Email alert
2019 Vol. 41, No. 9
Display Method:
2019, 41(9): 2033-2038.
doi: 10.11999/JEIT180919
Abstract:
In Cooperative Jamming (CJ) system, the Power Amplifier (PA) in the jamming transmitter works in nonlinear region, which results in a large number of nonlinear components in the Self-Interference (SI) signal received by the near-end receiver. To solve the problem of nonlinear distortion suppression, a nonlinear model is established at the receiver. Then, the reconstructed nonlinear signal based on the estimated parameters is subtracted from the received signal to suppress the nonlinear interference in CJ. Simulation and experimental results indicate that the nonlinear suppression scheme proposed in this paper can further suppress the nonlinear interference under the residual frequency offset in CJ, and verify the effectiveness and feasibility of the proposed scheme.
In Cooperative Jamming (CJ) system, the Power Amplifier (PA) in the jamming transmitter works in nonlinear region, which results in a large number of nonlinear components in the Self-Interference (SI) signal received by the near-end receiver. To solve the problem of nonlinear distortion suppression, a nonlinear model is established at the receiver. Then, the reconstructed nonlinear signal based on the estimated parameters is subtracted from the received signal to suppress the nonlinear interference in CJ. Simulation and experimental results indicate that the nonlinear suppression scheme proposed in this paper can further suppress the nonlinear interference under the residual frequency offset in CJ, and verify the effectiveness and feasibility of the proposed scheme.
2019, 41(9): 2039-2046.
doi: 10.11999/JEIT180770
Abstract:
To satisfy the diversity of requirements for different network slices and realize dynamic allocation of wireless virtual resource, an algorithm for network slice joint user association and power allocation is proposed in Non-Orthogonal Multiple Access(NOMA) C-RAN. Firstly, by considering imperfect Channel State Information(CSI), a joint user association and power allocation algorithm is designed to maximize the average total throughput in C-RAN with the constraints of slice and user minimum required rate, outage probability and fronthaul capacity limits. Secondly, a joint user association and power allocation algorithm is designed according to the current slot by transforming the probabilistic mixed optimalization problem into a non-probabilistic optimalization problem and using Lyapunov optimization. Finally, for user association problem, a greedy algorithm is proposed to find a feasible suboptimal solution; The power allocation problem is transformed into a convex optimization problem by using successive convex approximation; Then a dual decomposition approach is exploited to obtain a power allocation strategy. Simulation results demonstrate that the proposed algorithm can effectively improve the average total throughput of system while guaranteeing the network slice and user requirement.
To satisfy the diversity of requirements for different network slices and realize dynamic allocation of wireless virtual resource, an algorithm for network slice joint user association and power allocation is proposed in Non-Orthogonal Multiple Access(NOMA) C-RAN. Firstly, by considering imperfect Channel State Information(CSI), a joint user association and power allocation algorithm is designed to maximize the average total throughput in C-RAN with the constraints of slice and user minimum required rate, outage probability and fronthaul capacity limits. Secondly, a joint user association and power allocation algorithm is designed according to the current slot by transforming the probabilistic mixed optimalization problem into a non-probabilistic optimalization problem and using Lyapunov optimization. Finally, for user association problem, a greedy algorithm is proposed to find a feasible suboptimal solution; The power allocation problem is transformed into a convex optimization problem by using successive convex approximation; Then a dual decomposition approach is exploited to obtain a power allocation strategy. Simulation results demonstrate that the proposed algorithm can effectively improve the average total throughput of system while guaranteeing the network slice and user requirement.
2019, 41(9): 2047-2054.
doi: 10.11999/JEIT181139
Abstract:
The Walsh-Hadamard transform can be used to solve binary domain error-containing equations, and the method can be used for blind identification of convolutional codes. However, when the number of system unknowns is large, the requirement of computer memory makes it difficult to apply this method to practice. Therefore, a convolutional code recognition method based on partitioned Walsh-Hadamard transform is proposed. By segmenting the high-dimensional coefficient vectors of the equations into two low-dimensional coefficient vectors, the problem of solving the high-dimensional equations by Walsh-Hadamard transformation is decomposed into the problem of solving the two low-dimensional equations, and it is proved that the combination of the solution vectors of the two low-dimensional equations is the solution of the high-dimensional equations. The algorithm reduces effectively the need for computer memory, and the simulation results verify the effectiveness of the proposed algorithm, and the algorithm has good error code adaptability.
The Walsh-Hadamard transform can be used to solve binary domain error-containing equations, and the method can be used for blind identification of convolutional codes. However, when the number of system unknowns is large, the requirement of computer memory makes it difficult to apply this method to practice. Therefore, a convolutional code recognition method based on partitioned Walsh-Hadamard transform is proposed. By segmenting the high-dimensional coefficient vectors of the equations into two low-dimensional coefficient vectors, the problem of solving the high-dimensional equations by Walsh-Hadamard transformation is decomposed into the problem of solving the two low-dimensional equations, and it is proved that the combination of the solution vectors of the two low-dimensional equations is the solution of the high-dimensional equations. The algorithm reduces effectively the need for computer memory, and the simulation results verify the effectiveness of the proposed algorithm, and the algorithm has good error code adaptability.
2019, 41(9): 2055-2062.
doi: 10.11999/JEIT181038
Abstract:
Considering the shortcomings of Differential Chaos Shift Keying (DCSK) transmission rate and to further improve the system error performance, a Short reference Orthogonal Multiuser DCSK(SOM-DCSK) communication system is proposed. The system shortens the reference signal to 1/P of each information bearing signal, and transmits multiple users by different delay times. Then the orthogonality of Hilbert transform is used in each information slot to achieve the purpose of transmitting a two-bit information signal. The Bite Error Rate (BER) formula of SOM-DCSK system in Additive White Gaussian Noise (AWGN) and Rayleigh fading channel is derived and experimentally simulated. The simulation results show that BER of the scheme has obvious improvement compared with the traditional multi-user system under the same conditions, and it has good practical value.
Considering the shortcomings of Differential Chaos Shift Keying (DCSK) transmission rate and to further improve the system error performance, a Short reference Orthogonal Multiuser DCSK(SOM-DCSK) communication system is proposed. The system shortens the reference signal to 1/P of each information bearing signal, and transmits multiple users by different delay times. Then the orthogonality of Hilbert transform is used in each information slot to achieve the purpose of transmitting a two-bit information signal. The Bite Error Rate (BER) formula of SOM-DCSK system in Additive White Gaussian Noise (AWGN) and Rayleigh fading channel is derived and experimentally simulated. The simulation results show that BER of the scheme has obvious improvement compared with the traditional multi-user system under the same conditions, and it has good practical value.
