Email alert
2019 Vol. 41, No. 11
Display Method:
2019, 41(11): 2541-2548.
doi: 10.11999/JEIT181136
Abstract:
The spectral efficiency and energy efficiency of the uplink of massive MIMO-OFDM system is studied using mixed-precision Analog-Digital Converter (ADC) and Zero-Forcing (ZF) reception algorithm at the receiver. By using the additive quantization noise model to analyze the performance of the system, the approximate closed expression of the spectral efficiency and energy efficiency of the whole system is derived, and the correctness of the expression is proved by simulation. The research results show that the spectral efficiency of the system is related to the transmission power of each user, the number of antennas at the receiver and the quantization accuracy of the receiver. Numerical and simulation results also show that the performance loss caused by the low-precision ADC can be compensated by increasing the number of antennas at the base station.
The spectral efficiency and energy efficiency of the uplink of massive MIMO-OFDM system is studied using mixed-precision Analog-Digital Converter (ADC) and Zero-Forcing (ZF) reception algorithm at the receiver. By using the additive quantization noise model to analyze the performance of the system, the approximate closed expression of the spectral efficiency and energy efficiency of the whole system is derived, and the correctness of the expression is proved by simulation. The research results show that the spectral efficiency of the system is related to the transmission power of each user, the number of antennas at the receiver and the quantization accuracy of the receiver. Numerical and simulation results also show that the performance loss caused by the low-precision ADC can be compensated by increasing the number of antennas at the base station.
2019, 41(11): 2549-2556.
doi: 10.11999/JEIT190128
Abstract:
For the service characteristics and Quality of Service (QoS) requirements of Machine Type Communications (MTC), short-packet/short-coded block transmission in MTC based on Non-Orthogonal Multiple Access (NOMA) is considered in this paper, and the resource optimization problem of the Ultra-Reliable and Low-Latency (URLL) in MTC based on NOMA is discussed. Currently, uplink transmission is a bottleneck of MTC based on NOMA. Firstly, considering the performance requirements supporting NOMA and high reliability and low latency in wireless cellular networks, a system model for uplink wireless resource optimization is established. Then, the uplink transmission delay is analyzed and the link reliability function based on distance is derived. Further, with the constraints of delay, reliability and bandwidth, a wireless resource allocation algorithm for maximizing the sum rates of central users is proposed, and also the convergence proof and complexity analysis of the algorithm are given. Finally, the simulation results show the performance advantages of the proposed optimal scheme.
For the service characteristics and Quality of Service (QoS) requirements of Machine Type Communications (MTC), short-packet/short-coded block transmission in MTC based on Non-Orthogonal Multiple Access (NOMA) is considered in this paper, and the resource optimization problem of the Ultra-Reliable and Low-Latency (URLL) in MTC based on NOMA is discussed. Currently, uplink transmission is a bottleneck of MTC based on NOMA. Firstly, considering the performance requirements supporting NOMA and high reliability and low latency in wireless cellular networks, a system model for uplink wireless resource optimization is established. Then, the uplink transmission delay is analyzed and the link reliability function based on distance is derived. Further, with the constraints of delay, reliability and bandwidth, a wireless resource allocation algorithm for maximizing the sum rates of central users is proposed, and also the convergence proof and complexity analysis of the algorithm are given. Finally, the simulation results show the performance advantages of the proposed optimal scheme.
2019, 41(11): 2557-2564.
doi: 10.11999/JEIT181191
Abstract:
The power control problem of mobile users in macro-femto heterogeneous cellular networks is studied. Firstly, an optimization model that maximizes the total energy efficiency of femtocells with the minimum received signal-to-noise ratio as the constraint is established. Then, a femtocell centralized Power Control algorithm based on Q-Learning (PCQL) is proposed. Based on reinforcement learning, the algorithm can adjust the transmit power of the user terminal without accurate channel state information simultaneously. The simulation results show that the algorithm can effectively control the power of the user terminal and improve system energy efficient.
The power control problem of mobile users in macro-femto heterogeneous cellular networks is studied. Firstly, an optimization model that maximizes the total energy efficiency of femtocells with the minimum received signal-to-noise ratio as the constraint is established. Then, a femtocell centralized Power Control algorithm based on Q-Learning (PCQL) is proposed. Based on reinforcement learning, the algorithm can adjust the transmit power of the user terminal without accurate channel state information simultaneously. The simulation results show that the algorithm can effectively control the power of the user terminal and improve system energy efficient.
2019, 41(11): 2565-2570.
doi: 10.11999/JEIT180408
Abstract:
Due to the limited transmission performance of cellular network and the buffering capabilities of the Base Station (BS), it is very difficult to achieve the Quality of Service (QoS) requirements of multi-user content requests. In this paper, a joint user association and content deployment algorithm is proposed for cellular Device-to-Device (D2D) communication network. Assuming that multiple users located in a specific area may have content requests for the same content, a clustering and content deployment mechanism is presented in order to achieve efficient content acquisition. A joint clustering and content deployment optimization model is formulated to minimize total user service delay, which can be solved by Lagrange partial relaxation, iterative algorithm and Kuhn-Munkres algorithm, and the joint clustering and content deployment optimization strategies can be obtained. Finally, the effectiveness of the proposed algorithm is verified by MATLAB simulation.
Due to the limited transmission performance of cellular network and the buffering capabilities of the Base Station (BS), it is very difficult to achieve the Quality of Service (QoS) requirements of multi-user content requests. In this paper, a joint user association and content deployment algorithm is proposed for cellular Device-to-Device (D2D) communication network. Assuming that multiple users located in a specific area may have content requests for the same content, a clustering and content deployment mechanism is presented in order to achieve efficient content acquisition. A joint clustering and content deployment optimization model is formulated to minimize total user service delay, which can be solved by Lagrange partial relaxation, iterative algorithm and Kuhn-Munkres algorithm, and the joint clustering and content deployment optimization strategies can be obtained. Finally, the effectiveness of the proposed algorithm is verified by MATLAB simulation.
2019, 41(11): 2571-2577.
doi: 10.11999/JEIT180937
Abstract:
In order to improve multicast’s spectrum energy-efficient of elastic optical network configured with Colorless, Directionless and Contentionless-Flexible Reconfigurable Optical Add/Drop Multiplexer (CDC-F ROADM) nodes, an All-optical Multicast Energy Efficiency Scheduling Algorithm (AMEESA) is proposed. In the routing phase, considering both energy consumption and link spectrum resource utilization, the link cost function is designed to establish the multicast tree with the least cost. In the spectrum allocation phase, a spectrum conversion method based on High Spectral Resolution (HSR) is designed by changing the spectrum slot index of adjacent links according to links availability of spectrum blocks. And an energy-saving spectrum conversion scheme is selected to allocate spectrum block resources for the multicast tree. Simulation analysis shows that the proposed algorithm can effectively improve the network energy efficiency and reduce the bandwidth blocking probability of IP multicast.
