Email alert
2023 Vol. 45, No. 5
Display Method:
2023, 45(5): 1529-1540.
doi: 10.11999/JEIT220887
Abstract:
HyperSpectral Image(HSI) has nanometer-level spectral discriminative ability, capturing the spectral and spatial information of the ground objects simultaneously, within the integration of three-dimensional image cube. The capability to finely sense the intrinsic properties of objects makes it universally applied to many fields, e.g., remote sensing & detection, medical imaging & diagnosis, military defense & security, etc. Different from traditional one-dimensional time-series signals and two-dimensional image signals, HSIs are third-order tensor signals, with the spectral bands in the third-mode being high-dimensional. To eliminate the deficiencies of existing techniques in solving HSI processing and interpretation problems, Graph Signal Processing (GSP) is introduced. A short overview of the theoretical and technological development of GSP is given, along with its typical applications in HSI feature extraction, restoration, and classification. Based on the survey of the existing research basis, the future challenges and potential approaches to solve them in the community are also pointed out and discussed.
HyperSpectral Image(HSI) has nanometer-level spectral discriminative ability, capturing the spectral and spatial information of the ground objects simultaneously, within the integration of three-dimensional image cube. The capability to finely sense the intrinsic properties of objects makes it universally applied to many fields, e.g., remote sensing & detection, medical imaging & diagnosis, military defense & security, etc. Different from traditional one-dimensional time-series signals and two-dimensional image signals, HSIs are third-order tensor signals, with the spectral bands in the third-mode being high-dimensional. To eliminate the deficiencies of existing techniques in solving HSI processing and interpretation problems, Graph Signal Processing (GSP) is introduced. A short overview of the theoretical and technological development of GSP is given, along with its typical applications in HSI feature extraction, restoration, and classification. Based on the survey of the existing research basis, the future challenges and potential approaches to solve them in the community are also pointed out and discussed.
2023, 45(5): 1541-1549.
doi: 10.11999/JEIT220139
Abstract:
Recently, ship target detection in Synthetic Aperture Radar (SAR) imagery based on deep learning has been widely developed. However, a large number of labeled samples are needed in traditionally supervised learning to train the network. Therefore, a semi-supervised SAR ship target detection approach based on Graph ATtention network (GAT) is proposed. Firstly, a symmetric convolutional neural network is designed to realize land-ocean segmentation. Secondly, the super-pixel segmentation is completed and the super-pixels are modeled as nodes of the GAT. The multi-scale features of a node are extracted by region of interest pooling layer. Attentional mechanisms are used in GAT to concatenate adaptively the neighbor node’s features and classify the unlabeled nodes. Finally, the super-pixels predicted as ship targets are located in SAR image and the fine detection results are obtained. The proposed method is verified on the measured high resolution SAR images dataset. The results show that this method can effectively detect ship targets with low false alarm rate by using a small number of labeled samples.
Recently, ship target detection in Synthetic Aperture Radar (SAR) imagery based on deep learning has been widely developed. However, a large number of labeled samples are needed in traditionally supervised learning to train the network. Therefore, a semi-supervised SAR ship target detection approach based on Graph ATtention network (GAT) is proposed. Firstly, a symmetric convolutional neural network is designed to realize land-ocean segmentation. Secondly, the super-pixel segmentation is completed and the super-pixels are modeled as nodes of the GAT. The multi-scale features of a node are extracted by region of interest pooling layer. Attentional mechanisms are used in GAT to concatenate adaptively the neighbor node’s features and classify the unlabeled nodes. Finally, the super-pixels predicted as ship targets are located in SAR image and the fine detection results are obtained. The proposed method is verified on the measured high resolution SAR images dataset. The results show that this method can effectively detect ship targets with low false alarm rate by using a small number of labeled samples.
2023, 45(5): 1550-1558.
doi: 10.11999/JEIT220303
Abstract:
Recent researches on semi-supervised bearing fault diagnosis based on Graph Neural Network (GNN) still have some problems, such as insufficient label information mining and relatively ideal diagnosis scenarios. In engineering practice, bearings are often operated under time-varying speeds such as startup and shutdown, and fault label samples become increasingly expensive. In response to the above challenges, a new method called semi-supervised bearing fault diagnosis using improved Graph ATtention network (GAT) under time-varying speeds is proposed. Based on K-Nearest Neighbor (KNN) algorithm and Smoothing Assumption (SA), the pseudo-label propagation strategy is designed to spread the label information to the neighborhood samples with similar distribution along the edge, so that the label information hidden in the limited samples can be fully utilized. Each vibration spectrum sample is considered as a node, and a semi-supervised learning model based on node-level GAN is constructed to explore further representative bearing fault features through the attention mechanism. The proposed method is applied to analyze two sets of bearing fault experimental data under time-varying speed, and the results show that the proposed method is able to diagnose accurately different fault modes of bearings at low label rates of no more than 2%, which is better than other commonly used semi-supervised learning methods of GNN.
Recent researches on semi-supervised bearing fault diagnosis based on Graph Neural Network (GNN) still have some problems, such as insufficient label information mining and relatively ideal diagnosis scenarios. In engineering practice, bearings are often operated under time-varying speeds such as startup and shutdown, and fault label samples become increasingly expensive. In response to the above challenges, a new method called semi-supervised bearing fault diagnosis using improved Graph ATtention network (GAT) under time-varying speeds is proposed. Based on K-Nearest Neighbor (KNN) algorithm and Smoothing Assumption (SA), the pseudo-label propagation strategy is designed to spread the label information to the neighborhood samples with similar distribution along the edge, so that the label information hidden in the limited samples can be fully utilized. Each vibration spectrum sample is considered as a node, and a semi-supervised learning model based on node-level GAN is constructed to explore further representative bearing fault features through the attention mechanism. The proposed method is applied to analyze two sets of bearing fault experimental data under time-varying speed, and the results show that the proposed method is able to diagnose accurately different fault modes of bearings at low label rates of no more than 2%, which is better than other commonly used semi-supervised learning methods of GNN.
2023, 45(5): 1559-1566.
doi: 10.11999/JEIT220970
Abstract:
In most target localization applications, achieving high spatial resolution on angle and range is requested. Addressing this demand, a novel Graph Signal Processing (GSP) based target localization method for monostatic Frequency Diverse Array (FDA) is proposed in this paper. Firstly, a directed graph model applicable to the FDA is established based on the array geometry and the signal correlations among array elements, in which echoes received in the array are mapped to a graph signal. By leveraging the concept of the graph Fourier transform, the obtained graph signal is decomposed into a set of spectrums, and then the joint angle and range estimation can be solved successfully using a well-designed two-dimensional spectral peak search. The simulation results illustrate the validity and effectiveness of the proposed method, and it is shown that the proposed method outperforms the existing methods in estimation accuracy and is capable to achieve performance improvement for the weak target in a low Signal-to-Noise Ratio (SNR) environment.
In most target localization applications, achieving high spatial resolution on angle and range is requested. Addressing this demand, a novel Graph Signal Processing (GSP) based target localization method for monostatic Frequency Diverse Array (FDA) is proposed in this paper. Firstly, a directed graph model applicable to the FDA is established based on the array geometry and the signal correlations among array elements, in which echoes received in the array are mapped to a graph signal. By leveraging the concept of the graph Fourier transform, the obtained graph signal is decomposed into a set of spectrums, and then the joint angle and range estimation can be solved successfully using a well-designed two-dimensional spectral peak search. The simulation results illustrate the validity and effectiveness of the proposed method, and it is shown that the proposed method outperforms the existing methods in estimation accuracy and is capable to achieve performance improvement for the weak target in a low Signal-to-Noise Ratio (SNR) environment.
