted 8×8 multiple-input multiple-output (MIMO) least mean square (LMS) equalization at the receiver to compensate for the linear damages such as the mode coupling (MC) and differential mode group delay (DMGD) caused by the fiber transmission. Compared with the commonly used time-domain or frequency-domain MIMO-LMS equalization, the joint use of time and frequency domain can realize 57.1% improvement in convergence speed and 25.1% reduction in convergence error. After 1 000 km delivery of FMF, the bit error ratio (BER) of each mode and polarization state of all the 80 channels is lower than the 20% soft-decision forward error correction (SD-FEC) threshold of 2.4×10-2, and the net transmission rate is up to 68.2 Tbit/s.
In this paper, a miniaturized phase-shifting power divider based on slow-wave substrate integrated waveguide (SW-SIW) is proposed for the functional integration of phase shifter and power divider. The design features two output branches with equal lengths and widths to achieve a 30° phase shift. One output branch is realized by a conventional substrate integrated waveguide (SIW), while the other output branch is loaded a complementary split-ring resonator (CSRR) on the upper metal surface to replace the continuous metal surface of the conventional SIW, and the CSRR is evolved from the classic CSRR structure. Meanwhile, a metallized via-hole is added inside the CSRR to reduce the phase instability caused by the loading of the CSRR, so a SW-SIW is realized, and the cut-off frequency and phase velocity are reduced. The measured results show that the reflection coefficient |S11| of the phase-shifting power divider is less than dB in the frequency band of 9.0~11.8 GHz, the relative operating bandwidth is 26.9%, and the insertion loss is less than 1.3 dB. The phase difference between the two output ports is stable at , and the amplitude difference is less than 1.4 dB realizing equal power distribution. The designed phase-shifting power divider has a small size and low manufacturing cost, which is suitable for application in phased-array antennas.
Time reversal technology has natural anti-eavesdropping capability due to its unique spatial and temporal focusing capability. In this paper, an optimization scheme to improve the security performance of time reversal OFDM system is studied. Firstly, the temporal and spatial focusing characteristics of time reversal pre-filtering are applied to improve the signal strength of the legitimate receiver relative to the eavesdropper, and then the power of the subcarriers is optimized to maximize the security rate. In order to further improve the secure capability of the system, the zero-space artificial noise is realized by employing the degree of freedom provided by the circular prefix, which can effectively interfere with the eavesdropper, while does not interfere with the legitimate receiver. Then, the subcarrier power and the covariance matrix of the artificial noise are optimized to maximize the security transmission rate. Simulation results show that the proposed scheme can significantly promote the security rate of the system.
In order to meet the market demand of medium and low voltage consumer electronics, the small size and high density Bipolar-CMOS-DMOS technology has been vigorously developed. Low loss and high reliability have become the focus and difficulty in the design of lateral double-diffused metal-oxide-semiconductor field effect transistors in Bipolar-CMOS-DMOS technology. This paper introduces a lateral double-diffused metal-oxide-semiconductor field effect transistor based on the high temperature oxidation layer structure, and studies and analyzes the degradation mechanism of its hot carrier injection. The high temperature oxidation layer structure is used to improve the traditional shallow trench isolation structure, in which the oxide steps embedded in the semiconductor have adverse effects on the hot carrier injection of the device. Thus improve the reliability of the device. The proposed structure shortens the current path length in the on state of the device and reduces the loss. In addition, this paper also proposes a self-aligned implantation process optimization method for the P-type body region. By increasing the implantation process of the high-energy body region, the depletion region in the accumulation area is expanded, the surface electric field of the drift region is reduced, and the breakdown voltage is improved. The proposed HTO-LDMOS has a breakdown voltage of 43 V, a specific on-resistance of 9.5 mΩ·mm2, and a linear region current degradation of 0.87% after 10 000 s.