In order to improve multicast’s spectrum energy-efficient of elastic optical network configured with Colorless, Directionless and Contentionless-Flexible Reconfigurable Optical Add/Drop Multiplexer (CDC-F ROADM) nodes, an All-optical Multicast Energy Efficiency Scheduling Algorithm (AMEESA) is proposed. In the routing phase, considering both energy consumption and link spectrum resource utilization, the link cost function is designed to establish the multicast tree with the least cost. In the spectrum allocation phase, a spectrum conversion method based on High Spectral Resolution (HSR) is designed by changing the spectrum slot index of adjacent links according to links availability of spectrum blocks. And an energy-saving spectrum conversion scheme is selected to allocate spectrum block resources for the multicast tree. Simulation analysis shows that the proposed algorithm can effectively improve the network energy efficiency and reduce the bandwidth blocking probability of IP multicast.
2019, 41(11): 2578-2584.
doi: 10.11999/JEIT190037
Abstract:
In order to reduce the computational complexity of Convolutional Neural Network(CNN), the two-dimensional fast filtering algorithm is introduced into the CNN, and a hardware architecture for implementing CNN layer-by-layer acceleration on FPGA is proposed. Firstly, the line buffer loop control unit is designed by using the cyclic transformation method to manage effectively different convolution windows and the input feature map data between different layers, and starts the convolution calculation acceleration unit by the flag signal to realize layer-by-layer acceleration. Secondly, a convolution calculation accelerating unit based on 4 parallel fast filtering algorithm is designed. The unit is realized by a less complex parallel filtering structure composed of several small filters. Using the handwritten digit set MNIST to test the designed CNN accelerator circuit, the results show that on the xilinx kintex7 platform, when the input clock is 100 MHz, the computational performance of the circuit reaches 20.49 GOPS, and the recognition rate is 98.68%. It can be seen that the computational performance of the circuit can be improved by reducing the amount of calculation of the CNN.
In order to reduce the computational complexity of Convolutional Neural Network(CNN), the two-dimensional fast filtering algorithm is introduced into the CNN, and a hardware architecture for implementing CNN layer-by-layer acceleration on FPGA is proposed. Firstly, the line buffer loop control unit is designed by using the cyclic transformation method to manage effectively different convolution windows and the input feature map data between different layers, and starts the convolution calculation acceleration unit by the flag signal to realize layer-by-layer acceleration. Secondly, a convolution calculation accelerating unit based on 4 parallel fast filtering algorithm is designed. The unit is realized by a less complex parallel filtering structure composed of several small filters. Using the handwritten digit set MNIST to test the designed CNN accelerator circuit, the results show that on the xilinx kintex7 platform, when the input clock is 100 MHz, the computational performance of the circuit reaches 20.49 GOPS, and the recognition rate is 98.68%. It can be seen that the computational performance of the circuit can be improved by reducing the amount of calculation of the CNN.
2019, 41(11): 2585-2591.
doi: 10.11999/JEIT190143
Abstract:
Bitstream generator in FPGA Electronic Design Automation(EDA) offers precise configuration information, which enables the application circuits to be implemented on the target device. On one hand, modern FPGAs tend to have larger device scale and more configuration bits, on the other hand, embedded applications (e.g. eFPGAs) require better configuration efficiency and smaller, more adaptive database. In order to meet these new requirements, a bit-stream generation method is proposed which firstly models the configurable resources by configuration modes and matches the netlist with these models, then hierarchical mapping strategy is used to search every bit on a dynamically generated database determined by the array floorplan. This method well meets the challenges that embedded applications may bring-the surge of configuration bit count and the changeable size of the array. Compared to flattened modelling and mapping method, its time complexity is reduced from O(n) to O(lgn).
Bitstream generator in FPGA Electronic Design Automation(EDA) offers precise configuration information, which enables the application circuits to be implemented on the target device. On one hand, modern FPGAs tend to have larger device scale and more configuration bits, on the other hand, embedded applications (e.g. eFPGAs) require better configuration efficiency and smaller, more adaptive database. In order to meet these new requirements, a bit-stream generation method is proposed which firstly models the configurable resources by configuration modes and matches the netlist with these models, then hierarchical mapping strategy is used to search every bit on a dynamically generated database determined by the array floorplan. This method well meets the challenges that embedded applications may bring-the surge of configuration bit count and the changeable size of the array. Compared to flattened modelling and mapping method, its time complexity is reduced from O(n) to O(lgn).
2019, 41(11): 2592-2598.
doi: 10.11999/JEIT181060
Abstract:
A novel technique for increasing the load response speed of Capacitor-Less Low-DropOut linear regulator (CL-LDO) is proposed to improve the transient response of CL-LDO when its load current changes. With an additional fast signal feedback path, the CL-LDO can achieve fast transient response so that the overshoot and undershoot of its output voltage can be dramatically reduced. A CL-LDO with fast response is realized in 0.18 μm CMOS and occupies an active area of 0.00529 mm2. The CL-LDO has an output voltage of 1.194 V when the input supply voltage ranges from 1.5 V to 2.5 V. When the load current changes from 100 μA to 10 mA with the rise and fall time of 1 μs, the output of LDO can be recovered from its overshoot and undershoot to a stable voltage within 489.537 ns and 960.918 ns, respectively. Compared with a traditional CL-LDO without this proposed technique, the transient response speed of this CL-LDO is increased by 7.41 times. The overshoot and undershoot of the output voltage is decreased by 35.3% and 78.1%, respectively.
A novel technique for increasing the load response speed of Capacitor-Less Low-DropOut linear regulator (CL-LDO) is proposed to improve the transient response of CL-LDO when its load current changes. With an additional fast signal feedback path, the CL-LDO can achieve fast transient response so that the overshoot and undershoot of its output voltage can be dramatically reduced. A CL-LDO with fast response is realized in 0.18 μm CMOS and occupies an active area of 0.00529 mm2. The CL-LDO has an output voltage of 1.194 V when the input supply voltage ranges from 1.5 V to 2.5 V. When the load current changes from 100 μA to 10 mA with the rise and fall time of 1 μs, the output of LDO can be recovered from its overshoot and undershoot to a stable voltage within 489.537 ns and 960.918 ns, respectively. Compared with a traditional CL-LDO without this proposed technique, the transient response speed of this CL-LDO is increased by 7.41 times. The overshoot and undershoot of the output voltage is decreased by 35.3% and 78.1%, respectively.
2019, 41(11): 2599-2605.
doi: 10.11999/JEIT190058
Abstract:
Considering the large computational complexity and the long-time calculation of Convolutional Neural Networks (CNN), an Field-Programmable Gate Array(FPGA)-based CNN hardware accelerator is proposed. Firstly, by deeply analyzing the forward computing principle and exploring the parallelism of convolutional layer, a hardware architecture in which parallel for the input channel and output channel, deep pipeline for the convolution window is presented. Then, a full parallel multi-addition tree is designed to accelerate convolution and efficient window buffer to implement deep pipelining operation of convolution window. The experimental results show that the energy efficiency ratio of proposed accelerator reaches 32.73 GOPS/W, which is 34% higher than the existing solutions, as the performance reaches 317.86 GOPS.
Considering the large computational complexity and the long-time calculation of Convolutional Neural Networks (CNN), an Field-Programmable Gate Array(FPGA)-based CNN hardware accelerator is proposed. Firstly, by deeply analyzing the forward computing principle and exploring the parallelism of convolutional layer, a hardware architecture in which parallel for the input channel and output channel, deep pipeline for the convolution window is presented. Then, a full parallel multi-addition tree is designed to accelerate convolution and efficient window buffer to implement deep pipelining operation of convolution window. The experimental results show that the energy efficiency ratio of proposed accelerator reaches 32.73 GOPS/W, which is 34% higher than the existing solutions, as the performance reaches 317.86 GOPS.