2023, 45(5): 1567-1574.
doi: 10.11999/JEIT220188
Abstract:
The marine physical environment and electromagnetic environment are becoming increasingly complex, making the weak and slow small target detection in the sea clutter background be both emphasis and difficulty of radar target detection research. Due to small radar cross sections and low energy of small targets on the sea surface, traditional energy-based detection methods have a performance bottleneck. Feature-based detection methods focus on extracting distinguishing features between pure sea clutter and target returns to achieve target detection, which improve effectively the detection performance. Using the correlation of amplitude of radar returns in frequency domain, the graph theory method to feature-based detection is introduced. Firstly, measured sea clutter data are block-whitened to suppress sea clutter. Then, the data from Doppler channels are extracted in the frequency domain. With the help of graph processing methods, a distance adjacency matrix of the extracted data is constructed, and then it is converted into a Laplacian matrix. The maximum eigenvalue of Laplacian matrix under different radar returns is calculated, and fused with the relative Doppler peak height, then a new test statistic is obtained. By comparing the value of test statistics, sea clutter and returns with targets can be distinguished. Verified by the measured Ice multiParameter Imaging X-band (IPIX) database, the proposed detector attains better detection performance.
The marine physical environment and electromagnetic environment are becoming increasingly complex, making the weak and slow small target detection in the sea clutter background be both emphasis and difficulty of radar target detection research. Due to small radar cross sections and low energy of small targets on the sea surface, traditional energy-based detection methods have a performance bottleneck. Feature-based detection methods focus on extracting distinguishing features between pure sea clutter and target returns to achieve target detection, which improve effectively the detection performance. Using the correlation of amplitude of radar returns in frequency domain, the graph theory method to feature-based detection is introduced. Firstly, measured sea clutter data are block-whitened to suppress sea clutter. Then, the data from Doppler channels are extracted in the frequency domain. With the help of graph processing methods, a distance adjacency matrix of the extracted data is constructed, and then it is converted into a Laplacian matrix. The maximum eigenvalue of Laplacian matrix under different radar returns is calculated, and fused with the relative Doppler peak height, then a new test statistic is obtained. By comparing the value of test statistics, sea clutter and returns with targets can be distinguished. Verified by the measured Ice multiParameter Imaging X-band (IPIX) database, the proposed detector attains better detection performance.
2023, 45(5): 1575-1584.
doi: 10.11999/JEIT220840
Abstract:
Considering the shortcomings of the Adaptive Visibility Graph (AVG) algorithm being too complex and the accuracy improvement is not significant, an Automatic Modulation Recognition(AMR) framework based on Single-channel Multi-scale Graph Neural Network (SMGNN) is proposed and interpretability studies are conducted on the various parts of the framework. Firstly, the multi-layer perceptron and one-dimensional convolutional adaptive are used to realize the mapping between single-channel signal sequences and graphs, which reduces effectively the complexity of AVG algorithms. Secondly, a multi-scale graph neural network is designed to fuse the features of different resolutions, which improves the accuracy of model recognition. Experiments show that the SMGNN algorithm proposed in this paper saves nearly half of the parameter amount compared with the AVG algorithm, and the recognition accuracy has been greatly improved.
Considering the shortcomings of the Adaptive Visibility Graph (AVG) algorithm being too complex and the accuracy improvement is not significant, an Automatic Modulation Recognition(AMR) framework based on Single-channel Multi-scale Graph Neural Network (SMGNN) is proposed and interpretability studies are conducted on the various parts of the framework. Firstly, the multi-layer perceptron and one-dimensional convolutional adaptive are used to realize the mapping between single-channel signal sequences and graphs, which reduces effectively the complexity of AVG algorithms. Secondly, a multi-scale graph neural network is designed to fuse the features of different resolutions, which improves the accuracy of model recognition. Experiments show that the SMGNN algorithm proposed in this paper saves nearly half of the parameter amount compared with the AVG algorithm, and the recognition accuracy has been greatly improved.
2023, 45(5): 1585-1592.
doi: 10.11999/JEIT221194
Abstract:
For the reconstruction of large-scale network data, a Distributed Batch Reconstruction algorithm via Sobolev Smoothness on Cartesian product graph (DBR-SSC) is proposed, which is based on the Graph Signal Processing (GSP) theory. In the proposed algorithm, the time-varying graph signal is firstly divided into multiple signal segments in time dimension, and a product graph is constructed from graphs at each time instant via Cartesian product. Secondly, the reconstruction of the time-varying graph signal in each segment is formulated as an optimization problem by exploiting the Sobolev difference smoothness on the Cartesian product graph. Finally, a distributed algorithm with high convergence rate is devised to solve the optimization problem. Two real world data sets are used for experiments, and it is shown that the proposed algorithm has low reconstruction error and high convergence rate.
For the reconstruction of large-scale network data, a Distributed Batch Reconstruction algorithm via Sobolev Smoothness on Cartesian product graph (DBR-SSC) is proposed, which is based on the Graph Signal Processing (GSP) theory. In the proposed algorithm, the time-varying graph signal is firstly divided into multiple signal segments in time dimension, and a product graph is constructed from graphs at each time instant via Cartesian product. Secondly, the reconstruction of the time-varying graph signal in each segment is formulated as an optimization problem by exploiting the Sobolev difference smoothness on the Cartesian product graph. Finally, a distributed algorithm with high convergence rate is devised to solve the optimization problem. Two real world data sets are used for experiments, and it is shown that the proposed algorithm has low reconstruction error and high convergence rate.
2023, 45(5): 1593-1601.
doi: 10.11999/JEIT220854
Abstract:
In the post-Moore era, Chiplet is the most hottest integration technique for heterogeneous integrated circuit, which is characterized by complex multi-core stacked structures. In order to solve the post-bonding test problem of Chiplet in different stacked structures, a Universal Test Access Port Controller (UTAPC) circuit is proposed based on IEEE 1838 standard protocol. Based on the traditional Test Access Port (TAP) controller, the Chiplet Dedicated Finite State Machine (CDFSM) is designed, also the Chiplet configuration registers and Chiplet test interface circuit are added. Under the influence of the configuration registers’ control signals generated by the CDFSM, the configuration signals outputted from the Chiplet configuration registers are used to control the Chiplet test interface circuit to set up the effective test path of Chiplet, which realized to access cores cross layers. The simulation results demonstrate that the proposed UTAPC circuit is suitable for the design for test of Chiplet with arbitrary stacked structures. It can not only choose to test cores flexibly, but also save the resources of test ports and test time, as well as improve the test efficiency.
In the post-Moore era, Chiplet is the most hottest integration technique for heterogeneous integrated circuit, which is characterized by complex multi-core stacked structures. In order to solve the post-bonding test problem of Chiplet in different stacked structures, a Universal Test Access Port Controller (UTAPC) circuit is proposed based on IEEE 1838 standard protocol. Based on the traditional Test Access Port (TAP) controller, the Chiplet Dedicated Finite State Machine (CDFSM) is designed, also the Chiplet configuration registers and Chiplet test interface circuit are added. Under the influence of the configuration registers’ control signals generated by the CDFSM, the configuration signals outputted from the Chiplet configuration registers are used to control the Chiplet test interface circuit to set up the effective test path of Chiplet, which realized to access cores cross layers. The simulation results demonstrate that the proposed UTAPC circuit is suitable for the design for test of Chiplet with arbitrary stacked structures. It can not only choose to test cores flexibly, but also save the resources of test ports and test time, as well as improve the test efficiency.