2019, 41(9): 2063-2070.
doi: 10.11999/JEIT180903
Abstract:
In order to overcome the shortcomings of low fault-tolerance and high computational complexity in the process of parameter identification such as code length and synchronization of Turbo code, a new algorithm based on Differential Likelihood Difference (DLD) at low Signal-to-Noise Ratio (SNR) is proposed. Firstly, the concept of DLD is defined, and the analysis matrix is constructed to identify the code length by using the characteristic that the DLD between two codes in Turbo frame terminal is positive ("+"); Secondly, a method based on the minimum error decision criterion to decide DLD "+" position is proposed to complete frame synchronization. From the engineering practice, the possible values of the number of registers are traversed to realize the recognition of the code rate, the number of registers and the interleaving length. Simulation results show that the proposed algorithm is effective in identifying parameters such as code length and frame synchronization, the position distribution of DLD "+" is consistent with the data structure characteristics of the analysis, and the threshold can effectively determine the position of DLD "+". At the same time, the algorithm has strong fault-tolerant performance. Under the condition of SNR –5 dB, the identification of code length, frame synchronization and other parameters can reach more than 90%, and the complexity of the algorithm is far less than the existing algorithms.
In order to overcome the shortcomings of low fault-tolerance and high computational complexity in the process of parameter identification such as code length and synchronization of Turbo code, a new algorithm based on Differential Likelihood Difference (DLD) at low Signal-to-Noise Ratio (SNR) is proposed. Firstly, the concept of DLD is defined, and the analysis matrix is constructed to identify the code length by using the characteristic that the DLD between two codes in Turbo frame terminal is positive ("+"); Secondly, a method based on the minimum error decision criterion to decide DLD "+" position is proposed to complete frame synchronization. From the engineering practice, the possible values of the number of registers are traversed to realize the recognition of the code rate, the number of registers and the interleaving length. Simulation results show that the proposed algorithm is effective in identifying parameters such as code length and frame synchronization, the position distribution of DLD "+" is consistent with the data structure characteristics of the analysis, and the threshold can effectively determine the position of DLD "+". At the same time, the algorithm has strong fault-tolerant performance. Under the condition of SNR –5 dB, the identification of code length, frame synchronization and other parameters can reach more than 90%, and the complexity of the algorithm is far less than the existing algorithms.
2019, 41(9): 2071-2078.
doi: 10.11999/JEIT180894
Abstract:
In order to solve the unreasonable virtual resource allocation caused by the dynamic change of service request and delay of information feedback in wireless virtualized network, a traffic-aware algorithm which exploits historical Service Function Chaining (SFC) queue information to predict future load state based on Long Short-Term Memory (LSTM) network is proposed. With the prediction results, the Virtual Network Function (VNF) deployment and the corresponding computing resource allocation problems are studied, and a VNFs’ deployment method based on Maximum and Minimum Ant Colony Algorithm (MMACA) is developed. On the premise of satisfying the minimum resource demand for future queue non-overflow, the on-demand allocation method is used to maximize the computing resource utilization. Simulation results show that the prediction model based on LSTM neural network in this paper obtains good prediction results and realizes online monitoring of the network. The Maximum and Minimum Ant Colony Algorithm based VNF deployment method reduces effectively the bit loss rate and the average end-to-end delay caused by overall VNFs’ scheduling at the same time.
In order to solve the unreasonable virtual resource allocation caused by the dynamic change of service request and delay of information feedback in wireless virtualized network, a traffic-aware algorithm which exploits historical Service Function Chaining (SFC) queue information to predict future load state based on Long Short-Term Memory (LSTM) network is proposed. With the prediction results, the Virtual Network Function (VNF) deployment and the corresponding computing resource allocation problems are studied, and a VNFs’ deployment method based on Maximum and Minimum Ant Colony Algorithm (MMACA) is developed. On the premise of satisfying the minimum resource demand for future queue non-overflow, the on-demand allocation method is used to maximize the computing resource utilization. Simulation results show that the prediction model based on LSTM neural network in this paper obtains good prediction results and realizes online monitoring of the network. The Maximum and Minimum Ant Colony Algorithm based VNF deployment method reduces effectively the bit loss rate and the average end-to-end delay caused by overall VNFs’ scheduling at the same time.
2019, 41(9): 2079-2086.
doi: 10.11999/JEIT180906
Abstract:
After the Virtual Network Function (VNF) in the 5G access network is deployed, the resource requirements are dynamically changed, resulting in the problem that the Physical Machine (PM) resource utilization in the network is too high or too low. To solve the above problem, the resource usage of PM in the network is divided into five different partitions, and a multi-priority VNF migration request queue scheduling model is proposed. Secondly, based on the model, a joint optimization model is established to minimize the VNF migration cost and minimize the network energy consumption. Finally, a multi-priority VNF migration cost and network energy joint optimization algorithm based on 5G access network is presented to solve the above model. The simulation results show that the algorithm can effectively improve the PM resources utilization, ensure the PM performance and balance the PM load while effectively realizing a compromise between VNF migration cost and network energy consumption.
After the Virtual Network Function (VNF) in the 5G access network is deployed, the resource requirements are dynamically changed, resulting in the problem that the Physical Machine (PM) resource utilization in the network is too high or too low. To solve the above problem, the resource usage of PM in the network is divided into five different partitions, and a multi-priority VNF migration request queue scheduling model is proposed. Secondly, based on the model, a joint optimization model is established to minimize the VNF migration cost and minimize the network energy consumption. Finally, a multi-priority VNF migration cost and network energy joint optimization algorithm based on 5G access network is presented to solve the above model. The simulation results show that the algorithm can effectively improve the PM resources utilization, ensure the PM performance and balance the PM load while effectively realizing a compromise between VNF migration cost and network energy consumption.
2019, 41(9): 2087-2094.
doi: 10.11999/JEIT180875
Abstract:
Surface Acoustic Wave (SAW) resonator measuring technology can be used in high temperature and high pressure, strong electromagnetic radiation and strong electromagnetic interference to realize wireless passive parameter detection. Based on the the non-stationary characteristics of the SAW signal, a kind of echo signal frequency estimation method, Digital Frequency Significant Place Tracking (DFSPT) method is put forward. Compared with the existing methods based on Fast Fourier Transform (FFT) and Singular Value Decomposition (SVD), the simulation results show that the method can determine the number of significant digits of digital frequency according to the difference of signal-to-noise ratio. Thus, it can increases stability and accuracy. The experiment of wireless SAW temperature sensor shows that the frequency estimation standard deviation of this method is small and the robustness is high.
Surface Acoustic Wave (SAW) resonator measuring technology can be used in high temperature and high pressure, strong electromagnetic radiation and strong electromagnetic interference to realize wireless passive parameter detection. Based on the the non-stationary characteristics of the SAW signal, a kind of echo signal frequency estimation method, Digital Frequency Significant Place Tracking (DFSPT) method is put forward. Compared with the existing methods based on Fast Fourier Transform (FFT) and Singular Value Decomposition (SVD), the simulation results show that the method can determine the number of significant digits of digital frequency according to the difference of signal-to-noise ratio. Thus, it can increases stability and accuracy. The experiment of wireless SAW temperature sensor shows that the frequency estimation standard deviation of this method is small and the robustness is high.