2019, 41(11): 2606-2613.
doi: 10.11999/JEIT181156
Abstract:
In order to accommodate the development of new communication technology, an integrated programmable microwave photonic filter with high shape-factor is proposed in this paper. This filter is based on Silicon-On-Insulator (SOI) and an eight-tap finite impulse response. By controlling the thermal heaters on the amplitude modulator and phase modulator of each tap, a rectangular filter with tunable bandwidth and high shape-factor greater than 0.55 is obtained. Furthermore, the tunability of central frequency, bandwidth and variable pass-band shape can be also realized. Small size, light weight and flexibility are advantages of the preposed filters, moreover, it can be applied to large bandwidth signal processing and an alternative method to part the channels. So it can be widely used in defense field and 5G networks.
In order to accommodate the development of new communication technology, an integrated programmable microwave photonic filter with high shape-factor is proposed in this paper. This filter is based on Silicon-On-Insulator (SOI) and an eight-tap finite impulse response. By controlling the thermal heaters on the amplitude modulator and phase modulator of each tap, a rectangular filter with tunable bandwidth and high shape-factor greater than 0.55 is obtained. Furthermore, the tunability of central frequency, bandwidth and variable pass-band shape can be also realized. Small size, light weight and flexibility are advantages of the preposed filters, moreover, it can be applied to large bandwidth signal processing and an alternative method to part the channels. So it can be widely used in defense field and 5G networks.
2019, 41(11): 2614-2622.
doi: 10.11999/JEIT181190
Abstract:
Chirp signals are widely used in communication and exploration. The parameter analysis of the chirp signals often uses a Wigner-Ville Distribution (WVD) based time-frequency analysis method, which achieves high time-frequency resolution. However, this method has defects in cross terms, high sidelobes, and spectral aliasing problems. To solve these problems, a time-frequency analysis method called Spatially Variant Apodiztion-rearrange Wigner Ville Distribution (SVA-rWVD) is proposed, which achieves low sidelobes by exploiting the Spatially Variant Apodization (SVA) techniques, and avoids the cross terms and the spectral aliasing problems by applying the Short Time Fourier Transform (STFT). Furthermore, a new time-frequency distribution is obtained from the proposed method. Extensive simulations show that the time-frequency distribution obtained by the proposed method not only reduces the sidelobe level to –40 dB but also eliminates cross terms and spectral aliasing for both single-component and multi-component chirp signals.
Chirp signals are widely used in communication and exploration. The parameter analysis of the chirp signals often uses a Wigner-Ville Distribution (WVD) based time-frequency analysis method, which achieves high time-frequency resolution. However, this method has defects in cross terms, high sidelobes, and spectral aliasing problems. To solve these problems, a time-frequency analysis method called Spatially Variant Apodiztion-rearrange Wigner Ville Distribution (SVA-rWVD) is proposed, which achieves low sidelobes by exploiting the Spatially Variant Apodization (SVA) techniques, and avoids the cross terms and the spectral aliasing problems by applying the Short Time Fourier Transform (STFT). Furthermore, a new time-frequency distribution is obtained from the proposed method. Extensive simulations show that the time-frequency distribution obtained by the proposed method not only reduces the sidelobe level to –40 dB but also eliminates cross terms and spectral aliasing for both single-component and multi-component chirp signals.
2019, 41(11): 2623-2631.
doi: 10.11999/JEIT190026
Abstract:
The Dense Focal Plane Array Feed (DFPAF), which integrates the characters of multi-beam feed with multiple independent horns and Phased Array Feed (PAF), can simultaneously provide more fixed shaped beams and wider field of view than multi-beam feed with multiple independent horns and PAF. It attracts more attention in radio telescope, radar, electronic reconnaissance, satellite communication and so on. Its unique structure promotes the studies on special design method recently. Combing the theory of array antenna and inherent characteristic of parabolic reflector antenna, a fast design method with robust processing procedure is proposed in this paper. The design principle, calculated results, and comparison between DFPAF and the most representative multi-beam feed with multiple independent horns are presented. All these provide a theoretical basis and reference data for the design of giant reflector with DFPAF.
The Dense Focal Plane Array Feed (DFPAF), which integrates the characters of multi-beam feed with multiple independent horns and Phased Array Feed (PAF), can simultaneously provide more fixed shaped beams and wider field of view than multi-beam feed with multiple independent horns and PAF. It attracts more attention in radio telescope, radar, electronic reconnaissance, satellite communication and so on. Its unique structure promotes the studies on special design method recently. Combing the theory of array antenna and inherent characteristic of parabolic reflector antenna, a fast design method with robust processing procedure is proposed in this paper. The design principle, calculated results, and comparison between DFPAF and the most representative multi-beam feed with multiple independent horns are presented. All these provide a theoretical basis and reference data for the design of giant reflector with DFPAF.
2019, 41(11): 2632-2638.
doi: 10.11999/JEIT181067
Abstract:
The G -matrix model method is usually used to achieve the brightness temperature reconstruction for the one-Dimensional (1-D) synthetic aperture microwave radiometer system. For the 1-D radiometer system, the imaging process mainly includes: the radiometer instrument observes the full field of view of the 2-D target scene maps, and obtains the 1-D samples of the visibility, and then inverts the system parameter matrix G to realize the reconstruction of the 1-D image of the target scene. Since the system sampling baselines are only distributed in the 1-D of the spatial frequency domain, in the process of the brightness temperature image reconstruction, the matrix G needs to realize 2-D to 1-D conversion. Therefore, two G -matrix modification methods are proposed to improve the imaging quality for the 1-D synthetic aperture microwave radiometer. For the 8-element ground radiometer prototype system and the 10-element salinity radiometer system, theoretical analysis and simulation experiments have verified that the G -matrix modification methods proposed in this paper can effectively improve the imaging results, and can effectively suppress the imaging error caused by the side-lobed degradation of the antenna patterns.
The G -matrix model method is usually used to achieve the brightness temperature reconstruction for the one-Dimensional (1-D) synthetic aperture microwave radiometer system. For the 1-D radiometer system, the imaging process mainly includes: the radiometer instrument observes the full field of view of the 2-D target scene maps, and obtains the 1-D samples of the visibility, and then inverts the system parameter matrix G to realize the reconstruction of the 1-D image of the target scene. Since the system sampling baselines are only distributed in the 1-D of the spatial frequency domain, in the process of the brightness temperature image reconstruction, the matrix G needs to realize 2-D to 1-D conversion. Therefore, two G -matrix modification methods are proposed to improve the imaging quality for the 1-D synthetic aperture microwave radiometer. For the 8-element ground radiometer prototype system and the 10-element salinity radiometer system, theoretical analysis and simulation experiments have verified that the G -matrix modification methods proposed in this paper can effectively improve the imaging results, and can effectively suppress the imaging error caused by the side-lobed degradation of the antenna patterns.