2023, 45(5): 1602-1610.
doi: 10.11999/JEIT220448
Abstract:
For radar maritime target detection method of feature class, the convex hull classification algorithm is usually used in existing three feature detectors to complete detection. It is found that the decision region generated by convex hull learning algorithm may not well reflect the distribution of sea clutter samples in feature space in actual application, which may cause a certain degree of performance loss. By contrast, the decision region generated by concave hull algorithm is dug from convex hull, which can fit the distribution of sea clutter samples better. Therefore, in this paper, the form of the decision region is transformed from convex hull to concave hull. On this basis, a small target detection method based on 3-D concave hull learning algorithm is proposed. However, the existing 3-D concave hull algorithm has the disadvantages of low efficiency and unable to realize constant false alarm detection. To solve this problem, this paper improves the algorithm by optimizing the selection method of digging point and adding a process named "external complement". Finally, the measured CSIR datasets and X-band experimental radar data verify that the performance of proposed detection methods is superior to existing detection methods when other parameters are the same. At the same time, the analysis of algorithm complexity proves the application potential of proposed method.
For radar maritime target detection method of feature class, the convex hull classification algorithm is usually used in existing three feature detectors to complete detection. It is found that the decision region generated by convex hull learning algorithm may not well reflect the distribution of sea clutter samples in feature space in actual application, which may cause a certain degree of performance loss. By contrast, the decision region generated by concave hull algorithm is dug from convex hull, which can fit the distribution of sea clutter samples better. Therefore, in this paper, the form of the decision region is transformed from convex hull to concave hull. On this basis, a small target detection method based on 3-D concave hull learning algorithm is proposed. However, the existing 3-D concave hull algorithm has the disadvantages of low efficiency and unable to realize constant false alarm detection. To solve this problem, this paper improves the algorithm by optimizing the selection method of digging point and adding a process named "external complement". Finally, the measured CSIR datasets and X-band experimental radar data verify that the performance of proposed detection methods is superior to existing detection methods when other parameters are the same. At the same time, the analysis of algorithm complexity proves the application potential of proposed method.
2023, 45(5): 1611-1618.
doi: 10.11999/JEIT220475
Abstract:
Even though the internal and external SAR system calibration is very accurate, the accuracy of airborne fully polarimetric SAR measurement still changes to some extent under different flight conditions. In the condition of non-stationary and high frequency, the accuracy deteriorates seriously. In order to solve this problem, an error model of fully polarimetric SAR in non-stationary environment is proposed, and then the influence of the slight variation of trajectory between channels on the polarization phase imbalance degree in time-sharing transceiver system is analyzed. It is pointed out that the same motion error will aggravate the phase imbalance with the increase of the band. The corresponding processing method is proposed accordingly. Finally, the effectiveness of this method is verified by simulation and S-band SAR data. The effectiveness and stability of this method are also verified by several application demonstrations.
Even though the internal and external SAR system calibration is very accurate, the accuracy of airborne fully polarimetric SAR measurement still changes to some extent under different flight conditions. In the condition of non-stationary and high frequency, the accuracy deteriorates seriously. In order to solve this problem, an error model of fully polarimetric SAR in non-stationary environment is proposed, and then the influence of the slight variation of trajectory between channels on the polarization phase imbalance degree in time-sharing transceiver system is analyzed. It is pointed out that the same motion error will aggravate the phase imbalance with the increase of the band. The corresponding processing method is proposed accordingly. Finally, the effectiveness of this method is verified by simulation and S-band SAR data. The effectiveness and stability of this method are also verified by several application demonstrations.
2023, 45(5): 1619-1626.
doi: 10.11999/JEIT220308
Abstract:
In-Band Full Duplex (IBFD) technique is expected to be the most potential scheme for modern wireless communication system, since it can prove the spectral efficiency. However, in the application process, it faces the great challenge of Self-Interference Cancellation (SIC). SIC can be realized separately or in combination from three aspects: propagation domain, analog domain and digital domain. This paper focuses on digital SIC in the IBFD. To solve the problem that the performance of the traditional digital SIC is limited by the non-ideality of the components of transceiver link, this paper adopts an IBFD system of Radio Frequency (RF) auxiliary link. By exploiting the boundedness of signal of interest and self-interfering signal, a digital SIC algorithm based on bounded component analysis is developed. Under the two channel scenarios of the Line Of Sight (LOS) and the Non-Line Of Sight (NLOS), the simulation and measured data are used to verify and analyze. The results show that compared with the least square method and independent component analysis method, the proposed bounded component analysis method improves the SIC effect and improves the bit error rate performance of the system.
In-Band Full Duplex (IBFD) technique is expected to be the most potential scheme for modern wireless communication system, since it can prove the spectral efficiency. However, in the application process, it faces the great challenge of Self-Interference Cancellation (SIC). SIC can be realized separately or in combination from three aspects: propagation domain, analog domain and digital domain. This paper focuses on digital SIC in the IBFD. To solve the problem that the performance of the traditional digital SIC is limited by the non-ideality of the components of transceiver link, this paper adopts an IBFD system of Radio Frequency (RF) auxiliary link. By exploiting the boundedness of signal of interest and self-interfering signal, a digital SIC algorithm based on bounded component analysis is developed. Under the two channel scenarios of the Line Of Sight (LOS) and the Non-Line Of Sight (NLOS), the simulation and measured data are used to verify and analyze. The results show that compared with the least square method and independent component analysis method, the proposed bounded component analysis method improves the SIC effect and improves the bit error rate performance of the system.
2023, 45(5): 1627-1634.
doi: 10.11999/JEIT220464
Abstract:
Phase noise limits the cancellation capability of a Full-Duplex (FD) transceiver receiver and degrades the demodulation performance of the signal-of-interest, even for the transceiver receiver which deploy one common oscillator for its transmitter and its receiver. In order to mitigate the phase noises contained in the multipath Self-Interference (SI) components, a multiple-downconversion FD transceiver receiver design is proposed to suppress these phase noises. The proposed multiple-downconversion FD transceiver receiver design includes a new FD transceiver architecture with multiple receive chains, and a phase noise cancellation algorithm. The FD transceiver architecture deploys multiple receive chains to downconvert the signal received by one antenna. Particularly, the oscillator signal of each receive chain is originated from the transmit oscillator with a unique delay, such that the phase noises contained in the multipath SI components can be compensated. The phase noise cancellation algorithm deploys the received signals in different receive chains to estimate the phase noise coefficients, which can cancel the residual phase noise after the multiple-downconversion. The cancellation capability of the cancellation algorithm is derived and analyzed. Analytical and simulation results demonstrate that, in the scenario that the number of receive chains is greater than the number of strong multiple SI components, the phase noise does not affect the cancellation capability of proposed FD transceiver.
Phase noise limits the cancellation capability of a Full-Duplex (FD) transceiver receiver and degrades the demodulation performance of the signal-of-interest, even for the transceiver receiver which deploy one common oscillator for its transmitter and its receiver. In order to mitigate the phase noises contained in the multipath Self-Interference (SI) components, a multiple-downconversion FD transceiver receiver design is proposed to suppress these phase noises. The proposed multiple-downconversion FD transceiver receiver design includes a new FD transceiver architecture with multiple receive chains, and a phase noise cancellation algorithm. The FD transceiver architecture deploys multiple receive chains to downconvert the signal received by one antenna. Particularly, the oscillator signal of each receive chain is originated from the transmit oscillator with a unique delay, such that the phase noises contained in the multipath SI components can be compensated. The phase noise cancellation algorithm deploys the received signals in different receive chains to estimate the phase noise coefficients, which can cancel the residual phase noise after the multiple-downconversion. The cancellation capability of the cancellation algorithm is derived and analyzed. Analytical and simulation results demonstrate that, in the scenario that the number of receive chains is greater than the number of strong multiple SI components, the phase noise does not affect the cancellation capability of proposed FD transceiver.