A high-concentration doped density of states model (HCD-DOS model) was established for amorphous indium gallium zinc oxide (a-IGZO) thin-film transistors (TFTs) with back-channel etch (BCE) technology. The effect of the key parameters of the density of states on the device performance was also investigated by numerical simulation to reveal the physical mechanism of the preparation process to repair the conductive channel in a-IGZO TFTs. Firstly, the molybdenum/copper bilayer structure with high bonding strength was used as gate/source/drain electrodes, and the bottom-gate top-contact (BG-TC) TFTs was prepared by introducing the BCE method. Secondly, the HCD-DOS model of a-IGZO TFTs suitable for BCE technology was developed. Subsequently, the key parameters of the density of states were investigated numerically based on the TCAD (Technology Computer Aided Design) simulator. The results demonstrated that different density of states parameters had different effects on the transfer characteristic curves, electrical characteristics, and electron concentration distribution inside the channel of the a-IGZO TFTs device. At last, the influence of SiO x passivation-layer deposition and N2O plasma treatment on the internal mechanism of the device was explored based on the HCD-DOS model. It was found that N2O plasma treatment had a significant effect on the density of states distribution and channel carrier concentration, which in turn caused the threshold voltage to drift.
In the digital signal processing unit of the mode-division multiplexed system, the multi-input and multi-output (MIMO) equalization technology is usually used to compensate for the signal bit error rate (BER) degradation disturbed by various mode-dependent noises.The performance of MIMO equalization algorithm depends heavily on the step size factor μ and the number of taps K, so before welding the equalizers, it’s important to determine the optimal value of μ-K combination in MIMO equalization algorithm.A genetic algorithm (GA) based MIMO equalization parameter optimization scheme, namely GA-MIMO, is proposed to improve the efficiency of the parameter optimization, which is used to reduce the computational costs required during parameter optimization with the minimum BER output.In order to verify the performance of GA-MIMO, a point-to-point communication experimental system based on 10 km six-mode fiber is constructed.The new scheme is used to compensate the parallelly transmitted six-channel data, and the performance is compared with the steepest descent method and iterative algorithm.The experimental results show that the proposed GA scheme achieves the hit rate of the optimal μ-K parameters in MIMO equalization up to 99.98%, and the global search function of GA algorithm helps save the number of calls to the equalization algorithm of 86.14% and 90.3% compared with the steepest descent algorithm and iterative algorithm, respectively, effectively reducing the computational cost of locating μ-K parameters.
At present, rumor detection methods on social platforms mainly focus on obtaining information from the propagation path, most of these methods only use text information as the initial propagation feature, which is difficult to capture the rich propagation structure representation. In this paper, according to the propagation path of rumors, text and user credibility features are extracted, and a multi-feature rumor detection model based on propagation tree is constructed. This model aggregates text propagation features through a graph convolutional network, and uses a multi-head attention module to mine the intra-layer dependencies of the text propagation tree. At the same time, a credibility sequence is constructed for each user in the user propagation tree, and the M-Attention module is used to capture effective user credibility features. The experimental results show that the experimental accuracy of Twitter15, Twitter16 and Weibo datasets reaches 89.3%, 91.7% and 96.4%, which are 4.8%, 4.2% and 3% higher than the current optimal propagation tree model Bi-GCN (Binary Graph Convolutional Network) accuracy respectively.
Association analysis between genes and phenotypes is crucial to reveal the inherent genetic association of organisms. Random walk-based algorithms can fuse multiple omics data, aggregate the label information of first-order or higher-order neighbors, complete the association information between different nodes in the network, improve the accuracy of association prediction and further discover the potential genetic associations between genes and phenotypes. However, existing random walk algorithms usually treat each node equally and ignore the varying importance of different nodes, as such non-important nodes can be excessively propagated and the model performance is compromised. To this end, an individual multiple random walks (iMRW) algorithm based on multi-omics data fusion is proposed. On the heterogeneous genetic network composed with genes, miRNAs and phenotype nodes, we design the individual multiple random walks strategy based on the network topology, assign nodes of different importance with different walking lengths. We then complete the genetic information of different nodes by fusing multi-source association matrix, Gaussian interaction profile kernel similarity and random walk, and accurately obtain the gene-phenotype association prediction matrix. Under different experimental settings, iMRW can achieve the best prediction performance compared with the state-of-the-art algorithms. The case study with respect to maize photosynthetic ability and starch content further confirm the usefulness and effectiveness of iMRW in identifying potential gene-phenotype associations.