2019, 41(11): 2639-2645.
doi: 10.11999/JEIT190010
Abstract:
Forwarding dense false target jamming disturbs the detection and recognition of real targets by generating multiple false targets in the range dimension. Because the false echo signal is highly correlated with the real signal, it is difficult for radar to recognize and suppress it effectively. Frequency agile radar improves greatly the low interception and anti-jamming ability of radar by randomly changing the carrier frequency of transmitting adjacent pulses. However, agile radar can not completely eliminate the interference, some target echo pulses may be submerged by the interference, agile radar can not complete coherent accumulation and target detection well either. To solve the above problems, an anti-jamming method of frequency agility combined with Hough transform is proposed. Firstly, the inter-pulse frequency agility technology is used to avoid most narrowband aiming and deceptive jamming. Then, according to the time discontinuity of the jamming signal, Hough transform and peak extraction are used to identify and suppress the jamming. Frequency agility is incompatible with the traditional Moving Target Detection(MTD). Target detection is accomplished by sparse reconstruction. The simulation and actual radar and jammer countermeasure experiments show that the proposed method can achieve good anti-jamming performance and target detection performance.
Forwarding dense false target jamming disturbs the detection and recognition of real targets by generating multiple false targets in the range dimension. Because the false echo signal is highly correlated with the real signal, it is difficult for radar to recognize and suppress it effectively. Frequency agile radar improves greatly the low interception and anti-jamming ability of radar by randomly changing the carrier frequency of transmitting adjacent pulses. However, agile radar can not completely eliminate the interference, some target echo pulses may be submerged by the interference, agile radar can not complete coherent accumulation and target detection well either. To solve the above problems, an anti-jamming method of frequency agility combined with Hough transform is proposed. Firstly, the inter-pulse frequency agility technology is used to avoid most narrowband aiming and deceptive jamming. Then, according to the time discontinuity of the jamming signal, Hough transform and peak extraction are used to identify and suppress the jamming. Frequency agility is incompatible with the traditional Moving Target Detection(MTD). Target detection is accomplished by sparse reconstruction. The simulation and actual radar and jammer countermeasure experiments show that the proposed method can achieve good anti-jamming performance and target detection performance.
Inverse Synthetic Aperture Radar Imaging with Non-Coherent Short Pulse Radar and Its Sparse Recovery
2019, 41(11): 2646-2653.
doi: 10.11999/JEIT180912
Abstract:
The microwave source of Non-Coherent Short Pulse (NCSP) radar transmits short pulse. Thus, for high velocity targets, the motion effect in the pulse duration can be neglected, and the echo signal does not need special motion compensation. In order to use the NCSP radar signal for Inverse Synthetic Aperture Radar (ISAR) imaging, the compensation coherent processing method is applied to removing the uncertainty of the envelope time and the initial phase uncertainty. Assuming that the echo is envelope-aligned and initially compensated by conventional methods, ISAR radar imaging can be performed using the Range-Doppler (RD) method, subsequently. The simulation verifies the feasibility of the compensation signal ISAR imaging. However, the carrier-frequency random jitter factor of NCSP radar causes random-modulated sidelobes in the Doppler dimension, which affect imaging quality. In this paper, the sparse recovery technique is used to perform sparse reconstruction of the target scattering center in the imaging space. The Orthogonal Matching Pursuit (OMP) algorithm and the Sparse Bayesian Learning (SBL) algorithm are used as the recovery algorithm for imaging simulation experiments. The simulation results show that the sparse recovery technique can suppress the imaging sidelobes caused by non-coherence and improve the imaging quality.
The microwave source of Non-Coherent Short Pulse (NCSP) radar transmits short pulse. Thus, for high velocity targets, the motion effect in the pulse duration can be neglected, and the echo signal does not need special motion compensation. In order to use the NCSP radar signal for Inverse Synthetic Aperture Radar (ISAR) imaging, the compensation coherent processing method is applied to removing the uncertainty of the envelope time and the initial phase uncertainty. Assuming that the echo is envelope-aligned and initially compensated by conventional methods, ISAR radar imaging can be performed using the Range-Doppler (RD) method, subsequently. The simulation verifies the feasibility of the compensation signal ISAR imaging. However, the carrier-frequency random jitter factor of NCSP radar causes random-modulated sidelobes in the Doppler dimension, which affect imaging quality. In this paper, the sparse recovery technique is used to perform sparse reconstruction of the target scattering center in the imaging space. The Orthogonal Matching Pursuit (OMP) algorithm and the Sparse Bayesian Learning (SBL) algorithm are used as the recovery algorithm for imaging simulation experiments. The simulation results show that the sparse recovery technique can suppress the imaging sidelobes caused by non-coherence and improve the imaging quality.
2019, 41(11): 2654-2660.
doi: 10.11999/JEIT190114
Abstract:
In order to improve missile-borne radar detection performance in modern electronic warfare, a radar waveform design method based on Nash equilibrium is proposed. Firstly, the radar and jammer game signal models are established in electronic warfare. Based on maximum Signal-to-Interference-plus-Noise Ratio (SINR), waveform strategies of radar and jammer are designed respectively. Secondly, the existence of Nash equilibrium solution is demonstrated by mathematical derivation and verified in experimental simulation. A multiple iterative water-filling method which repeatedly eliminates strict disadvantages is designed to achieve Nash equilibrium. The maxmin scheme of disequilibrium game is deduced by two-step water-filling method. Finally, the radar detection performance of optimization strategies is tested by simulation experiments. Simulation results reveal that the radar waveform design based on Nash equilibrium is beneficial to improve the radar detection performance under game conditions. Compared with no-game and maxmin strategies, the radar detection probability of Nash equilibrium strategy can be increased by 12.02% and 3.82%, respectively. It is proved that the Nash equilibrium strategy of this paper is closer to the Pareto optimality.
In order to improve missile-borne radar detection performance in modern electronic warfare, a radar waveform design method based on Nash equilibrium is proposed. Firstly, the radar and jammer game signal models are established in electronic warfare. Based on maximum Signal-to-Interference-plus-Noise Ratio (SINR), waveform strategies of radar and jammer are designed respectively. Secondly, the existence of Nash equilibrium solution is demonstrated by mathematical derivation and verified in experimental simulation. A multiple iterative water-filling method which repeatedly eliminates strict disadvantages is designed to achieve Nash equilibrium. The maxmin scheme of disequilibrium game is deduced by two-step water-filling method. Finally, the radar detection performance of optimization strategies is tested by simulation experiments. Simulation results reveal that the radar waveform design based on Nash equilibrium is beneficial to improve the radar detection performance under game conditions. Compared with no-game and maxmin strategies, the radar detection probability of Nash equilibrium strategy can be increased by 12.02% and 3.82%, respectively. It is proved that the Nash equilibrium strategy of this paper is closer to the Pareto optimality.
2019, 41(11): 2661-2668.
doi: 10.11999/JEIT190163
Abstract:
For Time-Division Multiple Access (TDMA) signals, the performance of Specific Emitter Identification (SEI) is primarily limited by burst duration. To remedy this shortcoming, a novel radiometric signature is presented, which reveals whether the users of the adjacent time slots are the same from a perspective of carrier phase, thereby providing the basis for data accumulation of the same user. First, the feature mechanism is introduced, as well as the extraction method. Thereafter, user identity detection of the adjacent slots is implemented with an adaptive threshold, which is derived from the distribution of the signature. Finally, a new SEI processing procedure is designed with data accumulation, which breaks the routine of identifying only one slot at a time. Simulation results demonstrate that the proposed signature is resilient against the noise, and can accurately detect the user identity of the adjacent slots. Compared with the traditional processing procedure, the proposed one can effectively improve the SEI performance of TDMA signals.