2023, 45(5): 1635-1643.
doi: 10.11999/JEIT220352
Abstract:
In recent years, the deployment of Unmanned Aerial Vehicles (UAVs) equipped with Mobile Edge Computing (MEC) servers to provide computing services for ground users has become an emerging method. Considering an UAV-assisted MEC system with multi-users, a scheme is investigated to minimize the average energy consumption for all users to complete their computation tasks via optimizing the trajectory of UAV and computation strategies of the users during the UAV’s whole flight duration. A Deep Reinforcement Learning (DRL)-based Soft Actor-Critic (SAC) algorithm is proposed to tackle the energy consumption optimization problem. With the iteration of the network training procedure, the best action is obtained according to the maximum entropy rule, which does not neglect any action with high reward value and thus can enhance the exploration and convergence performance of the proposed algorithm. Simulation results reveal that the proposed SAC algorithm can effectively decrease the average energy consumption of all users and achieves better stability and convergence performance, as compared to some existing baseline algorithms.
In recent years, the deployment of Unmanned Aerial Vehicles (UAVs) equipped with Mobile Edge Computing (MEC) servers to provide computing services for ground users has become an emerging method. Considering an UAV-assisted MEC system with multi-users, a scheme is investigated to minimize the average energy consumption for all users to complete their computation tasks via optimizing the trajectory of UAV and computation strategies of the users during the UAV’s whole flight duration. A Deep Reinforcement Learning (DRL)-based Soft Actor-Critic (SAC) algorithm is proposed to tackle the energy consumption optimization problem. With the iteration of the network training procedure, the best action is obtained according to the maximum entropy rule, which does not neglect any action with high reward value and thus can enhance the exploration and convergence performance of the proposed algorithm. Simulation results reveal that the proposed SAC algorithm can effectively decrease the average energy consumption of all users and achieves better stability and convergence performance, as compared to some existing baseline algorithms.
2023, 45(5): 1644-1650.
doi: 10.11999/JEIT220761
Abstract:
In view of the surge in data traffic and the diversified needs of users in the future 6G network, using UAV to assist the cellular network can provide users with better services. This paper proposes a UAV trajectory planning and resource allocation joint optimization method based on content-aware. Hot content is cached on UAV. Under the condition of satisfying the user’s content demand, user association and UAV trajectory are jointly optimized to maximize the minimum average service rate of users. Since the established optimization problem is non-convex, a block coordinate descent method is proposed to decompose the original problem into two sub-problems and the trajectory planning problem is solved by continuous convex optimization method. The simulation results show that the proposed method can effectively improve the minimum user average service rate and the network depth coverage level.
In view of the surge in data traffic and the diversified needs of users in the future 6G network, using UAV to assist the cellular network can provide users with better services. This paper proposes a UAV trajectory planning and resource allocation joint optimization method based on content-aware. Hot content is cached on UAV. Under the condition of satisfying the user’s content demand, user association and UAV trajectory are jointly optimized to maximize the minimum average service rate of users. Since the established optimization problem is non-convex, a block coordinate descent method is proposed to decompose the original problem into two sub-problems and the trajectory planning problem is solved by continuous convex optimization method. The simulation results show that the proposed method can effectively improve the minimum user average service rate and the network depth coverage level.
2023, 45(5): 1651-1659.
doi: 10.11999/JEIT220442
Abstract:
To address the path planning and inter-aircraft collision avoidance problems in Unmanned Aerial Vehicle (UAV) formation reconstruction in complex electromagnetic environment, the repulsion function is improved by using distance factor on the basis of traditional UV virtual potential field, and a non-uniform UV virtual potential field is constructed to assist UAVs in collision avoidance in this paper. The improved UV non-uniform virtual potential field can make the collision avoidance path of UAV smoother, and the UAV can fly a longer distance in the same time. In addition, the distance between UAVs is calculated by the wireless ultraviolet ranging method, and the traditional artificial potential field method is improved by combining the ultraviolet non-uniform potential field to realize the formation reconstruction of UAVs. Simulation results show that this algorithm can effectively solve the path oscillation and local minimum problems under the traditional algorithm, while the collision avoidance efficiency is significantly improved compared with the traditional artificial potential field algorithm, and the distance is shortened by 6% in the preset environment while the time to reach the target point is advanced by 40%. The results show that the algorithm can effectively achieve the expected inter-aircraft collision avoidance effect in UAV formation reconstruction.
To address the path planning and inter-aircraft collision avoidance problems in Unmanned Aerial Vehicle (UAV) formation reconstruction in complex electromagnetic environment, the repulsion function is improved by using distance factor on the basis of traditional UV virtual potential field, and a non-uniform UV virtual potential field is constructed to assist UAVs in collision avoidance in this paper. The improved UV non-uniform virtual potential field can make the collision avoidance path of UAV smoother, and the UAV can fly a longer distance in the same time. In addition, the distance between UAVs is calculated by the wireless ultraviolet ranging method, and the traditional artificial potential field method is improved by combining the ultraviolet non-uniform potential field to realize the formation reconstruction of UAVs. Simulation results show that this algorithm can effectively solve the path oscillation and local minimum problems under the traditional algorithm, while the collision avoidance efficiency is significantly improved compared with the traditional artificial potential field algorithm, and the distance is shortened by 6% in the preset environment while the time to reach the target point is advanced by 40%. The results show that the algorithm can effectively achieve the expected inter-aircraft collision avoidance effect in UAV formation reconstruction.
2023, 45(5): 1660-1668.
doi: 10.11999/JEIT220436
Abstract:
A hybrid mapping scheme is proposed in this paper to improve further the Bit Error Rate (BER) performance of Orthogonal Frequency Division Multiplexing with Index Modulation (OFDM-IM) system for high Signal-to-Noise Ratio (SNR). By taking advantage of the fact that the BER of index bits is lower than that of data bits in the high SNR region, the scheme uses two allocation strategies of index bits with different numbers, which increases the proportion of index bits. Correspondingly, the Subcarrier Activation Patterns (SAPs) are divided into super SAPs and normal SAPs. Compared with normal SAPs, super SAPs have seen increased index bits and decreased data bits by 1 bit respectively. The decreased data bits corresponding to a super SAP are supplemented with a parity-check bit and then mapped into symbols, which improves the diversity order of OFDM-IM system and increases the minimum Euclidean distance between symbols. The simulation results show that, compared with the previous mapping scheme, the hybrid mapping scheme can gain about 1～3 dB when BER is 10–4, and improve effectively the BER performance of OFDM-IM system.
A hybrid mapping scheme is proposed in this paper to improve further the Bit Error Rate (BER) performance of Orthogonal Frequency Division Multiplexing with Index Modulation (OFDM-IM) system for high Signal-to-Noise Ratio (SNR). By taking advantage of the fact that the BER of index bits is lower than that of data bits in the high SNR region, the scheme uses two allocation strategies of index bits with different numbers, which increases the proportion of index bits. Correspondingly, the Subcarrier Activation Patterns (SAPs) are divided into super SAPs and normal SAPs. Compared with normal SAPs, super SAPs have seen increased index bits and decreased data bits by 1 bit respectively. The decreased data bits corresponding to a super SAP are supplemented with a parity-check bit and then mapped into symbols, which improves the diversity order of OFDM-IM system and increases the minimum Euclidean distance between symbols. The simulation results show that, compared with the previous mapping scheme, the hybrid mapping scheme can gain about 1～3 dB when BER is 10–4, and improve effectively the BER performance of OFDM-IM system.
2023, 45(5): 1669-1677.
doi: 10.11999/JEIT220347
Abstract:
To address the single point failure of group key management in the existing hierarchical Unmanned Aerial Vehicle (UAV) network, and the incapability of timely group key calculation and update caused by offline group members, a decentralized group key management scheme that supports asynchronous computing is proposed. The Asynchronous Ratchet Tree (ART) protocol is adopted to realize the pre-deployment of the group key, in which asynchronous calculation and autonomous update of the group key can be performed by each member. The decentralization feature of the blockchain technology is employed to solve the single point failure problem, which improves the transparency and fairness of group key management. The performance evaluation shows that, compared with existing schemes, the cluster member UAV in this scheme has lower computational overhead and communication overhead, and are suitable for application in a hierarchical UAV network environment.