2019, 41(9): 2095-2102.
doi: 10.11999/JEIT180922
Abstract:
A broadband metasurface antenna array with wideband low Radar Cross Section (RCS) is proposed. Two kind metasurface antennas with nearly the same radiation performance are designed and fabricated in a chessboard configuration. For x -polarized incidence, the reflected energy from the two elements is dissipated based on destructive phase difference. For y -polarized incidence, the incident energy is absorbed by matching load. By this means, inherent wideband low RCS is achieved for both polarizations without redundant structures. The proposed antenna array is fabricated and measured. Simulated and measured results show that the working frequency band is 6.0～8.5 GHz. Meanwhile under x polarization the antenna monostatic RCS is reduced significantly 6 dB RCS reduction is achieved over the range of 6.2～10.5 GHz and the peak reduction is up to 21.07 dB. Under y polarization the antenna monostatic RCS is reduced over 3 dB RCS reduction in bandwidth. Both measured and simulated results verify the proposed antenna array is characterized with wideband low-RCS without degrading the radiation performance.
A broadband metasurface antenna array with wideband low Radar Cross Section (RCS) is proposed. Two kind metasurface antennas with nearly the same radiation performance are designed and fabricated in a chessboard configuration. For x -polarized incidence, the reflected energy from the two elements is dissipated based on destructive phase difference. For y -polarized incidence, the incident energy is absorbed by matching load. By this means, inherent wideband low RCS is achieved for both polarizations without redundant structures. The proposed antenna array is fabricated and measured. Simulated and measured results show that the working frequency band is 6.0～8.5 GHz. Meanwhile under x polarization the antenna monostatic RCS is reduced significantly 6 dB RCS reduction is achieved over the range of 6.2～10.5 GHz and the peak reduction is up to 21.07 dB. Under y polarization the antenna monostatic RCS is reduced over 3 dB RCS reduction in bandwidth. Both measured and simulated results verify the proposed antenna array is characterized with wideband low-RCS without degrading the radiation performance.
Measurement Error Correction of the Orthogonal Magnetic Loop Antenna for Lightning-direction Finding
2019, 41(9): 2103-2107.
doi: 10.11999/JEIT181016
Abstract:
The measurement accuracy for lightning direction finding by the Orthogonal Magnetic Loop Antenna (OMLA) is continuously improved, which results in the Angle Measurement Error (AME) caused by the OMLA machining error increasing. A theoretical model is established for the relationship between the machining error and AME of OMLA. With the compensation coefficient and equivalent non-orthogonal angle error, a AME correction method for OMLA is proposed. The AME of the conventional measurement way and the corrected measurement way are compared through three groups of data experimentally. The experimental results show that the AME by the corrected measurement way is significantly reduced by about 50%. Therefore, this correction method can help the OMLA with the same hardware condition to obtain higher measurement accuracy for lightning direction finding.
The measurement accuracy for lightning direction finding by the Orthogonal Magnetic Loop Antenna (OMLA) is continuously improved, which results in the Angle Measurement Error (AME) caused by the OMLA machining error increasing. A theoretical model is established for the relationship between the machining error and AME of OMLA. With the compensation coefficient and equivalent non-orthogonal angle error, a AME correction method for OMLA is proposed. The AME of the conventional measurement way and the corrected measurement way are compared through three groups of data experimentally. The experimental results show that the AME by the corrected measurement way is significantly reduced by about 50%. Therefore, this correction method can help the OMLA with the same hardware condition to obtain higher measurement accuracy for lightning direction finding.
2019, 41(9): 2108-2114.
doi: 10.11999/JEIT181083
Abstract:
Magnetic induction detection technology is a non-contact and non-invasive electrical impedance detection technology. Multi-frequency synchronous detection can simultaneously obtain the impedance information of the tested object at different frequencies. Firstly, the principle of multi-frequency synchronous excitation and detection of magnetic induction signal are studied. Five-frequency excitation signal is synthesized based on Walsh function. Secondly, the performance of synthesized multi-frequency synchronous detection is analyzed, and a synthesized multi-frequency magnetic induction signal synchronous detection system is designed. Finally, the detection experiments of NaCl solution with different conductivities are carried out by synthesizing five-frequency excitation signal and synchronous detection system. The results show that the measurement results of five main harmonics of synthesized five-frequency excitation signal have good linearity. It provides an excitation-detection method for multi-frequency synchronous detection of magnetic induction signal.
Magnetic induction detection technology is a non-contact and non-invasive electrical impedance detection technology. Multi-frequency synchronous detection can simultaneously obtain the impedance information of the tested object at different frequencies. Firstly, the principle of multi-frequency synchronous excitation and detection of magnetic induction signal are studied. Five-frequency excitation signal is synthesized based on Walsh function. Secondly, the performance of synthesized multi-frequency synchronous detection is analyzed, and a synthesized multi-frequency magnetic induction signal synchronous detection system is designed. Finally, the detection experiments of NaCl solution with different conductivities are carried out by synthesizing five-frequency excitation signal and synchronous detection system. The results show that the measurement results of five main harmonics of synthesized five-frequency excitation signal have good linearity. It provides an excitation-detection method for multi-frequency synchronous detection of magnetic induction signal.
2019, 41(9): 2115-2122.
doi: 10.11999/JEIT181131
Abstract:
End-fire array antenna is extremely suitable for forward-looking or backward-looking blind compensation of airborne radar due to its low wind resistance and high-gain characteristics, while the forward-looking or backward-looking placement of antenna can not avoid the problem of range-dependent clutter. In this paper, in view of the fact that the conventional Space-Time INterpolation Technique(STINT) can not be directly applied to end-fire array clutter compensation in range ambiguity situation, a novel method of end-fire array clutter compensation based on space-time interpolation is proposed based on the characteristics of clutter spectrum for end-fire array airborne radar. The method takes full account of the ambiguous clutter of each range gate and takes the arc corresponding to the main lobe of the long-range stationary clutter ridge as the interpolation reference subspace. Furthermore, it also refines the constrained object of moving target constraints, which achieves effective compensation for the non-stationary clutter of end-fire array in range ambiguity situation. Computer simulation results verify the effectiveness of the proposed method.
End-fire array antenna is extremely suitable for forward-looking or backward-looking blind compensation of airborne radar due to its low wind resistance and high-gain characteristics, while the forward-looking or backward-looking placement of antenna can not avoid the problem of range-dependent clutter. In this paper, in view of the fact that the conventional Space-Time INterpolation Technique(STINT) can not be directly applied to end-fire array clutter compensation in range ambiguity situation, a novel method of end-fire array clutter compensation based on space-time interpolation is proposed based on the characteristics of clutter spectrum for end-fire array airborne radar. The method takes full account of the ambiguous clutter of each range gate and takes the arc corresponding to the main lobe of the long-range stationary clutter ridge as the interpolation reference subspace. Furthermore, it also refines the constrained object of moving target constraints, which achieves effective compensation for the non-stationary clutter of end-fire array in range ambiguity situation. Computer simulation results verify the effectiveness of the proposed method.
2019, 41(9): 2123-2130.
doi: 10.11999/JEIT180904
Abstract:
For the passive radar based on LTE signal, the received signal contains direct-path and multipath clutters interference of multiple co-channel base station, and the traditional passive radar signal processing flow is improved, and the processing steps of co-channel base station interference are added. A blind source separation algorithm based on convolutive mixtures is proposed. The algorithm can suppress the clutters interference of co-channel base station. It is assumed that the mixing matrix is a vector linear time-invariant filter matrix. The mutual information is used as a cost function. By finding the gradient of mutual information, it is iterated by the steepest descent method. The separation criterion is to minimize the mutual information between the separated signals. The simulation results show that the proposed algorithm can effectively suppress the clutters interference of the LTE signal co-channel base station, and provide a basis for the subsequent clutters cancellation processing of the main base station.