For Time-Division Multiple Access (TDMA) signals, the performance of Specific Emitter Identification (SEI) is primarily limited by burst duration. To remedy this shortcoming, a novel radiometric signature is presented, which reveals whether the users of the adjacent time slots are the same from a perspective of carrier phase, thereby providing the basis for data accumulation of the same user. First, the feature mechanism is introduced, as well as the extraction method. Thereafter, user identity detection of the adjacent slots is implemented with an adaptive threshold, which is derived from the distribution of the signature. Finally, a new SEI processing procedure is designed with data accumulation, which breaks the routine of identifying only one slot at a time. Simulation results demonstrate that the proposed signature is resilient against the noise, and can accurately detect the user identity of the adjacent slots. Compared with the traditional processing procedure, the proposed one can effectively improve the SEI performance of TDMA signals.
2019, 41(11): 2669-2674.
doi: 10.11999/JEIT180870
Abstract:
In order to achieve routing optimization in the Software Defined Network (SDN) environment, deep reinforcement learning is imposed to the SDN routing process and a mechanism based on deep reinforcement learning is proposed to optimize routing. This mechanism can improve network performance such as delay, throughput, and realize black-box optimization in continuous time, which surely reduces network operation and maintenance costs. Besides, the proposed routing optimization mechanism is evaluated through a series of experiments. The experimental results show that the proposed SDN routing optimization mechanism has good convergence and effectiveness, and can provide better routing configurations and performance stability than traditional routing protocols.
In order to achieve routing optimization in the Software Defined Network (SDN) environment, deep reinforcement learning is imposed to the SDN routing process and a mechanism based on deep reinforcement learning is proposed to optimize routing. This mechanism can improve network performance such as delay, throughput, and realize black-box optimization in continuous time, which surely reduces network operation and maintenance costs. Besides, the proposed routing optimization mechanism is evaluated through a series of experiments. The experimental results show that the proposed SDN routing optimization mechanism has good convergence and effectiveness, and can provide better routing configurations and performance stability than traditional routing protocols.
2019, 41(11): 2675-2683.
doi: 10.11999/JEIT190145
Abstract:
With the introduction of Network Function Virtualization (NFV), the operating costs of operators can be greatly reduced. However, most existing Service Function Chain (SFC) orchestration researches can not optimize the resources utilization while guaranteeing the performance of service delay. A spatial and temporal optimal method of Service Function Chain (SFC) orchestration based on an overlay network structure is proposed. Based on the consideration of the restrictions such as computing resource, network resource and fine-grained end to end delay, this method separates the computing resource and network resource. The resources cost and related delay of SFC can be abstracted into the links weight of overlay network, which can help to convert the SFC orchestration problem into the shortest path problem that can be easily solved. As for the SFC requests set requiring batch processing, an Overlay Network based Simulated Annealing iterative optimal orchestration algorithm(ONSA) is designed. The simulation results demonstrate that the proposed orchestration scheme can reduce the end-to-end delay, the utilization ratio of link bandwidth resource and the operational expenditure by 29.5%, 12.4% and 15.2%, and the acceptance ratio of requests set can be improved by 22.3%. The performance of Virtual Network Function (VNF) load balancing can be significantly improved.
With the introduction of Network Function Virtualization (NFV), the operating costs of operators can be greatly reduced. However, most existing Service Function Chain (SFC) orchestration researches can not optimize the resources utilization while guaranteeing the performance of service delay. A spatial and temporal optimal method of Service Function Chain (SFC) orchestration based on an overlay network structure is proposed. Based on the consideration of the restrictions such as computing resource, network resource and fine-grained end to end delay, this method separates the computing resource and network resource. The resources cost and related delay of SFC can be abstracted into the links weight of overlay network, which can help to convert the SFC orchestration problem into the shortest path problem that can be easily solved. As for the SFC requests set requiring batch processing, an Overlay Network based Simulated Annealing iterative optimal orchestration algorithm(ONSA) is designed. The simulation results demonstrate that the proposed orchestration scheme can reduce the end-to-end delay, the utilization ratio of link bandwidth resource and the operational expenditure by 29.5%, 12.4% and 15.2%, and the acceptance ratio of requests set can be improved by 22.3%. The performance of Virtual Network Function (VNF) load balancing can be significantly improved.
2019, 41(11): 2684-2690.
doi: 10.11999/JEIT180970
Abstract:
To support the execution of computation-intensive, delay-sensitive computing task by moving down the computing and processing capability in mobile edge computing becomes the current trend. However, when serving a large number of mobile users, how to use effectively the edge nodes with limited computing resources to ensure Quality of service (QoS) of end-user has become a key issue. To solve this problem, the edge cloud and remote cloud are combined to build a layered edge cloud computing architecture. Based on this architecture, with the goal of minimizing mobile device energy consumption and task execution time, the problem which is proved to be convex is formulated to minimize the weight sum of energy and delay. A computation offloading and resource allocation mechanism based on multiplier method is proposed. Simulations are conducted to evaluate the proposed mechanism. Compared with local computing and computation offloading mechanism, the proposed mechanism can effectively reduce the energy consumption of mobile device and the delay of system by up to 60% and 10%, respectively.
To support the execution of computation-intensive, delay-sensitive computing task by moving down the computing and processing capability in mobile edge computing becomes the current trend. However, when serving a large number of mobile users, how to use effectively the edge nodes with limited computing resources to ensure Quality of service (QoS) of end-user has become a key issue. To solve this problem, the edge cloud and remote cloud are combined to build a layered edge cloud computing architecture. Based on this architecture, with the goal of minimizing mobile device energy consumption and task execution time, the problem which is proved to be convex is formulated to minimize the weight sum of energy and delay. A computation offloading and resource allocation mechanism based on multiplier method is proposed. Simulations are conducted to evaluate the proposed mechanism. Compared with local computing and computation offloading mechanism, the proposed mechanism can effectively reduce the energy consumption of mobile device and the delay of system by up to 60% and 10%, respectively.
2019, 41(11): 2691-2698.
doi: 10.11999/JEIT190166
Abstract:
Online contract signing is becoming more and more popular in e-commerce. It is not easy to sign a contract between two parties who do not trust each other. Many of these protocols involve the participation of third parties, but they are not advantageous in efficiency and prone to security problems. Currently, contract signing agreements with third-party participation are replaced by block chain technology, but the public verification of block chain challenges the sensitive information of both the signer and the contract to be signed. And most of the agreements are for the signing of contracts between the two parties. With the increase of the number of signatories, the communication cost and complexity of the agreements increase sharply. Combined with the existing protocols, this paper proposes an efficient multi-party contract signing protocol. In the protocol, an efficient aggregation signature scheme based on no certificate is used to improve the signature verification efficiency of the signer under the block chain, and only the temporary key of the signer is disclosed on the block chain to reduce the system overhead. The protocol satisfies the requirements of correctness, security, fairness, privacy and high efficiency.