To address the single point failure of group key management in the existing hierarchical Unmanned Aerial Vehicle (UAV) network, and the incapability of timely group key calculation and update caused by offline group members, a decentralized group key management scheme that supports asynchronous computing is proposed. The Asynchronous Ratchet Tree (ART) protocol is adopted to realize the pre-deployment of the group key, in which asynchronous calculation and autonomous update of the group key can be performed by each member. The decentralization feature of the blockchain technology is employed to solve the single point failure problem, which improves the transparency and fairness of group key management. The performance evaluation shows that, compared with existing schemes, the cluster member UAV in this scheme has lower computational overhead and communication overhead, and are suitable for application in a hierarchical UAV network environment.
2023, 45(5): 1678-1687.
doi: 10.11999/JEIT220240
Abstract:
With the rapid development of Internet of Things (IoT), the deep integration of Artificial Intelligence (AI) and Edge Computing (EC) has formed Edge AI. However, since IoT devices are computationally and communicationally constrained and these devices often require privacy-preserving, it is still a challenge to accelerate Edge AI while protecting privacy. Federated Learning (FL), an emerging distributed learning paradigm, has great potential in terms of privacy preservation and improving model performance, but communication and local training are inefficient. To address the above challenges, a FL acceleration framework AccFed is proposed in this paper. Firstly, a Device-Edge-Cloud synergy training algorithm based on model partitioning is proposed to accelerate FL local training according to the different network states; Then, a multi-iteration and reaggregation algorithm is designed to accelerate FL aggregation; Finally, experimental results show that AccFed outperforms the control group in terms of training accuracy, convergence speed, training time, etc.
With the rapid development of Internet of Things (IoT), the deep integration of Artificial Intelligence (AI) and Edge Computing (EC) has formed Edge AI. However, since IoT devices are computationally and communicationally constrained and these devices often require privacy-preserving, it is still a challenge to accelerate Edge AI while protecting privacy. Federated Learning (FL), an emerging distributed learning paradigm, has great potential in terms of privacy preservation and improving model performance, but communication and local training are inefficient. To address the above challenges, a FL acceleration framework AccFed is proposed in this paper. Firstly, a Device-Edge-Cloud synergy training algorithm based on model partitioning is proposed to accelerate FL local training according to the different network states; Then, a multi-iteration and reaggregation algorithm is designed to accelerate FL aggregation; Finally, experimental results show that AccFed outperforms the control group in terms of training accuracy, convergence speed, training time, etc.
2023, 45(5): 1688-1696.
doi: 10.11999/JEIT220364
Abstract:
To improve the secure transmission performance of physical layer, a novel Multiple Parameters Weighted-type FRactional Fourier Transform (MP-WFRFT) secure communication method based on Two-Dimensional Cosine Power-Activation (2D-CPA) discrete hyperchaotic encryption is proposed. Firstly, the activation function and cosine function are introduced into the one-dimensional cubic chaos map as nonlinear factors to construct the 2D chaos map. The nonlinear factor can perturb the iterative process of the original cubic chaotic map so as to obtain a fuller phase orbit. The dynamic characteristics of the proposed two-dimensional chaotic mapping are verified by using bifurcation diagram, phase diagram and Lyapunov exponential spectrum. The results show that the constructed 2D chaotic sequence has good randomness and can enter hyperchaotic state. Then, the amplitude transformation matrix, phase rotation matrix and MP-WFRFT parameter pool are constructed by using 2D-CPA hyperchaotic sequence respectively to complete the constellation amplitude phase encryption and MP-WFRFT dynamic transformation encryption process, the statistical characteristics of data are further eliminated and the anti-parameter scanning performance of MP-WFRFT transform is improved. Numerical simulation results show that the constellation diagram of the encrypted data is Gaussian-like and that the transmission system is well sensitive to the key.
To improve the secure transmission performance of physical layer, a novel Multiple Parameters Weighted-type FRactional Fourier Transform (MP-WFRFT) secure communication method based on Two-Dimensional Cosine Power-Activation (2D-CPA) discrete hyperchaotic encryption is proposed. Firstly, the activation function and cosine function are introduced into the one-dimensional cubic chaos map as nonlinear factors to construct the 2D chaos map. The nonlinear factor can perturb the iterative process of the original cubic chaotic map so as to obtain a fuller phase orbit. The dynamic characteristics of the proposed two-dimensional chaotic mapping are verified by using bifurcation diagram, phase diagram and Lyapunov exponential spectrum. The results show that the constructed 2D chaotic sequence has good randomness and can enter hyperchaotic state. Then, the amplitude transformation matrix, phase rotation matrix and MP-WFRFT parameter pool are constructed by using 2D-CPA hyperchaotic sequence respectively to complete the constellation amplitude phase encryption and MP-WFRFT dynamic transformation encryption process, the statistical characteristics of data are further eliminated and the anti-parameter scanning performance of MP-WFRFT transform is improved. Numerical simulation results show that the constellation diagram of the encrypted data is Gaussian-like and that the transmission system is well sensitive to the key.
2023, 45(5): 1697-1705.
doi: 10.11999/JEIT220404
Abstract:
In the case of large gathering of users such as sports venues or sudden disasters, the ground base stations often face the problem of overloading or even paralysing. In this case, the multi-Unmanned Aerial Vehicle (UAV) auxiliary network system can provide the signal compensation for ground base stations and enhance effectively the communication quality in local areas. However, the topology changes induced by the mobility of UAV and the network flows, will lead to frequent intermittent connections or even transmission failures. Therefore, the efficient deployment of UAV base stations, as well as the optimization of network performance, become urgent issues. In this paper, an improved Particle Swarm Optimization (PSO) UAV assisted network deployment optimization algorithm based on the Beetle Antennae Search (BAS), the Intelligent and Efficient Algorithm (IEA), is proposed to improve PSO algorithm by using the individual seeking advantages of BAS algorithm. And for the first time, the double threshold constraint is applied to ensure the communication quality of users, which makes the network performance under the multi-UAV system improved. The simulation results show that, compared with the traditional algorithms, the IEA algorithm proposed in this paper achieves an obvious improvement in terms of the system throughput, the user’s average throughput as well as the spectral efficiency.
In the case of large gathering of users such as sports venues or sudden disasters, the ground base stations often face the problem of overloading or even paralysing. In this case, the multi-Unmanned Aerial Vehicle (UAV) auxiliary network system can provide the signal compensation for ground base stations and enhance effectively the communication quality in local areas. However, the topology changes induced by the mobility of UAV and the network flows, will lead to frequent intermittent connections or even transmission failures. Therefore, the efficient deployment of UAV base stations, as well as the optimization of network performance, become urgent issues. In this paper, an improved Particle Swarm Optimization (PSO) UAV assisted network deployment optimization algorithm based on the Beetle Antennae Search (BAS), the Intelligent and Efficient Algorithm (IEA), is proposed to improve PSO algorithm by using the individual seeking advantages of BAS algorithm. And for the first time, the double threshold constraint is applied to ensure the communication quality of users, which makes the network performance under the multi-UAV system improved. The simulation results show that, compared with the traditional algorithms, the IEA algorithm proposed in this paper achieves an obvious improvement in terms of the system throughput, the user’s average throughput as well as the spectral efficiency.
2023, 45(5): 1706-1713.
doi: 10.11999/JEIT220466
Abstract:
A Reconfigurable Intelligent Surface (RIS)-assisted spectrum sharing Cognitive Radio (CR) Multiple-Input Multiple-Output (MIMO) wireless secure communication system is considered. In the case of eavesdropper, the secondary transmitter equippes with multiple antennas communicates with the secondary user. Firstly, by using the statistical channel state information, the deterministic equivalent expression of the ergodic security rate is derived. Then, on the premise of meeting the constraints of total transmission power and interference power, an alternating optimization algorithm combining Taylor series expansion method and Lagrange multiplier method is proposed to optimize jointly the transmission covariance matrix and phase shift matrix. Finally, simulation results verify the effectiveness of the proposed algorithm.