For the passive radar based on LTE signal, the received signal contains direct-path and multipath clutters interference of multiple co-channel base station, and the traditional passive radar signal processing flow is improved, and the processing steps of co-channel base station interference are added. A blind source separation algorithm based on convolutive mixtures is proposed. The algorithm can suppress the clutters interference of co-channel base station. It is assumed that the mixing matrix is a vector linear time-invariant filter matrix. The mutual information is used as a cost function. By finding the gradient of mutual information, it is iterated by the steepest descent method. The separation criterion is to minimize the mutual information between the separated signals. The simulation results show that the proposed algorithm can effectively suppress the clutters interference of the LTE signal co-channel base station, and provide a basis for the subsequent clutters cancellation processing of the main base station.
2019, 41(9): 2137-2142.
doi: 10.11999/JEIT181090
Abstract:
The MUltiple SIgnal Classification (MUSIC) algorithm is a classical spatial spectrum estimation algorithm. Taking L-shaped array as an example, an improved 2D-MUSIC algorithm is proposed for the problem that 2D-MUSIC algorithm often fails to estimate accurately targets in close proximity among multiple targets when the signal-to-noise ratio is low.The algorithm identifies the target location through spectrum peak search by first performing conjugate recombination on the covariance matrix generated by the classical 2D-MUSIC algorithm, then calculating the mean of sum of square of the recombined one and the original one as the new matrix, whose corresponding noise subspace then weighted by applying appropriate coefficients to obtain a new noise subspace. The computer simulation results show that compared with the 2D-MUSIC algorithm, the improved algorithm performs well on DOA estimation for the targets in close proximity among multiple targets when the received signal has low signal-to-noise ratio, which improves the resolution of 2D-DOA estimation with L-shaped array, with better engineering application value.
The MUltiple SIgnal Classification (MUSIC) algorithm is a classical spatial spectrum estimation algorithm. Taking L-shaped array as an example, an improved 2D-MUSIC algorithm is proposed for the problem that 2D-MUSIC algorithm often fails to estimate accurately targets in close proximity among multiple targets when the signal-to-noise ratio is low.The algorithm identifies the target location through spectrum peak search by first performing conjugate recombination on the covariance matrix generated by the classical 2D-MUSIC algorithm, then calculating the mean of sum of square of the recombined one and the original one as the new matrix, whose corresponding noise subspace then weighted by applying appropriate coefficients to obtain a new noise subspace. The computer simulation results show that compared with the 2D-MUSIC algorithm, the improved algorithm performs well on DOA estimation for the targets in close proximity among multiple targets when the received signal has low signal-to-noise ratio, which improves the resolution of 2D-DOA estimation with L-shaped array, with better engineering application value.
2019, 41(9): 2143-2150.
doi: 10.11999/JEIT181107
Abstract:
For the fact that exsiting MIMO transmit beampattern design methods suffer from huge computational burden, a novel MIMO transmit beampattern design method based on atomic norm is proposed. According to the signal model of atomic norm, firstly a multi-rank transmit beamformer and a set of orthogonal signals are selected. Then the transmit beampattern matching design problem is formulated into an atomic norm minimization problem. The multi-rank transmit beamformer is achieved by Vandermonde decomposition method of positive semidefinite Toeplitz matrix, which is attained by the solution of the atomic norm minimization problem with Semi-Definite Programming (SDP). Finally, the transmit waveforms can be acquired from the resulting multi-rank transmit beamformer and existing orthogonal waveforms. The theoretical analysis and simulation results verify that the proposed method satisfies the uniform element power constraint and low Peak to Average Power Ratio (PAPR). Simultaneously, compared with current methods, the proposed method has lower computational burden and comparable matching performance.
For the fact that exsiting MIMO transmit beampattern design methods suffer from huge computational burden, a novel MIMO transmit beampattern design method based on atomic norm is proposed. According to the signal model of atomic norm, firstly a multi-rank transmit beamformer and a set of orthogonal signals are selected. Then the transmit beampattern matching design problem is formulated into an atomic norm minimization problem. The multi-rank transmit beamformer is achieved by Vandermonde decomposition method of positive semidefinite Toeplitz matrix, which is attained by the solution of the atomic norm minimization problem with Semi-Definite Programming (SDP). Finally, the transmit waveforms can be acquired from the resulting multi-rank transmit beamformer and existing orthogonal waveforms. The theoretical analysis and simulation results verify that the proposed method satisfies the uniform element power constraint and low Peak to Average Power Ratio (PAPR). Simultaneously, compared with current methods, the proposed method has lower computational burden and comparable matching performance.
2019, 41(9): 2131-2136.
doi: 10.11999/JEIT181091
Abstract:
Range sidelobes may lead to weak targets masked by strong targets and false alarm. This paper proposes a sequential optimization method for the sidelobe suppression of cognitive radar. First, the region to detect is divided according to range cell. Second, the transmit waveform and receive filter are optimized jointly based on the principle of minimum mean square error against one range cell. The optimized transmit and receive systems are used in Radar Cross Section (RCS) estimation for the scatter in the current range cell. The above process is carried out in each range cell in the scene sequentially. The acquired RCS estimate is used in the sidelobe suprresion for the following range cells. The RCS estimation for all the range cells in the scene is obtained in a bootstrapping way successively and updated circularly. The proposed method forms a closed loop detection system. The transmitting and receiving systems are adjusted according to the feedback scene information in real time. The sensing ability about the environment can be enhanced. The detection performance and robustness against noise can be improved. The efficiency and validity are verified by the simulation results.
Range sidelobes may lead to weak targets masked by strong targets and false alarm. This paper proposes a sequential optimization method for the sidelobe suppression of cognitive radar. First, the region to detect is divided according to range cell. Second, the transmit waveform and receive filter are optimized jointly based on the principle of minimum mean square error against one range cell. The optimized transmit and receive systems are used in Radar Cross Section (RCS) estimation for the scatter in the current range cell. The above process is carried out in each range cell in the scene sequentially. The acquired RCS estimate is used in the sidelobe suprresion for the following range cells. The RCS estimation for all the range cells in the scene is obtained in a bootstrapping way successively and updated circularly. The proposed method forms a closed loop detection system. The transmitting and receiving systems are adjusted according to the feedback scene information in real time. The sensing ability about the environment can be enhanced. The detection performance and robustness against noise can be improved. The efficiency and validity are verified by the simulation results.