Online contract signing is becoming more and more popular in e-commerce. It is not easy to sign a contract between two parties who do not trust each other. Many of these protocols involve the participation of third parties, but they are not advantageous in efficiency and prone to security problems. Currently, contract signing agreements with third-party participation are replaced by block chain technology, but the public verification of block chain challenges the sensitive information of both the signer and the contract to be signed. And most of the agreements are for the signing of contracts between the two parties. With the increase of the number of signatories, the communication cost and complexity of the agreements increase sharply. Combined with the existing protocols, this paper proposes an efficient multi-party contract signing protocol. In the protocol, an efficient aggregation signature scheme based on no certificate is used to improve the signature verification efficiency of the signer under the block chain, and only the temporary key of the signer is disclosed on the block chain to reduce the system overhead. The protocol satisfies the requirements of correctness, security, fairness, privacy and high efficiency.
2019, 41(11): 2699-2707.
doi: 10.11999/JEIT190127
Abstract:
Side channel attack is the primary way to leak information between tenants in current cloud computing environment. However, existing Service Function Chain (SFC) deployment methods do not fully consider the side channel attack problem faced by the Virtual Network Function (VNF) in the multi-tenant environment. A SFC deployment method is proposed against side channel attack. A tenant classification strategy based on average time and a deployment strategy considering historical information are introduced. Under the resource constraints of the SFC, the optimization model is established with the goal of minimizing the number of servers that the tenant can cover. And a deployment algorithm is designed based on the greedy choice. The experimental results show that, compared with other deployment methods, this method can significantly improve the difficulty and cost of malicious tenant to realize co-residence, and reduces the risk of side channel attack faced by tenants.
Side channel attack is the primary way to leak information between tenants in current cloud computing environment. However, existing Service Function Chain (SFC) deployment methods do not fully consider the side channel attack problem faced by the Virtual Network Function (VNF) in the multi-tenant environment. A SFC deployment method is proposed against side channel attack. A tenant classification strategy based on average time and a deployment strategy considering historical information are introduced. Under the resource constraints of the SFC, the optimization model is established with the goal of minimizing the number of servers that the tenant can cover. And a deployment algorithm is designed based on the greedy choice. The experimental results show that, compared with other deployment methods, this method can significantly improve the difficulty and cost of malicious tenant to realize co-residence, and reduces the risk of side channel attack faced by tenants.
2019, 41(11): 2708-2714.
doi: 10.11999/JEIT190129
Abstract:
Heterogeneous hybrid group signcryption can not only solve the confidentiality and unforgeability of data transmission under different cryptosystems, but also encrypt data of any length. Firstly, the security of a hybrid group signcryption scheme under heterogeneous cryptosystem is analyzed, and it is pointed out that the scheme does not satisfy the correctness, confidentiality and unforgeability. And a new efficient heterogeneous hybrid group signcryption scheme is proposed. Secondly, it is proved that the proposed scheme is safe under the random oracle model. Finally, the efficiency analysis shows that the proposed scheme reduces the computational cost while realizing all the functions of the original scheme.
Heterogeneous hybrid group signcryption can not only solve the confidentiality and unforgeability of data transmission under different cryptosystems, but also encrypt data of any length. Firstly, the security of a hybrid group signcryption scheme under heterogeneous cryptosystem is analyzed, and it is pointed out that the scheme does not satisfy the correctness, confidentiality and unforgeability. And a new efficient heterogeneous hybrid group signcryption scheme is proposed. Secondly, it is proved that the proposed scheme is safe under the random oracle model. Finally, the efficiency analysis shows that the proposed scheme reduces the computational cost while realizing all the functions of the original scheme.
2019, 41(11): 2715-2721.
doi: 10.11999/JEIT181196
Abstract:
To solve the problem for the large amount of tasks, complex constraint conditions and manual which is hard to generation shifts of airport foreign airline service personnel. A shift generation model is studied and constructed for multi-task hierarchical qualification which including employees have hierarchical qualifications for tasks and shift needs to meet all kinds of labor laws and regulations and others constraints to minimize the total working time of shifts for optimum. Tabu search algorithm is designed to solve the model. Experiments, based on the actual scheduling data set of the foreign airlines service department of capital airport, verify the practicability and effectiveness of the model and the algorithm. The results show that compared to the existing manual shifts schemes, shifts obtained by using the model can fulfill all constraint conditions, shorten the total working time, reduce the number of employees and improve the utilization rate of airport resources.
To solve the problem for the large amount of tasks, complex constraint conditions and manual which is hard to generation shifts of airport foreign airline service personnel. A shift generation model is studied and constructed for multi-task hierarchical qualification which including employees have hierarchical qualifications for tasks and shift needs to meet all kinds of labor laws and regulations and others constraints to minimize the total working time of shifts for optimum. Tabu search algorithm is designed to solve the model. Experiments, based on the actual scheduling data set of the foreign airlines service department of capital airport, verify the practicability and effectiveness of the model and the algorithm. The results show that compared to the existing manual shifts schemes, shifts obtained by using the model can fulfill all constraint conditions, shorten the total working time, reduce the number of employees and improve the utilization rate of airport resources.
2019, 41(11): 2722-2728.
doi: 10.11999/JEIT180974
Abstract:
With the attribute feature information of the fuzzy membership matrix and cluster centers after the iteration not fully utilized, the results of Fuzzy C-Means (FCM) Clustering and related modified algorithms are determined based on the principle of maximum fuzzy membership, causing bad influence on the clustering accuracy. To solve this problem, the improvement ideas are proposed: to improve classification principle of FCM. The formula definition of attribute similarity in binary topological subspaces is given. Then, the improved FCM algorithm based on the Similarity of Attribute Space (FCM-SAS) is proposed: First, samples with fuzzy membership degree lower than the clustering reliability are selected as suspicious samples. Next, the attribute similarity between the suspicious samples and the cluster centers after clustering are calculated. Finally, cluster labels of suspicious samples based on the principle of maximum attribute similarity are updated. The validity and superiority of the proposed algorithm is verified by the UCI sample set experiments and comparisons with other modified algorithms based on the principle of maximum fuzzy membership.
With the attribute feature information of the fuzzy membership matrix and cluster centers after the iteration not fully utilized, the results of Fuzzy C-Means (FCM) Clustering and related modified algorithms are determined based on the principle of maximum fuzzy membership, causing bad influence on the clustering accuracy. To solve this problem, the improvement ideas are proposed: to improve classification principle of FCM. The formula definition of attribute similarity in binary topological subspaces is given. Then, the improved FCM algorithm based on the Similarity of Attribute Space (FCM-SAS) is proposed: First, samples with fuzzy membership degree lower than the clustering reliability are selected as suspicious samples. Next, the attribute similarity between the suspicious samples and the cluster centers after clustering are calculated. Finally, cluster labels of suspicious samples based on the principle of maximum attribute similarity are updated. The validity and superiority of the proposed algorithm is verified by the UCI sample set experiments and comparisons with other modified algorithms based on the principle of maximum fuzzy membership.