A Reconfigurable Intelligent Surface (RIS)-assisted spectrum sharing Cognitive Radio (CR) Multiple-Input Multiple-Output (MIMO) wireless secure communication system is considered. In the case of eavesdropper, the secondary transmitter equippes with multiple antennas communicates with the secondary user. Firstly, by using the statistical channel state information, the deterministic equivalent expression of the ergodic security rate is derived. Then, on the premise of meeting the constraints of total transmission power and interference power, an alternating optimization algorithm combining Taylor series expansion method and Lagrange multiplier method is proposed to optimize jointly the transmission covariance matrix and phase shift matrix. Finally, simulation results verify the effectiveness of the proposed algorithm.
2023, 45(5): 1714-1721.
doi: 10.11999/JEIT220426
Abstract:
To demodulate the Weighted FRactional Fourier Transform encrypted Chaotic Direct Sequence Spread Spectrum (WFRFT-CD3S) signal over multipath fading channels, a generalized channel differential demodulation algorithm is proposed. The transmitter of the WFRFT-CD3S system modulates differentially the message bits. The receiver regards the product of the differential code and the channel as a generalized channel, and constructs a frequency-domain matched filter through the local spreading sequence to estimate the generalized channel impulse response. The receiver combines the path energies and recovers the message bits by de-differentiating the estimates of the generalized channel impulse response. The bit error rate of the proposed algorithm is analyzed theoretically, and the theoretical results are verified by numerical simulation. The numerical simulation results show that the proposed demodulation algorithm can demodulate the multipath WFRFT-CD3S signal under low signal-to-noise ratio, which ensures the anti-energy detection capability of the WFRFT-CD3S system.
To demodulate the Weighted FRactional Fourier Transform encrypted Chaotic Direct Sequence Spread Spectrum (WFRFT-CD3S) signal over multipath fading channels, a generalized channel differential demodulation algorithm is proposed. The transmitter of the WFRFT-CD3S system modulates differentially the message bits. The receiver regards the product of the differential code and the channel as a generalized channel, and constructs a frequency-domain matched filter through the local spreading sequence to estimate the generalized channel impulse response. The receiver combines the path energies and recovers the message bits by de-differentiating the estimates of the generalized channel impulse response. The bit error rate of the proposed algorithm is analyzed theoretically, and the theoretical results are verified by numerical simulation. The numerical simulation results show that the proposed demodulation algorithm can demodulate the multipath WFRFT-CD3S signal under low signal-to-noise ratio, which ensures the anti-energy detection capability of the WFRFT-CD3S system.
2023, 45(5): 1722-1730.
doi: 10.11999/JEIT220712
Abstract:
Secure Multiparty Computation (SMC) of sets has wide applications in joint data analysis, secure search over sensitive data, data security exchange. Based on geometric coding of rational numbers and the scalar product protocol, two secure computation protocols for computing the intersection and the union of two multisets with private rational numbers are proposed for the first time. The simulation paradigm is used to prove the privacy- preserving properties of proposed protocols in the semi-honest model, and the protocols’ efficiency is verified by theoretical analysis and programming test. Compared with existing protocols, the proposed protocols do not need to specify a universal set, which can protect the privacy of set potential. Moreover, the multiplication operation is mainly used in the implementation of the protocols, which achieves the security of information theory.
Secure Multiparty Computation (SMC) of sets has wide applications in joint data analysis, secure search over sensitive data, data security exchange. Based on geometric coding of rational numbers and the scalar product protocol, two secure computation protocols for computing the intersection and the union of two multisets with private rational numbers are proposed for the first time. The simulation paradigm is used to prove the privacy- preserving properties of proposed protocols in the semi-honest model, and the protocols’ efficiency is verified by theoretical analysis and programming test. Compared with existing protocols, the proposed protocols do not need to specify a universal set, which can protect the privacy of set potential. Moreover, the multiplication operation is mainly used in the implementation of the protocols, which achieves the security of information theory.
2023, 45(5): 1731-1736.
doi: 10.11999/JEIT221145
Abstract:
Constructing quantum codes with good parameters is an important part of quantum error-correcting codes research. In this paper,\begin{document}$ {2^m} $\end{document} ![]()
![]()
\begin{document}$ R = {F_{{4^m}}} + v{F_{{4^m}}} $\end{document} ![]()
![]()
\begin{document}$ \phi $\end{document} ![]()
![]()
\begin{document}$ \phi (C) $\end{document} ![]()
![]()
\begin{document}$ {2^m} $\end{document} ![]()
![]()
Constructing quantum codes with good parameters is an important part of quantum error-correcting codes research. In this paper,
2023, 45(5): 1737-1746.
doi: 10.11999/JEIT220153
Abstract:
To cope with the moving objects in dynamic environments and make the robots truly understand the surroundings, a visual Simultaneous Localization And Mapping (SLAM) system is proposed to estimate simultaneously trajectory and object-level dense 3D semantic maps in dynamic environments. Object detection and optical flow results are leveraged to identify those actually moving objects. To improve semantic mapping accuracy, an unsupervised algorithm is employed to segment 3D point cloud into meaningful clusters with semantic cues. The semantic maps are further used to improve object detection model, by fine-tuning with hard examples coming from semantic maps in challenging conditions. Extensive qualitative and quantitative experiments which compare the proposed method to comparable state-of-the-art approaches show that the proposed method achieves improved accuracy and robustness in dynamic scenes.
To cope with the moving objects in dynamic environments and make the robots truly understand the surroundings, a visual Simultaneous Localization And Mapping (SLAM) system is proposed to estimate simultaneously trajectory and object-level dense 3D semantic maps in dynamic environments. Object detection and optical flow results are leveraged to identify those actually moving objects. To improve semantic mapping accuracy, an unsupervised algorithm is employed to segment 3D point cloud into meaningful clusters with semantic cues. The semantic maps are further used to improve object detection model, by fine-tuning with hard examples coming from semantic maps in challenging conditions. Extensive qualitative and quantitative experiments which compare the proposed method to comparable state-of-the-art approaches show that the proposed method achieves improved accuracy and robustness in dynamic scenes.
2023, 45(5): 1747-1757.
doi: 10.11999/JEIT220369
Abstract:
To address the problem of insufficient target feature mining in nighttime marine boat detection based on low light remote sensing images, a new sparsity index is designed to minimize the misclassification of boat lights samples and background noises samples, and a detection algorithm for boat lights based on sparse coding and dictionary learning is proposed in this paper. The proposed algorithm is applied to the northern sea area of the Gulf of Mexico, the sea area south of Tianjin Port, and the sea area east of Shanghai Port, and the detection precision is 96.36%, 95.12%, 86.26%, recall rate is 96.36%, 92.86%, 94.19%, and the harmonic mean is 96.36%, 93.98%, 90.05% respectively. Furthermore, the proposed algorithm is compared with three typical marine boat lights detection method for low light remote sensing images during night, demonstrating that the proposed algorithm has a superior performance and provides a new idea for marine boat detection during night.
To address the problem of insufficient target feature mining in nighttime marine boat detection based on low light remote sensing images, a new sparsity index is designed to minimize the misclassification of boat lights samples and background noises samples, and a detection algorithm for boat lights based on sparse coding and dictionary learning is proposed in this paper. The proposed algorithm is applied to the northern sea area of the Gulf of Mexico, the sea area south of Tianjin Port, and the sea area east of Shanghai Port, and the detection precision is 96.36%, 95.12%, 86.26%, recall rate is 96.36%, 92.86%, 94.19%, and the harmonic mean is 96.36%, 93.98%, 90.05% respectively. Furthermore, the proposed algorithm is compared with three typical marine boat lights detection method for low light remote sensing images during night, demonstrating that the proposed algorithm has a superior performance and provides a new idea for marine boat detection during night.