2019, 41(9): 2151-2155.
doi: 10.11999/JEIT180884
Abstract:
Based on the theory of Ding - generalized circle, a new class of generalized cyclotomic sequences of\begin{document}$ 2{p^m}$\end{document} ![]()
![]()
\begin{document}$ p$\end{document} ![]()
![]()
Based on the theory of Ding - generalized circle, a new class of generalized cyclotomic sequences of
2019, 41(9): 2156-2163.
doi: 10.11999/JEIT180849
Abstract:
The security of lightweight block cipher Simeck against integral attack is evaluated in this paper. First, a 16-round and a 20-round high-order integral distinguisher of Simeck48 and Simeck64 are constructed by decrypting the existed integral distinguisher forward. Then, combined with the meet-in-the-middle strategy and subkey relationship, the integral attacks on 24-round Simeck48 and 29-round Simeck64 are first proposed utilizing the equivalent-subkey and partial-sum technologies based on the new integral distinguishers. The data, time and memory complexity of attacking 24-round Simeck48 are 246, 295 and 282.52 while the data, time and memory complexity of attacking 29-round Simeck64 are 263, 2127.3 and 2109.02. These new attacks improve greatly the results of the previous integral attack on Simeck. Compared with the known results of the integral attack on Simeck, the number of rounds of the integral attacks on Simeck48 and Simeck64 is increased by 3-round and 5-round, respectively.
The security of lightweight block cipher Simeck against integral attack is evaluated in this paper. First, a 16-round and a 20-round high-order integral distinguisher of Simeck48 and Simeck64 are constructed by decrypting the existed integral distinguisher forward. Then, combined with the meet-in-the-middle strategy and subkey relationship, the integral attacks on 24-round Simeck48 and 29-round Simeck64 are first proposed utilizing the equivalent-subkey and partial-sum technologies based on the new integral distinguishers. The data, time and memory complexity of attacking 24-round Simeck48 are 246, 295 and 282.52 while the data, time and memory complexity of attacking 29-round Simeck64 are 263, 2127.3 and 2109.02. These new attacks improve greatly the results of the previous integral attack on Simeck. Compared with the known results of the integral attack on Simeck, the number of rounds of the integral attacks on Simeck48 and Simeck64 is increased by 3-round and 5-round, respectively.
2019, 41(9): 2164-2171.
doi: 10.11999/JEIT180905
Abstract:
In order to defend against Side-Channel Attacks (SCA) in Network Slicing (NS), the existing defense methods based on dynamic migration have the problem that the conditions for sharing of physical resources between different virtual nodes are not strict enough, a virtual node migration method is proposed for sensing side-channel risk. According to the characteristics of SCA, the entropy method is used to evaluate the side-channel risks and migrate the virtual node from a server with large deviation from average risk. The Markov decision process is used to describe the migration of virtual nodes for network slicing, and the Sarsa learning algorithm is used to solve the optimal migration scheme. The simulation results show that this method can separates malicious network slice instances from other target network slice instances to achieve the purpose of defense side channel attacks.
In order to defend against Side-Channel Attacks (SCA) in Network Slicing (NS), the existing defense methods based on dynamic migration have the problem that the conditions for sharing of physical resources between different virtual nodes are not strict enough, a virtual node migration method is proposed for sensing side-channel risk. According to the characteristics of SCA, the entropy method is used to evaluate the side-channel risks and migrate the virtual node from a server with large deviation from average risk. The Markov decision process is used to describe the migration of virtual nodes for network slicing, and the Sarsa learning algorithm is used to solve the optimal migration scheme. The simulation results show that this method can separates malicious network slice instances from other target network slice instances to achieve the purpose of defense side channel attacks.
2019, 41(9): 2172-2179.
doi: 10.11999/JEIT181117
Abstract:
Network multi-alarm information fusion processing is one of the most important methods to implement effectively network dynamic threat analysis. Focusing on this, a mechanism for dynamic threat tracking and quantitative analysis by using network system multi-alarm information is proposed. Firstly, the attack graph theory is used to construct the system dynamic threat attribute attack graph. Secondly, based on the privilege escalation principle, Antecedent Predictive Algorithm(APA), the Consequent Predictive Algorithm(CPA) and the Comprehensive Alarm Information Inference Algorithm(CAIIA) are designed to integrate the multi-alarm information fusion and do threat analysis. Then, the network dynamic threat tracking graph is generated to visualize the threat change situation. Finally, the effectiveness of the mechanism and algorithm is validates through experiments.
Network multi-alarm information fusion processing is one of the most important methods to implement effectively network dynamic threat analysis. Focusing on this, a mechanism for dynamic threat tracking and quantitative analysis by using network system multi-alarm information is proposed. Firstly, the attack graph theory is used to construct the system dynamic threat attribute attack graph. Secondly, based on the privilege escalation principle, Antecedent Predictive Algorithm(APA), the Consequent Predictive Algorithm(CPA) and the Comprehensive Alarm Information Inference Algorithm(CAIIA) are designed to integrate the multi-alarm information fusion and do threat analysis. Then, the network dynamic threat tracking graph is generated to visualize the threat change situation. Finally, the effectiveness of the mechanism and algorithm is validates through experiments.
2019, 41(9): 2180-2186.
doi: 10.11999/JEIT181103
Abstract:
Most existing searchable encryption schemes only support the search for keyword sets, and the data users can not quickly identify the file keyword information returned by the server. Meanwhile, considering the server has strong computing power, it may judge keyword information from single keywords and the identity of the data consumer is not verified. In this paper, the data user and data owner are delegated server to verify whether the data ueer is a legitimate user; if legal, the delegated server can detect the validity of the return ciphertext with data user. The data user uses the server public key, keywords and pseudo-keywords to generate trapdoor, in order to ensure the indistinguishable of the keywords, a delegated multi-keyword searchable encryption scheme is designed, which is resistant to keyword guessing of data user authentication. Meanwhile, when the data owner encrypts, the public key of the cloud server, the delegated server, and the data user can be used to prevent collusion attacks. In the random oracle model the security of the proposed scheme is proved. The experiment results show that the scheme is efficient under the multi-keyword environment.
Most existing searchable encryption schemes only support the search for keyword sets, and the data users can not quickly identify the file keyword information returned by the server. Meanwhile, considering the server has strong computing power, it may judge keyword information from single keywords and the identity of the data consumer is not verified. In this paper, the data user and data owner are delegated server to verify whether the data ueer is a legitimate user; if legal, the delegated server can detect the validity of the return ciphertext with data user. The data user uses the server public key, keywords and pseudo-keywords to generate trapdoor, in order to ensure the indistinguishable of the keywords, a delegated multi-keyword searchable encryption scheme is designed, which is resistant to keyword guessing of data user authentication. Meanwhile, when the data owner encrypts, the public key of the cloud server, the delegated server, and the data user can be used to prevent collusion attacks. In the random oracle model the security of the proposed scheme is proved. The experiment results show that the scheme is efficient under the multi-keyword environment.
2019, 41(9): 2187-2193.
doi: 10.11999/JEIT181014
Abstract:
The basis of the identification of network security situation element is to perform the feature extraction of situation data effectively. Considering the problem that the Back Propagation(BP) neural networks have excessive dependence on data labels when it has a learning of massive security situation information data, a network security situation element identification method is proposed, which combines deep stack encoder and BP algorithm. It trains the network layer by layer through unsupervised learning algorithm. On this basis the deep track encoder by stacking can be obtained. The unsupervised training of the network is realized when using the encoder to extract the characteristic of the data sets. It is verified by simulation experiments that the method can improve the performance and accuracy of situational awareness effectively.