2019, 41(11): 2729-2735.
doi: 10.11999/JEIT190115
Abstract:
Considering the failure of the Conditional adversarial Domain AdaptatioN(CDAN) to fully utilize the sample transferability, which still struggle with some hard-to-transfer source samples disturbed the distribution of the target domain samples, a Transfer Weight based Conditional adversarial Domain AdaptatioN(TW-CDAN) is proposed. Firstly, the discriminant results in the domain discriminant model as the main factor are employed to measure the transfer performance. Then the weight is applied to class loss and minimum entropy loss. It is for eliminating the influence of hard-to-transfer samples of the model. Finally, experiments are carried out using the six domain adaptation tasks of the Office-31 dataset and the 12 domain adaptation tasks of the Office-Home dataset. The proposed method improves the 14 domain adaptation tasks and increases the average accuracy by 1.4% and 3.1% respectively.
Considering the failure of the Conditional adversarial Domain AdaptatioN(CDAN) to fully utilize the sample transferability, which still struggle with some hard-to-transfer source samples disturbed the distribution of the target domain samples, a Transfer Weight based Conditional adversarial Domain AdaptatioN(TW-CDAN) is proposed. Firstly, the discriminant results in the domain discriminant model as the main factor are employed to measure the transfer performance. Then the weight is applied to class loss and minimum entropy loss. It is for eliminating the influence of hard-to-transfer samples of the model. Finally, experiments are carried out using the six domain adaptation tasks of the Office-31 dataset and the 12 domain adaptation tasks of the Office-Home dataset. The proposed method improves the 14 domain adaptation tasks and increases the average accuracy by 1.4% and 3.1% respectively.
2019, 41(11): 2736-2743.
doi: 10.11999/JEIT181054
Abstract:
Most current transfer learning methods are modeled by utilizing the source data with the assumption that all data in the source domain are equally related to the target domain. In many practical applications, however, this assumption may induce negative learning effect when it becomes invalid. To tackle this issue, by minimizing the integrated squared error of the probability distribution of the source and target domain classification errors, the Classification-error Consensus Regularization (CCR) is proposed. Furthermore, CCR-based Adaptive knowledge Transfer Learning (CATL) method is developed to quickly determine the correlative source data and the corresponding weights. The proposed method can alleviate the negative transfer learning effect while improving the efficiency of knowledge transfer. The experimental results on the real image and text datasets validate the advantages of the CATL method.
Most current transfer learning methods are modeled by utilizing the source data with the assumption that all data in the source domain are equally related to the target domain. In many practical applications, however, this assumption may induce negative learning effect when it becomes invalid. To tackle this issue, by minimizing the integrated squared error of the probability distribution of the source and target domain classification errors, the Classification-error Consensus Regularization (CCR) is proposed. Furthermore, CCR-based Adaptive knowledge Transfer Learning (CATL) method is developed to quickly determine the correlative source data and the corresponding weights. The proposed method can alleviate the negative transfer learning effect while improving the efficiency of knowledge transfer. The experimental results on the real image and text datasets validate the advantages of the CATL method.
2019, 41(11): 2744-2751.
doi: 10.11999/JEIT181110
Abstract:
To solve the problem that small moving object is difficult to be detected in video surveillance, a track-based detection algorithm is proposed. Firstly, in order to reduce missing alarm, an adaptive foreground extraction method combining regional texture features and difference probability is presented. Then, for reducing false alarm, the probability computing model of track correlation is designed to establish the correlation of suspected objects between frames, and double-threshold are set to distinguish between true and false positive. Experimental results show that compared with many classical algorithms, this algorithm can accurately detect small moving object within the quantitative range with lower missing and false alarm.
To solve the problem that small moving object is difficult to be detected in video surveillance, a track-based detection algorithm is proposed. Firstly, in order to reduce missing alarm, an adaptive foreground extraction method combining regional texture features and difference probability is presented. Then, for reducing false alarm, the probability computing model of track correlation is designed to establish the correlation of suspected objects between frames, and double-threshold are set to distinguish between true and false positive. Experimental results show that compared with many classical algorithms, this algorithm can accurately detect small moving object within the quantitative range with lower missing and false alarm.
2019, 41(11): 2752-2759.
doi: 10.11999/JEIT181088
Abstract:
Facial expression is the most intuitive description of changes in psychological emotions, and different people have great differences in facial expressions. The existing facial expression recognition methods use facial statistical features to distinguish among different expressions, but these methods are short of deep exploration for facial detail information. According to the definition of facial behavior coding by psychologists, it can be seen that the local detail information of the face determines the meaning of facial expression. Therefore, a facial expression recognition method based on multi-scale detail enhancement is proposed, because facial expression is much more affected by the image details than other information, the method proposed in this paper extracts the image detail information with the Gaussian pyramid firstly, thus the image is enhanced in detail to enrich the facial expression information. Secondly, for the local characteristics of facial expressions, a local gradient feature calculation method is proposed based on hierarchical structure to describe the local shape features of facial feature points. Finally, facial expressions are classified using a Support Vector Machine (SVM). The experimental results in the CK+ expression database show that the method not only proves the important role of image detail in facial expression recognition, but also obtains very good recognition results under small-scale training data. The average recognition rate of expressions reaches 98.19%.
Facial expression is the most intuitive description of changes in psychological emotions, and different people have great differences in facial expressions. The existing facial expression recognition methods use facial statistical features to distinguish among different expressions, but these methods are short of deep exploration for facial detail information. According to the definition of facial behavior coding by psychologists, it can be seen that the local detail information of the face determines the meaning of facial expression. Therefore, a facial expression recognition method based on multi-scale detail enhancement is proposed, because facial expression is much more affected by the image details than other information, the method proposed in this paper extracts the image detail information with the Gaussian pyramid firstly, thus the image is enhanced in detail to enrich the facial expression information. Secondly, for the local characteristics of facial expressions, a local gradient feature calculation method is proposed based on hierarchical structure to describe the local shape features of facial feature points. Finally, facial expressions are classified using a Support Vector Machine (SVM). The experimental results in the CK+ expression database show that the method not only proves the important role of image detail in facial expression recognition, but also obtains very good recognition results under small-scale training data. The average recognition rate of expressions reaches 98.19%.
2019, 41(11): 2760-2767.
doi: 10.11999/JEIT180926
Abstract:
By using the intra-view and inter-view correlations and the motion vector-sharing, a depth map error concealment approach is proposed for 3D video coding based on the High Efficiency Video Coding (3D-HEVC) to combat the packet loss of the depth video transmission. Based on the Hierarchical B-frame Prediction (HBP) structure in 3D-HEVC and textured features of the depth map, all the lost coding units are firstly categorized into two classes, i.e., motion blocks and static blocks. Then, according to the outer boundary matching criterion combining the texture structure, the optimal motion/disparity vector is chosen for the damaged motion blocks to conduct the motion/disparity compensation based error concealment. Whereas, the direct copy is applied to concealling the damaged static blocks quickly. Finally, for the concealed blocks whose qualities are not ideal, the new motion/disparity compensation blocks reconstructed by the reference frames recombination are applied to improning the qualities of those blocks. The experimental results show that the repaired depth map concealed by the proposed approach can achieve 0.25～2.03 dB gain in term of the Peak-Signal-to-Noise Ratio (PSNR) and 0.001～0.006 gain in term of Structural Similarity Index Measure(SSIM). Moreover, the subjective visual quality of the repaired area is better in lines with the original depth maps.