RISC-V instruction set architecture (ISA), as a new streamlined ISA, has developed rapidly due to its characteristics of free, open source, and freedom. Since the research on RISC-V at home and abroad mainly focuses on hardware development, the software ecosystem is still weak compared to mature ISAs. Implementing a set of high-performance basic math libraries for the RISC-V instruction set can further enrich the RISC-V software ecosystem. This paper realizes the transplantation of Sunway math library to RISC-V based on automatic transplantation technology, and provides the first basic math library system using vector instruction optimization for RISC-V instruction architecture. This paper proposes an automatic branch look-up table method and a path marker insertion method for vector registers, focusing on solving the problem of register multiplexing in the process of register mapping between different architectures, realizing the correct and efficient mapping of registers, and automatically transplanting 69 mathematical functions according to different instruction equivalence conversion strategies. The test results show that the RISC-V basic math library function can achieve correct calculation, the maximum error is 1.90ULP, and the average performance of functions is 157.03 beats.
Cellular automata are widely considered as a fundamental architecture for nano-computers and quantum computers based on molecular self-assembly technology. In such a situation, the complexity of cellular automata will directly affect their efficiency of parallel distributed computing, together with the feasibility of physical implementation. The simplest model of all asynchronous cellular automata in the literatures employs three cellular states and three transition rules, which can construct all logic circuits and thus hold the computational universality equivalent to Turing machine (Turing universality). In order to further reduce the complexity of universal asynchronous cellular automata, this paper proposes a new model, which requires only three cellular states and two transition rules. The smaller number of transition rules is mainly attributed to the new circuit elements and the design of large-scale distributed circuits. Different from the logic gates of synchronous circuits, the novel circuit elements can effectively utilize the random fluctuations of signals, whereby they may promise potential applications via nano-technologies such as single-electron tunnel transistors. In addition to Turing universality, this paper explicitly provides a scalable and distributed scheme to construct parallel logic circuits, which enables our proposed asynchronous cellular automaton to realize the same parallel computing capability as synchronous cellular automata.
Since the traditional sentiment classification methods for text comments usually ignore the influence of user personality on sentiment classification results, a sentiment classification method for text comments based on user personality and semantic-structural features is proposed. According to the advantage of Big Five personality model on effectively expressing the user personality, the user personality feature is obtained from the comment texts by calculating the personality scores from different dimensions. Moreover, the advantages of bidirectional gated recurrent unit (BiGRU) and convolutional neural network (CNN) on effectively extracting the contextual semantic features and the local structural features are taken, and a new text semantic-structural feature acquisition method based on BiGRU, CNN and two-layer attention mechanism is proposed. Finally, in order to distinguish the influence of the features with different types, the hybrid attention layer is introduced to obtain the final text vector representation by integrating the user personality feature and the textural semantic-structural feature effectively. The experimental results on the datasets of IMDB, Yelp-2, Yelp-5 and Ekman show that BF_BiGAC achieves good performance when the measurements of Accuracy and weighted macro F 1 (F w) are used. Specifically, it achieves the improvements of 0.020, 0.012, 0.017 and 0.011 compared to sentiment classification method concatenating BiGRU and CNN (BiGRU_CNN) on accuracy, and achieves the improvements of 0.022, 0.013, 0.028 and 0.023 compared to sentiment classification method concatenating CNN and BiGRU (ConvBiLSTM) on Fw. Moreover, when comparing with the pre-trained models of BERT and RoBERTa, BF_BiGAC achieves higher executing efficiency while ensuring the classification accuracy.
Aiming at the problems of self-supervised monocular depth estimation in current traffic scenarios, such as weak feature expression ability, fuzzy local details of depth map and low accuracy of depth estimation, a self-supervised monocular depth estimation method based on dual attention mechanism and adaptive cost volume is proposed. Firstly, a dual attention mechanism combining channel attention and spatial attention is used to adaptively weight the extracted scene features to enhance the feature expression ability of the feature extraction network. Secondly, according to the adaptively constructed cost volume of extracting global features, the network is guided to learn fine depth features, which improves the learning ability of the network model for the local details of the depth map and solves the problem of low accuracy of existing depth estimation methods. Experimental results on public datasets KITTI and Cityscapes show that the proposed method is superior to the current mainstream methods.
Polynomial multiplication operations limit the practical applications of lattice-based post-quantum cryptography. In order to improve the performance and efficiency of post-quantum cryptography Crystal_Kyber algorithm, and reduce the running time and reduce the influence of polynomial multiplication,this paper designs a new butterfly operation unit to optimize the Kyber scheme with prime modulus . First of all,the algorithm is executed by 16-way parallel scheduling of the new butterfly operation unit, which shortens the calculation cycle. Secondly, using pipeline technology and improved K2RED algorithm, the design and implementation of a new butterfly operation unit for reducing resource consumption. Ultimately, the data is stored in the way of multi-RAM, and the multi-channel RAM is optimized to allow data to be stored alternately in RAM and improve the resource reuse rate. The experimental results show that the optimized NTT (number theoretic transform), INTT (Inverse NTT), PWM (point-wise multiplication) efficiency reaches 200 MHz, and the combined execution Kyber efficiency reaches 175 MHz, which is superior to other schemes and has good performance.