The basis of the identification of network security situation element is to perform the feature extraction of situation data effectively. Considering the problem that the Back Propagation(BP) neural networks have excessive dependence on data labels when it has a learning of massive security situation information data, a network security situation element identification method is proposed, which combines deep stack encoder and BP algorithm. It trains the network layer by layer through unsupervised learning algorithm. On this basis the deep track encoder by stacking can be obtained. The unsupervised training of the network is realized when using the encoder to extract the characteristic of the data sets. It is verified by simulation experiments that the method can improve the performance and accuracy of situational awareness effectively.
2019, 41(9): 2194-2200.
doi: 10.11999/JEIT180916
Abstract:
A method for visualizing the weights of a reconstructed model is proposed to analyze a deep convolutional network works. Firstly, a specific input is used in the original neural network during the forward propagation to get the prior information for model reconstruction. Then some of the structure of the original network is changed for further parameter calculation. After that, the parameters of the reconstructed model are calculated with a group of orthogonal vectors. Finally, the parameters are put into a special order to make them visualized. Experimental results show that the model reconstructed with the proposed method is totally equivalent to the original model during the forward propagation in the classification process. The feature of the weights of the reconstructed model can be observed clearly and the principle of the neural network can be analyzed with the feature.
A method for visualizing the weights of a reconstructed model is proposed to analyze a deep convolutional network works. Firstly, a specific input is used in the original neural network during the forward propagation to get the prior information for model reconstruction. Then some of the structure of the original network is changed for further parameter calculation. After that, the parameters of the reconstructed model are calculated with a group of orthogonal vectors. Finally, the parameters are put into a special order to make them visualized. Experimental results show that the model reconstructed with the proposed method is totally equivalent to the original model during the forward propagation in the classification process. The feature of the weights of the reconstructed model can be observed clearly and the principle of the neural network can be analyzed with the feature.
2019, 41(9): 2201-2208.
doi: 10.11999/JEIT181129
Abstract:
In order to solve the problem of sensor scheduling in the multi-task scenario, a multi-sensor scheduling method for target cooperative detection and tracking is proposed. Firstly, the sensor scheduling model is built based on the Partially Observable Markov Decision Process (POMDP) and an objective function is designed based on Posterior Carmér-Rao Lower Bound (PCRLB). Then, considering sensor switching time and the change of target number, the randomly distributed particles are used to calculate the detection probability of new target, and the sensor scheduling methods are given for the situations with fixed target number and time-varying target number. At last, to meet the real-time requirement of online scheduling, an Adaptive Multi-swarm Cooperative Differential Evolution (AMCDE) algorithm is used to solve the sensor scheduling scheme. Simulation results show that the method can effectively deal with multi-task scenarios and realize reasonable scheduling of multi-sensor resources.
In order to solve the problem of sensor scheduling in the multi-task scenario, a multi-sensor scheduling method for target cooperative detection and tracking is proposed. Firstly, the sensor scheduling model is built based on the Partially Observable Markov Decision Process (POMDP) and an objective function is designed based on Posterior Carmér-Rao Lower Bound (PCRLB). Then, considering sensor switching time and the change of target number, the randomly distributed particles are used to calculate the detection probability of new target, and the sensor scheduling methods are given for the situations with fixed target number and time-varying target number. At last, to meet the real-time requirement of online scheduling, an Adaptive Multi-swarm Cooperative Differential Evolution (AMCDE) algorithm is used to solve the sensor scheduling scheme. Simulation results show that the method can effectively deal with multi-task scenarios and realize reasonable scheduling of multi-sensor resources.
2019, 41(9): 2209-2216.
doi: 10.11999/JEIT180748
Abstract:
In order to overcome the accumulation error in Micro-Electro-Mechanical System-Inertial Navigation System (MEMS-INS) and the jump error in iBeacon fingerprint positioning, an iBencon/MEMS-INS data fusion location algorithm based on Unscented Kalman Filter (UKF) is proposed. The new algorithm solves the distance between the iBeacon anchor and the locating target. The solution of attitude matrix and position are obtained respectively by using accelerometer and gyroscope data. Bluetooth anchor position vector, the carrier speed error and other information constitute state variables. Inertial navigation location and bluetooth system distance information constitute measure variables. Based on state variables and measure variables, the UKF is designed to realize iBencon/MEMS-INS data fusion indoor positioning. The experimental results show that the proposed algorithm can effectively solve the problem of the large accumulation error of INS and the jump error of iBeacon fingerprint positioning, and this algorithm can realize 1.5 m positioning accuracy.
In order to overcome the accumulation error in Micro-Electro-Mechanical System-Inertial Navigation System (MEMS-INS) and the jump error in iBeacon fingerprint positioning, an iBencon/MEMS-INS data fusion location algorithm based on Unscented Kalman Filter (UKF) is proposed. The new algorithm solves the distance between the iBeacon anchor and the locating target. The solution of attitude matrix and position are obtained respectively by using accelerometer and gyroscope data. Bluetooth anchor position vector, the carrier speed error and other information constitute state variables. Inertial navigation location and bluetooth system distance information constitute measure variables. Based on state variables and measure variables, the UKF is designed to realize iBencon/MEMS-INS data fusion indoor positioning. The experimental results show that the proposed algorithm can effectively solve the problem of the large accumulation error of INS and the jump error of iBeacon fingerprint positioning, and this algorithm can realize 1.5 m positioning accuracy.
2019, 41(9): 2217-2223.
doi: 10.11999/JEIT180931
Abstract:
In order to improve the quality of reconstruction image by Block Compressed Sensing (BCS), a Total Variation Iterative Threshold regularization image reconstruction algorithm (BCS-TVIT) is proposed. Combining the properties of local smoothing and bounded variation of the image, BCS-TVIT uses the minimization l0 norm and total variation to construct the objective function. To solve the problem that l0 norm term and the block measurement constraint can not be optimized directly, the iterative threshold method is used to minimize the l0 norm of the reconstructed image, and the convex set projection is employed to guarantee the block measurement constraint condition. Experiments show that BCS-TVIT has better performance than BCS-SPL in PSNR by 2 dB. Meanwhile, BCS-TVIT can eliminate the " bright spot” effect of BCS-SPL, having better visual effect. Comparing with the minimum total variation, the proposed algorithm increases PSNR by 1 dB, and the reconstruction time is reduced by two orders of magnitude.
In order to improve the quality of reconstruction image by Block Compressed Sensing (BCS), a Total Variation Iterative Threshold regularization image reconstruction algorithm (BCS-TVIT) is proposed. Combining the properties of local smoothing and bounded variation of the image, BCS-TVIT uses the minimization l0 norm and total variation to construct the objective function. To solve the problem that l0 norm term and the block measurement constraint can not be optimized directly, the iterative threshold method is used to minimize the l0 norm of the reconstructed image, and the convex set projection is employed to guarantee the block measurement constraint condition. Experiments show that BCS-TVIT has better performance than BCS-SPL in PSNR by 2 dB. Meanwhile, BCS-TVIT can eliminate the " bright spot” effect of BCS-SPL, having better visual effect. Comparing with the minimum total variation, the proposed algorithm increases PSNR by 1 dB, and the reconstruction time is reduced by two orders of magnitude.