2023, 45(5): 1758-1765.
doi: 10.11999/JEIT220441
Abstract:
Considering the problems of the existing in the process of multi-party human-computer interaction system, such as lack of propriety and low autonomy, a dialogue psychological model based on Stackelberg game is proposed in this paper. The multi-party human-computer interaction model is used to simulate the psychological game process in interpersonal communication. Taking into account the communication characteristics between the leader and the follower in the multi-party interaction, the game model of single leader and multiple followers is adopted to formalize it. The robot is played the role of the follower and considered the benefit brought by subordinate relationship in the multi-party Stackelberg game to make the effective decision-making strategy. The experimental result shows that the robot played the role of follower is polite and replied at the right time in communication with the multi-party, which improves further the rationality and autonomy in response for the robot.
Considering the problems of the existing in the process of multi-party human-computer interaction system, such as lack of propriety and low autonomy, a dialogue psychological model based on Stackelberg game is proposed in this paper. The multi-party human-computer interaction model is used to simulate the psychological game process in interpersonal communication. Taking into account the communication characteristics between the leader and the follower in the multi-party interaction, the game model of single leader and multiple followers is adopted to formalize it. The robot is played the role of the follower and considered the benefit brought by subordinate relationship in the multi-party Stackelberg game to make the effective decision-making strategy. The experimental result shows that the robot played the role of follower is polite and replied at the right time in communication with the multi-party, which improves further the rationality and autonomy in response for the robot.
2023, 45(5): 1766-1773.
doi: 10.11999/JEIT220320
Abstract:
An efficient photodetector placement method using Linear Optimization Fuzzy C-Means (LOFCM) and Artificial Neural Networks (ANN) is proposed for the problem that the current photodetector placement method is computationally intensive, energy intensive, susceptible to human factors and difficult to predict accurately indoor daylight illuminance. In this method, working surface photodetector layout is obtained by the LOFCM algorithm that using Fuzzy C-Means (FCM) to filter data after using Linear Optimization (LO) to sparse weight matrix. Subsequently, a non-linear mathematical model is trained between the working surface photodetector layouts and the four sets of auxiliary photodetector layouts using the ANN respectively. The experimental results show that the proposed LOFCM-based algorithm can reduce the number of working surface photodetector by 37.5% compared to other methods while ensuring the accuracy of calculating the average illumination and uniformity of the working surface. In addition, the wall and window auxiliary photodetector layout obtain a high prediction performance.
An efficient photodetector placement method using Linear Optimization Fuzzy C-Means (LOFCM) and Artificial Neural Networks (ANN) is proposed for the problem that the current photodetector placement method is computationally intensive, energy intensive, susceptible to human factors and difficult to predict accurately indoor daylight illuminance. In this method, working surface photodetector layout is obtained by the LOFCM algorithm that using Fuzzy C-Means (FCM) to filter data after using Linear Optimization (LO) to sparse weight matrix. Subsequently, a non-linear mathematical model is trained between the working surface photodetector layouts and the four sets of auxiliary photodetector layouts using the ANN respectively. The experimental results show that the proposed LOFCM-based algorithm can reduce the number of working surface photodetector by 37.5% compared to other methods while ensuring the accuracy of calculating the average illumination and uniformity of the working surface. In addition, the wall and window auxiliary photodetector layout obtain a high prediction performance.
2023, 45(5): 1774-1785.
doi: 10.11999/JEIT220362
Abstract:
Considering the problems of breast tumor size and shape change, blurred boundary and severe class imbalance between foreground and background, a multi-scale residual dual-domain attention fusion network is proposed. In this network, multi-scale residual blocks composed of multi-scale convolution are used as the basic building modules. Multi-scale residual block improves the network's ability to recognize targets of different sizes and the model’s robustness by extracting multi-scale features and optimizing gradient propagation. Meanwhile, the dual-domain attention units are integrated into the network to improve the ability of edge recognition and boundary preservation. The hybrid loss function with adaptive weight is proposed, it can improve the optimization direction of the network, alleviate the influence of the extreme imbalance of positive and negative samples. The experimental results show that the average Dice value of the method proposed in this paper reaches 0.8063, which is 5.3% higher than that of U-shaped Network (UNet), and the number of parameters is reduced by 73.36%, which has better segmentation performance.
Considering the problems of breast tumor size and shape change, blurred boundary and severe class imbalance between foreground and background, a multi-scale residual dual-domain attention fusion network is proposed. In this network, multi-scale residual blocks composed of multi-scale convolution are used as the basic building modules. Multi-scale residual block improves the network's ability to recognize targets of different sizes and the model’s robustness by extracting multi-scale features and optimizing gradient propagation. Meanwhile, the dual-domain attention units are integrated into the network to improve the ability of edge recognition and boundary preservation. The hybrid loss function with adaptive weight is proposed, it can improve the optimization direction of the network, alleviate the influence of the extreme imbalance of positive and negative samples. The experimental results show that the average Dice value of the method proposed in this paper reaches 0.8063, which is 5.3% higher than that of U-shaped Network (UNet), and the number of parameters is reduced by 73.36%, which has better segmentation performance.
2023, 45(5): 1786-1794.
doi: 10.11999/JEIT220354
Abstract:
micro RiboNucleic Acid (miRNA) plays an important role in the process of gene expression and transcription, and is closely related to the production of diseases. Biological experimental methods for disease miRNA association prediction are costly and time-consuming. To extract contextual information in heterogeneous networks of diseases and miRNAs, a high-performance Variational Gated AutoEncoder Network (VGAE-N) and a gated multi-layer perceptron are designed based on gated mechanisms and convolutions, and then a deep variational gated neural model is constructed for inferring disease miRNA associations. Multisource information first is integrated between miRNAs and diseases, and then the comprehensive similarity matrix is obtained between miRNAs and diseases. Based on a comprehensive network and the miRNA disease adjacency matrix, topological information is further extracted for miRNAs and diseases, respectively. Based on the miRNA disease adjacency matrix, the nonnegative matrix decomposition is used to extract low dimensional denoising features of miRNA and diseases. The experimental results show that the proposed model can effectively conduct miRNA disease association prediction, and provide reliable technical support for biological experiments.
micro RiboNucleic Acid (miRNA) plays an important role in the process of gene expression and transcription, and is closely related to the production of diseases. Biological experimental methods for disease miRNA association prediction are costly and time-consuming. To extract contextual information in heterogeneous networks of diseases and miRNAs, a high-performance Variational Gated AutoEncoder Network (VGAE-N) and a gated multi-layer perceptron are designed based on gated mechanisms and convolutions, and then a deep variational gated neural model is constructed for inferring disease miRNA associations. Multisource information first is integrated between miRNAs and diseases, and then the comprehensive similarity matrix is obtained between miRNAs and diseases. Based on a comprehensive network and the miRNA disease adjacency matrix, topological information is further extracted for miRNAs and diseases, respectively. Based on the miRNA disease adjacency matrix, the nonnegative matrix decomposition is used to extract low dimensional denoising features of miRNA and diseases. The experimental results show that the proposed model can effectively conduct miRNA disease association prediction, and provide reliable technical support for biological experiments.