Aiming at the problem that it is difficult for base stations to collect and manage instantaneous channel state information in high dynamic vehicle networking environment, a spectrum allocation algorithm for vehicle networking based on multi-agent deep reinforcement learning is proposed. The algorithm aims to maximize the network throughput under the constraints of vehicle communication delay and reliability, and uses the learning algorithm to improve the spectrum and power allocation strategy. Firstly, the implicit cooperative agent is trained by improving DQN model and EXP3 strategy. Secondly, the nonstationary problem caused by multi-agent concurrent learning is solved by using hysteretic Q-learning and concurrent experience replay trajectory. The simulation results show that the average successful delivery rate of the payload of the proposed algorithm can reach 95.89%, which is 16.48% higher than the random baseline algorithm. It can quickly obtain the approximate optimal solution, and has significant advantages in reducing the signaling overhead of the Internet of vehicles communication system.
In view of the accuracy of existing 3D object detection algorithms based on Pseudo-LiDar is far lower than that based on real LiDAR (Light Detection and ranging), this paper studies the reconstruction of Pseudo-LiDar and proposes a 3D object detection algorithm suitable for Pseudo-LiDar. Considering that the Pseudo-LiDAR obtained by image depth is dense and gradually sparse along the increase of depth, a depth related Pseudo-LiDAR sparsification method is proposed to reduce the subsequent calculation amount while retaining more useful Pseudo-LiDAR in the middle and long distance, so as to realize the reconstruction of Pseudo-LiDAR. Furthermore, a 3D object detection algorithm based on object feature distribution convergence under the guidance of LiDar point cloud and semantic association is proposed. During network training, a laser point cloud branch is introduced to guide the generation of Pseudo-LiDAR object features, so that the generated Pseudo-LiDar object feature distribution converges to the feature distribution of laser point cloud object, thereby correcting the detection error caused by the difference between the two data sources. Aiming at the insufficient semantic association between Pseudo-LiDar in the 3D candidate bounding-box obtained by RPN (Region Proposal Network) network, an attention perception module is designed to embed the semantic association between points through the attention mechanism in the feature representation of Pseudo-LiDar, so as to improve the accuracy of 3D object detection. The experimental results on KITTI 3D object detection dataset show when the existing 3D object detection network adopts the reconstructed Pseudo-LiDar, the detection accuracy is improved by 2.61%. Furthermore, the proposed 3D object detection network with the feature distribution convergence and semantic association improves the accuracy by 0.57%. Compared with other excellent methods, it also improves the detection accuracy.
Hyperspectral images (HSIs) have high spectral resolution and rich spectral information, which can obtain the physical and chemical information of the target of interest by using a large number of narrow-band waves. HSIs can effectively distinguish different substances by corresponding spectral features, and complete the task of target detection. However, the problem of target and background confusion caused by limited samples, a small amount of prior information, high dimensional similar background, and differences between different classes make hyperspectral target detection (HTD) still face challenges. To this end, we propose a region-guided and dual-attention discriminative learning network (RADN) for HTD to solve the problem of intra-class differences and inter-class similarities under a few samples. It can reduce the computational complexity caused by high-dimensional redundant features and improve detection accuracy. In this paper, we introduce the empirical region-guided network for training. We employ the spectrally constrained unsupervised clustering network to determine the network input. To selectively focus on salient features and regions of interest, we add a dual-channel attention mechanism in the generator and discriminator to assist in the estimation of complex background distributions; We introduce an inter-class spectral prior loss function in the network and further reduce the interference of high-dimensional complex background and spectral changes to the target. Experimental results and analysis show that RADN outperforms existing state-of-the-art algorithms on different datasets.