2019, 41(9): 2224-2230.
doi: 10.11999/JEIT180826
Abstract:
Currently, many saliency-detection methods focus on 2D-image. But, these methods cannot be applied in RGB-D image. Based on this situation, new methods which are suitable for RGB-D image are needed. This paper presents a novel algorithm based on Extreme Learning Machine(ELM), feature-extraction and depth-detection. Firstly, feature-extraction is used for getting a feature, which contains 4-scale superpixels and 4096 dimensions. Secondly, according to the 4-sacle superpixels, the RGB, LAB and LBP feature of RGB image are computed, and LBE feature of depth image. Thirdly, weak salient map with LBE and dark-channel features are computed, and the foreground objects is strengthened in every circle. Fourthly, according to weak salient map, both foreground seeds and background seeds are chosen, and then, put these seeds into ELM to compute the first stage salient map. Finally, depth-detection and graph-cut are used for optimizing the first stage salient map and getting the second stage salient map.
Currently, many saliency-detection methods focus on 2D-image. But, these methods cannot be applied in RGB-D image. Based on this situation, new methods which are suitable for RGB-D image are needed. This paper presents a novel algorithm based on Extreme Learning Machine(ELM), feature-extraction and depth-detection. Firstly, feature-extraction is used for getting a feature, which contains 4-scale superpixels and 4096 dimensions. Secondly, according to the 4-sacle superpixels, the RGB, LAB and LBP feature of RGB image are computed, and LBE feature of depth image. Thirdly, weak salient map with LBE and dark-channel features are computed, and the foreground objects is strengthened in every circle. Fourthly, according to weak salient map, both foreground seeds and background seeds are chosen, and then, put these seeds into ELM to compute the first stage salient map. Finally, depth-detection and graph-cut are used for optimizing the first stage salient map and getting the second stage salient map.
2019, 41(9): 2231-2238.
doi: 10.11999/JEIT180738
Abstract:
To solve the incompleteness of the salient region obtained by the existing saliency detection method in the frequency domain, a frequency saliency detection method of multi-scale analysis is proposed. Firstly, the quaternion hypercomplex is constructed by the input image feature channels. Then, the multi-scale decomposition of the quaternion amplitude spectrum is performed by wavelet transform, and the multi-scale visual saliency map is calculated. Finally, the better saliency map is fused based on the evaluation function, and central bias is used to generate the final visual saliency map. The experimental results show that the proposed method can effectively suppress the background interference, find significant target quickly and accurately, and have high detection accuracy.
To solve the incompleteness of the salient region obtained by the existing saliency detection method in the frequency domain, a frequency saliency detection method of multi-scale analysis is proposed. Firstly, the quaternion hypercomplex is constructed by the input image feature channels. Then, the multi-scale decomposition of the quaternion amplitude spectrum is performed by wavelet transform, and the multi-scale visual saliency map is calculated. Finally, the better saliency map is fused based on the evaluation function, and central bias is used to generate the final visual saliency map. The experimental results show that the proposed method can effectively suppress the background interference, find significant target quickly and accurately, and have high detection accuracy.
2019, 41(9): 2239-2246.
doi: 10.11999/JEIT180740
Abstract:
In order to improve the accuracy rate of person re-identification, a pedestrian attribute hierarchy recognition neural network is proposed based on attention model. Compared with the existing algorithms, the model has the following three advantages. Firstly, the attention model is used in this paper to identify the pedestrian attributes, and to extract of pedestrian attribute information and degree of significance. Secondly, the attention model in used in this paper to classify the attributes according to the significance of the pedestrian attributes and the amount of informationcontained. Thirdly, this paper analyzes the correlation between attributes, and adjusts the next level identification strategy according to the recognition results of the upper level. It can improve the recognition accuracy of small target attributes, and the accuracy of pedestrian recognition is improved. The experimental results show that the proposed model can effectively improve the first accuracy rate (rank-1) of person re-identification compared with the existing methods. On the Market1501 dataset, the first accuracy rate is 93.1%, and the first accuracy rate is 81.7% on the DukeMTMC dataset.
In order to improve the accuracy rate of person re-identification, a pedestrian attribute hierarchy recognition neural network is proposed based on attention model. Compared with the existing algorithms, the model has the following three advantages. Firstly, the attention model is used in this paper to identify the pedestrian attributes, and to extract of pedestrian attribute information and degree of significance. Secondly, the attention model in used in this paper to classify the attributes according to the significance of the pedestrian attributes and the amount of informationcontained. Thirdly, this paper analyzes the correlation between attributes, and adjusts the next level identification strategy according to the recognition results of the upper level. It can improve the recognition accuracy of small target attributes, and the accuracy of pedestrian recognition is improved. The experimental results show that the proposed model can effectively improve the first accuracy rate (rank-1) of person re-identification compared with the existing methods. On the Market1501 dataset, the first accuracy rate is 93.1%, and the first accuracy rate is 81.7% on the DukeMTMC dataset.
2019, 41(9): 2247-2255.
doi: 10.11999/JEIT181018
Abstract:
In recent years, the Siamese networks has drawn great attention in visual tracking community due to its balanced accuracy and speed. However, most Siamese networks model are not updated, which causes tracking errors. In view of this deficiency, an algorithm based on the Siamese network with double templates is proposed. First, the base template R which is the initial frame target with stable response map score and the dynamic template T which is using the improved APCEs model update strategy to determine are kept. Then, the candidate targets region and the two template matching results are analyzed, meanwhile the result response maps are fused, which could ensure more accurate tracking results. The experimental results on the OTB2013 and OTB2015 datasets show that comparing with the 5 current mainstream tracking algorithms, the tracking accuracy and success rate of the proposed algorithm are superior. The proposed algorithm not only displays better tracking effects under the conditions of scale variation, in-plane rotation, out-of-plane rotation, occlusion, and illumination variation, but also achieves real-time tracking by a speed of 46 frames per second.
In recent years, the Siamese networks has drawn great attention in visual tracking community due to its balanced accuracy and speed. However, most Siamese networks model are not updated, which causes tracking errors. In view of this deficiency, an algorithm based on the Siamese network with double templates is proposed. First, the base template R which is the initial frame target with stable response map score and the dynamic template T which is using the improved APCEs model update strategy to determine are kept. Then, the candidate targets region and the two template matching results are analyzed, meanwhile the result response maps are fused, which could ensure more accurate tracking results. The experimental results on the OTB2013 and OTB2015 datasets show that comparing with the 5 current mainstream tracking algorithms, the tracking accuracy and success rate of the proposed algorithm are superior. The proposed algorithm not only displays better tracking effects under the conditions of scale variation, in-plane rotation, out-of-plane rotation, occlusion, and illumination variation, but also achieves real-time tracking by a speed of 46 frames per second.