By using the intra-view and inter-view correlations and the motion vector-sharing, a depth map error concealment approach is proposed for 3D video coding based on the High Efficiency Video Coding (3D-HEVC) to combat the packet loss of the depth video transmission. Based on the Hierarchical B-frame Prediction (HBP) structure in 3D-HEVC and textured features of the depth map, all the lost coding units are firstly categorized into two classes, i.e., motion blocks and static blocks. Then, according to the outer boundary matching criterion combining the texture structure, the optimal motion/disparity vector is chosen for the damaged motion blocks to conduct the motion/disparity compensation based error concealment. Whereas, the direct copy is applied to concealling the damaged static blocks quickly. Finally, for the concealed blocks whose qualities are not ideal, the new motion/disparity compensation blocks reconstructed by the reference frames recombination are applied to improning the qualities of those blocks. The experimental results show that the repaired depth map concealed by the proposed approach can achieve 0.25～2.03 dB gain in term of the Peak-Signal-to-Noise Ratio (PSNR) and 0.001～0.006 gain in term of Structural Similarity Index Measure(SSIM). Moreover, the subjective visual quality of the repaired area is better in lines with the original depth maps.
2019, 41(11): 2768-2776.
doi: 10.11999/JEIT181138
Abstract:
To solve the problems that Two-Dimensional Principal Component Analysis (2DPCA) can not implement the on-line feature extraction and can not represent the complete structure information, an Incremental 2DPCA (I2DPCA) without estimating covariance matrices is presented by an iterative estimation method, not to deal with the image covariance matrices by the eigenvalue decomposition or the singular value decomposition. The complexity will be greatly reduced and the on-line feature extraction speed can be improved. The proposed I2DPCA can only extract the horizontal features, and thus another Incremental Row-Column 2DPCA (IRC2DPCA) is proposed to incrementally extract the longitudinal ones from the feature matrices of the I2DPCA. The IRC2DPCA can preserve the horizontal and longitudinal features and implement the dimensionality reduction in both row and column directions. Finally, a series of experiments are carried out with the self-built block dataset, ORL and Yale face datasets, respectively. The results show that the proposed algorithms have significantly improved the performances of the convergence rate, the classification rate and the complexity. The convergence rate is over 99%, the classification rate can reach 97.6% and the average processing speed is about 29 frames per second, and it can meet the on-line feature extraction requirements for incremental learning.
To solve the problems that Two-Dimensional Principal Component Analysis (2DPCA) can not implement the on-line feature extraction and can not represent the complete structure information, an Incremental 2DPCA (I2DPCA) without estimating covariance matrices is presented by an iterative estimation method, not to deal with the image covariance matrices by the eigenvalue decomposition or the singular value decomposition. The complexity will be greatly reduced and the on-line feature extraction speed can be improved. The proposed I2DPCA can only extract the horizontal features, and thus another Incremental Row-Column 2DPCA (IRC2DPCA) is proposed to incrementally extract the longitudinal ones from the feature matrices of the I2DPCA. The IRC2DPCA can preserve the horizontal and longitudinal features and implement the dimensionality reduction in both row and column directions. Finally, a series of experiments are carried out with the self-built block dataset, ORL and Yale face datasets, respectively. The results show that the proposed algorithms have significantly improved the performances of the convergence rate, the classification rate and the complexity. The convergence rate is over 99%, the classification rate can reach 97.6% and the average processing speed is about 29 frames per second, and it can meet the on-line feature extraction requirements for incremental learning.
2019, 41(11): 2777-2786.
doi: 10.11999/JEIT190056
Abstract:
An image semantic segmentation model based on region and deep residual network is proposed. Region based methods use multi-scale to create overlapping regions, which can identify multi-scale objects and obtain fine object segmentation boundary. Fully convolutional methods learn features automatically by using Convolutional Neural Network (CNN) to perform end-to-end training for pixel classification tasks, but typically produce coarse segmentation boundaries. The advantages of these two methods are combined: firstly, candidate regions are generated by region generation network, and then the image is fed through the deep residual network with dilated convolution to obtain the feature map. Then the candidate regions and the feature maps are combined to get the features of the regions, and the features are mapped to each pixel in the regions. Finally, the global average pooling layer is used to classify pixels. Multiple different models are obtained by training with different sizes of candidate region inputs. When testing, the final segmentation are obtained by fusing the classification results of these models. The experimental results on SIFT FLOW and PASCAL Context datasets show that the proposed method has higher average accuracy than some state-of-the-art algorithms.
An image semantic segmentation model based on region and deep residual network is proposed. Region based methods use multi-scale to create overlapping regions, which can identify multi-scale objects and obtain fine object segmentation boundary. Fully convolutional methods learn features automatically by using Convolutional Neural Network (CNN) to perform end-to-end training for pixel classification tasks, but typically produce coarse segmentation boundaries. The advantages of these two methods are combined: firstly, candidate regions are generated by region generation network, and then the image is fed through the deep residual network with dilated convolution to obtain the feature map. Then the candidate regions and the feature maps are combined to get the features of the regions, and the features are mapped to each pixel in the regions. Finally, the global average pooling layer is used to classify pixels. Multiple different models are obtained by training with different sizes of candidate region inputs. When testing, the final segmentation are obtained by fusing the classification results of these models. The experimental results on SIFT FLOW and PASCAL Context datasets show that the proposed method has higher average accuracy than some state-of-the-art algorithms.
2019, 41(11): 2787-2794.
doi: 10.11999/JEIT181076
Abstract:
As an extension of Compressed Sensing(CS), Matrix Completion(MC) is widely applied to different fields. Recently, the Riemannian optimization based MC algorithm attracts a lot of attention from researchers due to its high accuracy in reconstruction and computational efficiency. Considering that the Riemannian optimization based MC algorithm assumes a fixed rank of the original matrix, and selects a random initial point for iteration, a novel algorithm is proposed, namely automatic rank estimation based Riemannian optimization matrix completion algorithm. In the proposed algorithm, the estimate of rank is obtained minimizing the objective function that involving the rank regulation, in addition, the iterative starting point is optimized based on Riemannian manifold. The Riemannian manifold based conjugate gradient method is then used to complete the matrix, thereby improving the reconstruction precision. The experimental results demonstrate that the image completion performance is significantly improved using the proposed algorithm, compared with several classical image completion methods.
As an extension of Compressed Sensing(CS), Matrix Completion(MC) is widely applied to different fields. Recently, the Riemannian optimization based MC algorithm attracts a lot of attention from researchers due to its high accuracy in reconstruction and computational efficiency. Considering that the Riemannian optimization based MC algorithm assumes a fixed rank of the original matrix, and selects a random initial point for iteration, a novel algorithm is proposed, namely automatic rank estimation based Riemannian optimization matrix completion algorithm. In the proposed algorithm, the estimate of rank is obtained minimizing the objective function that involving the rank regulation, in addition, the iterative starting point is optimized based on Riemannian manifold. The Riemannian manifold based conjugate gradient method is then used to complete the matrix, thereby improving the reconstruction precision. The experimental results demonstrate that the image completion performance is significantly improved using the proposed algorithm, compared with several classical image completion methods.