Keypoint detection of human body is a hot research area in computer vision. At present there exist some problems for keypoint detection in gymnastics actions, such as insufficient detection accuracy and lack of capability to detect detailed body parts. In order to improve the detection accuracy, this paper proposes a multi-resolution network that has a larger receptive field in the shallow layers and can utilize high-resolution channel to enhance the extraction of detailed features. To achieve the detection of keypoints of hands and feet, an incremental learning network is designed. The network fuses the shallow features of the multi-resolution network and computes deep features using a gymnastics actions self-built dataset, so that the detection ability of keypoints on hands and feet is improved. Finally, the output results of the two sub-networks are concated. Computer simulations demonstrate that the multi-resolution network achieves an accuracy rate of 94.4% on the COCO2017 keypoint detection dataset, and the incremental learning network can accurately detect keypoints of detailed body parts with fewer training data.
Unsupervised person re-identification aims to match query pedestrian images with images in the gallery without the need for identity labels. Currently, mainstream unsupervised person re-identification methods typically utilize clustering algorithms to generate pseudo-labels, which are subsequently exploited to train deep neural networks. However, due to the model’s inferior representation ability at early stages and the limitations of the clustering algorithms, a vast of noise is inevitably introduced into the pseudo-labels, which seriously misleads the training process and impedes the model performance. In this paper, we propose a novel pseudo-label regularization loss (PLRL) to remedy the detrimental effect of pseudo-label noises. Concretely, firstly, this paper proposes a clustering-guided attention mechanism (CGA) to estimate the confidence of pseudo-labels based on the semantic relevance between pseudo-labels and clustering centers. The CGA score is able to identify noisy labels and assign more weight to correct labels, which effectively reduces the influence of pseudo-label noise in the overall loss function. Meanwhile, for the sake of fully utilizing the discriminative power of pseudo-labels, this paper performs soft sample mining using pseudo-labels, which constructs positive and negative sample pairs in mini-batches and calculates a continuous weight score for each pair. By incorporating the confidence of pseudo-labels and the similarity of samples into the contrastive loss, the newly designed pseudo-label regularization loss can effectively alleviate the influence of pseudo-label noise in the training process, thereby improving the accuracy and robustness of the model. Experiments and ablation studies on multiple public datasets demonstrate its effectiveness and superiority, with the mAP on Market1501, DukeMTMC-reID, and MSMT17 datasets reaching 85.9%, 75.1%, and 29.3%, respectively.
In the digital age, the generation and accumulation of massive amounts of data are exploding, driving the demand for storage capacity sharply. Conventional magnetic recording disks (CMR) are considered the preferred solution for massive data storage due to their high capacity and low cost. However, the presence of the superparamagnetic effect (SPE) limits further improvements in CMR disk density. To overcome this limitation, shingled magnetic recording (SMR) technology was developed. Based on the conventional hard disc architecture, this technology dramatically increases disk areal density by overlapping tracks. However, SMR disks produce unpredictable write amplification effects when performing random writes, which can severely impact I/O performance. To solve this problem, the industry then proposed interlaced magnetic recording (IMR) technology, which uses an optimized track layout and heat-assisted magnetic recording technology to effectively balance storage capacity and performance. In this paper, we first introduce the technical principles, disk types, and applications of SMR and IMR in detail, and quantitatively analyzes the write interference problem that affects their I/O performance. It then focuses on solutions to optimize their I/O performance at the device level, analyzing and summarizing the advantages, disadvantages, and optimization objectives of different strategies in each type of solution. An overview of device-oriented system-level and application-level optimization solutions, such as file systems, redundant array of independent disk (RAID) technologies, and databases, is then provided. Finally, possible research directions for optimizing the performance of SMR and IMR disks in the future are discussed.
Due to the slowdown of transistor scaling, it has become increasingly difficult to enhance the performance of a single GPU (Graphics Processing Units). Therefore, multi-GPU systems have become the main means to improve the performance of GPU systems. However, due to the constraints of off-chip physical design, the bandwidth imbalance between processors in multi-GPU systems leads to non-uniform memory access (NUMA) problems, which seriously affects the performance of multi-GPU systems. In order to reduce the performance loss caused by non-uniform memory access, this paper first analyzes the causes of non-uniform memory access and compares existing solutions for non-uniform memory access. For non-uniform memory access with different dimensions, this paper summarizes optimization solutions for non-uniform memory access from two directions: reducing remote access traffic and improving remote access performance. Finally, combining the advantages and disadvantages of these solutions, this paper proposes the future development direction of non-uniform memory access optimization for multi-GPU systems.