2019, 41(9): 2256-2263.
doi: 10.11999/JEIT180978
Abstract:
In the field of computer vision, predicting human motion is very necessary for timely human–computer interaction and personnel tracking. In order to improve the performance of human–computer interaction and personnel tracking, an encoder-decoder model called Bi–directional Gated Recurrent Unit Encoder–Decoder (EBiGRU–D) based on Gated Recurrent Unit (GRU) is proposed to learn 3D human motion and give a prediction of motion over a period of time. EBiGRU–D is a deep Recurrent Neural Network (RNN) in which the encoder is a Bidirectional GRU (BiGRU) unit and the decoder is a unidirectional GRU unit. BiGRU allows raw data to be simultaneously input from both the forward and reverse directions and then encoded into a state vector, which is then sent to the decoder for decoding. BiGRU associates the current output with the state of the front and rear time, so that the output fully considers the characteristics of the time before and after, so that the prediction is more accurate. Experimental results on the human3.6m dataset demonstrate that EBiGRU–D not only improves greatly the error of 3D human motion prediction but also increases greatly the time for accurate prediction.
In the field of computer vision, predicting human motion is very necessary for timely human–computer interaction and personnel tracking. In order to improve the performance of human–computer interaction and personnel tracking, an encoder-decoder model called Bi–directional Gated Recurrent Unit Encoder–Decoder (EBiGRU–D) based on Gated Recurrent Unit (GRU) is proposed to learn 3D human motion and give a prediction of motion over a period of time. EBiGRU–D is a deep Recurrent Neural Network (RNN) in which the encoder is a Bidirectional GRU (BiGRU) unit and the decoder is a unidirectional GRU unit. BiGRU allows raw data to be simultaneously input from both the forward and reverse directions and then encoded into a state vector, which is then sent to the decoder for decoding. BiGRU associates the current output with the state of the front and rear time, so that the output fully considers the characteristics of the time before and after, so that the prediction is more accurate. Experimental results on the human3.6m dataset demonstrate that EBiGRU–D not only improves greatly the error of 3D human motion prediction but also increases greatly the time for accurate prediction.
2019, 41(9): 2264-2272.
doi: 10.11999/JEIT180900
Abstract:
ElectroEncephaloGram (EEG) is regarded as a " gold standard” of fatigue detection and drivers’ vigilance states can be detected through the analysis of EEG signals. However, due to the characteristics of non-linear, non-stationary and low spatial resolution of EEG signals, traditional machine learning methods still have the disadvantages of low recognition rate and complicated feature extraction operations in EEG-based fatigue detection task. To tackle this problem, a fatigue detection method with transfer learning based on the Electrode-Frequency Distribution Maps (EFDMs) of EEG signals is proposed. A deep convolutional neural network is designed and pre-trained with SEED dataset, and then it is used for fatigue detection with transfer learning strategy. Experimental results show that the proposed convolutional neural network can automatically obtain vigilance related features from EFDMs, and achieve much better recognition results than traditional machine learning methods. Moreover, based on the transfer learning strategy, this model can also be used for other recognition tasks, which is helpful for promoting the application of EEG signals to the driver fatigue detection system.
ElectroEncephaloGram (EEG) is regarded as a " gold standard” of fatigue detection and drivers’ vigilance states can be detected through the analysis of EEG signals. However, due to the characteristics of non-linear, non-stationary and low spatial resolution of EEG signals, traditional machine learning methods still have the disadvantages of low recognition rate and complicated feature extraction operations in EEG-based fatigue detection task. To tackle this problem, a fatigue detection method with transfer learning based on the Electrode-Frequency Distribution Maps (EFDMs) of EEG signals is proposed. A deep convolutional neural network is designed and pre-trained with SEED dataset, and then it is used for fatigue detection with transfer learning strategy. Experimental results show that the proposed convolutional neural network can automatically obtain vigilance related features from EFDMs, and achieve much better recognition results than traditional machine learning methods. Moreover, based on the transfer learning strategy, this model can also be used for other recognition tasks, which is helpful for promoting the application of EEG signals to the driver fatigue detection system.
2019, 41(9): 2273-2279.
doi: 10.11999/JEIT181073
Abstract:
In view of the nonlinear and stochastic characteristics of short-term traffic flow data, this article proposes a prediction model and algorithm based on hybrid Auto-Regressive Integrated Moving Average (ARIMA) and Genetic Particle Swarm Optimization Wavelet Neural Network (GPSOWNN) in order to improve its prediction accuracy and rate of convergence. In terms of model construction, the ARIMA model prediction value and the historical data of the first three moments with strong correlation with gray correlation coefficient greater than 0.6 are used as input of the Wavelet Neural Network(WNN), and the structure of the model is simplified considering both the stationary and non-stationary historical data. In terms of algorithm, by using the genetic particle swarm optimization algorithm to select optimally the initial values of the wavelet neural network, the results can speed up the convergence of network training under the condition that it is not easy to fall into local optimum. The experimental results show that the proposed model is superior to hybrid ARIMA and GPSOWNN in terms of prediction accuracy, the genetic particle swarm optimization algorithm is superior to the genetic algorithm optimization model in terms of convergence speed.
In view of the nonlinear and stochastic characteristics of short-term traffic flow data, this article proposes a prediction model and algorithm based on hybrid Auto-Regressive Integrated Moving Average (ARIMA) and Genetic Particle Swarm Optimization Wavelet Neural Network (GPSOWNN) in order to improve its prediction accuracy and rate of convergence. In terms of model construction, the ARIMA model prediction value and the historical data of the first three moments with strong correlation with gray correlation coefficient greater than 0.6 are used as input of the Wavelet Neural Network(WNN), and the structure of the model is simplified considering both the stationary and non-stationary historical data. In terms of algorithm, by using the genetic particle swarm optimization algorithm to select optimally the initial values of the wavelet neural network, the results can speed up the convergence of network training under the condition that it is not easy to fall into local optimum. The experimental results show that the proposed model is superior to hybrid ARIMA and GPSOWNN in terms of prediction accuracy, the genetic particle swarm optimization algorithm is superior to the genetic algorithm optimization model in terms of convergence speed.
2019, 41(9): 2280-2286.
doi: 10.11999/JEIT180925
Abstract:
Considering discrete-time chaotic dynamics systems, a new algorithm is proposed which is based on matrix eigenvalues and eigenvectors to configure Lyapunov exponents to be positive. The eigenvalues and eigenvectors of the discrete controlled matrix are calculated to design a general controller with positive Lyapunov exponents. The theory proves the boundedness of the system orbit and the finiteness of the Lyapunov exponents. The numerical simulation analysis of the linear feedback operator and the perturbation feedback operator verifies the correctness, versatility and effectiveness of the algorithm. Performance evaluations show that, compared with Chen-Lai methods, the proposed method can construct chaotic system with lower computation complexity and the running time is shorter and the outputs demonstrate strong randomness. Thus, a discrete chaotic system with no degradation and no merger is realized.
Considering discrete-time chaotic dynamics systems, a new algorithm is proposed which is based on matrix eigenvalues and eigenvectors to configure Lyapunov exponents to be positive. The eigenvalues and eigenvectors of the discrete controlled matrix are calculated to design a general controller with positive Lyapunov exponents. The theory proves the boundedness of the system orbit and the finiteness of the Lyapunov exponents. The numerical simulation analysis of the linear feedback operator and the perturbation feedback operator verifies the correctness, versatility and effectiveness of the algorithm. Performance evaluations show that, compared with Chen-Lai methods, the proposed method can construct chaotic system with lower computation complexity and the running time is shorter and the outputs demonstrate strong randomness. Thus, a discrete chaotic system with no degradation and no merger is realized.