2023, 45(5): 1795-1806.
doi: 10.11999/JEIT220470
Abstract:
Considering the characteristics of irregular retinal blood vessel topology, complex morphology and diverse scale changes, a Multi-resolution Fusion Input U-Netword (MFIU-Net) is proposed to achieve accurate segmentation of retinal blood vessels. A rough segmentation network based on multi-resolution fusion input is designed to generate high-resolution features.. The improved ResNeSt is used to replace the traditional convolution to optimize the boundary features of blood vessel segmentation, and the parallel spatial activation module is embedded to capture more semantic and spatial information. Another U-shaped fine segmentation network is constructed to improve the microscopic representation and recognition ability of the model. Firstly, the multi-scale dense feature pyramid module to extract the multi-scale feature information of blood vessels is adopted at the bottom layer. Secondly, the feature adaptive module is used to enhance the feature fusion between coarse and fine networks to suppress irrelevant background noise. Thirdly, a detail-oriented double loss function fusion is designed to guide the network to focus on learning features. Experiments are carried out on the fundus data Digital Retinal Images for Vessel Extraction (DRIVE), STructured Analysis of the REtinal (STARE) and Child Heart and Health Study (CHASE_DB1), the accuracy rates are 97.00%, 97.47% and 97.48%, the sensitivity is 82.73%, 82.86% and 83.24%, and the Area Under Cure (AUC) values are 98.74%, 98.90% and 98.93%, respectively. The overall performance of its model is better than that of existing algorithms.
Considering the characteristics of irregular retinal blood vessel topology, complex morphology and diverse scale changes, a Multi-resolution Fusion Input U-Netword (MFIU-Net) is proposed to achieve accurate segmentation of retinal blood vessels. A rough segmentation network based on multi-resolution fusion input is designed to generate high-resolution features.. The improved ResNeSt is used to replace the traditional convolution to optimize the boundary features of blood vessel segmentation, and the parallel spatial activation module is embedded to capture more semantic and spatial information. Another U-shaped fine segmentation network is constructed to improve the microscopic representation and recognition ability of the model. Firstly, the multi-scale dense feature pyramid module to extract the multi-scale feature information of blood vessels is adopted at the bottom layer. Secondly, the feature adaptive module is used to enhance the feature fusion between coarse and fine networks to suppress irrelevant background noise. Thirdly, a detail-oriented double loss function fusion is designed to guide the network to focus on learning features. Experiments are carried out on the fundus data Digital Retinal Images for Vessel Extraction (DRIVE), STructured Analysis of the REtinal (STARE) and Child Heart and Health Study (CHASE_DB1), the accuracy rates are 97.00%, 97.47% and 97.48%, the sensitivity is 82.73%, 82.86% and 83.24%, and the Area Under Cure (AUC) values are 98.74%, 98.90% and 98.93%, respectively. The overall performance of its model is better than that of existing algorithms.
C2 Transformer U-Net: A Medical Image Segmentation Model for Cross-modality and Contextual Semantics
2023, 45(5): 1807-1816.
doi: 10.11999/JEIT220445
Abstract:
Cross-modal medical images can provide more semantic information at the same lesion. In view of the U-Net network uses mainly single-modal images for segmentation, the cross-modal and contextual semantic correlations are not fully considered. Therefore, a cross-modal and contextual semantic-oriented medical image segmentation C2 Transformer U-Net model is proposed. The main idea of this model is: first, a backbone and auxiliary U-Net network structure is proposed in the encoder part to extract semantic information of different modalities; Then, the Multi-modal Context semantic Awareness Processor (MCAP) is designed to extract effectively the semantic information of the same lesion across modalities. After adding the two modal images using the backbone network in the skip connection, it is passed to the Transformer decoder. This enhances the expression ability of the model to the lesion; Secondly, the pre-activated residual unit and Transformer architecture are used in the encoder-decoder. On the one hand, the contextual feature information of the lesion is extracted, and on the other hand, the network pays more attention to the location information of the lesion when making full use of low-level and high-level features; Finally, the effectiveness of the algorithm is verified by using a clinical multi-modal lung medical image dataset. Comparative experimental results show that the Acc, Pre, Recall, Dice, Voe and Rvd of the proposed model for lung lesion segmentation are: 97.95%, 94.94%, 94.31%, 96.98%, 92.57% and 93.35%. For the segmentation of lung lesions with complex shapes, it has high accuracy and relatively low redundancy. Overall, it outperforms existing state-of-the-art methods.
Cross-modal medical images can provide more semantic information at the same lesion. In view of the U-Net network uses mainly single-modal images for segmentation, the cross-modal and contextual semantic correlations are not fully considered. Therefore, a cross-modal and contextual semantic-oriented medical image segmentation C2 Transformer U-Net model is proposed. The main idea of this model is: first, a backbone and auxiliary U-Net network structure is proposed in the encoder part to extract semantic information of different modalities; Then, the Multi-modal Context semantic Awareness Processor (MCAP) is designed to extract effectively the semantic information of the same lesion across modalities. After adding the two modal images using the backbone network in the skip connection, it is passed to the Transformer decoder. This enhances the expression ability of the model to the lesion; Secondly, the pre-activated residual unit and Transformer architecture are used in the encoder-decoder. On the one hand, the contextual feature information of the lesion is extracted, and on the other hand, the network pays more attention to the location information of the lesion when making full use of low-level and high-level features; Finally, the effectiveness of the algorithm is verified by using a clinical multi-modal lung medical image dataset. Comparative experimental results show that the Acc, Pre, Recall, Dice, Voe and Rvd of the proposed model for lung lesion segmentation are: 97.95%, 94.94%, 94.31%, 96.98%, 92.57% and 93.35%. For the segmentation of lung lesions with complex shapes, it has high accuracy and relatively low redundancy. Overall, it outperforms existing state-of-the-art methods.
2023, 45(5): 1817-1823.
doi: 10.11999/JEIT220578
Abstract:
To realize the generation of the navigation knowledge and the running control driven by goal for the intelligent and autonomous vehicle, a biological inspired Goal-Oriented (GO) navigation model based on spatial exploration and construction of cognitive map is discussed in this paper. This model is made up of three parts, including spatial exploration, construction of cognitive map and control of goal-oriented navigation. During spatial exploration, the model from Grid Cells (GCs) to Place Cells (PCs) and visual place cells’ model are fused to represent current state, and Q-learning algorithm is used to build and update the state-action. As a result, the goal-oriented navigation knowledge is learned. Then, during the construction of cognitive map, the gravity center estimation principle is used to deal with the obtained spatial exploration knowledge, which can produce the direction information corresponding to the different place cells’ state. Finally, during goal-oriented navigation process, the vehicle controls its running direction based on the cognitive map. Therefore, the goal-oriented navigation can be realized. Simulation validates that this model is available. The vehicle can construct cognitive map after sufficient spatial exploration and realizes goal-oriented navigation based on the cognitive map. Besides, the vehicle can effectively avoid obstacles during running.
To realize the generation of the navigation knowledge and the running control driven by goal for the intelligent and autonomous vehicle, a biological inspired Goal-Oriented (GO) navigation model based on spatial exploration and construction of cognitive map is discussed in this paper. This model is made up of three parts, including spatial exploration, construction of cognitive map and control of goal-oriented navigation. During spatial exploration, the model from Grid Cells (GCs) to Place Cells (PCs) and visual place cells’ model are fused to represent current state, and Q-learning algorithm is used to build and update the state-action. As a result, the goal-oriented navigation knowledge is learned. Then, during the construction of cognitive map, the gravity center estimation principle is used to deal with the obtained spatial exploration knowledge, which can produce the direction information corresponding to the different place cells’ state. Finally, during goal-oriented navigation process, the vehicle controls its running direction based on the cognitive map. Therefore, the goal-oriented navigation can be realized. Simulation validates that this model is available. The vehicle can construct cognitive map after sufficient spatial exploration and realizes goal-oriented navigation based on the cognitive map. Besides, the vehicle can effectively avoid obstacles during running.
2023, 45(5): 1824-1832.
doi: 10.11999/JEIT220401
Abstract:
For the training time of deep learning network is long, as well as larger distribution difference between source domain and target domain data of rolling bearings under different loads, a fast classification method of rolling bearing state based on improved broad model transfer learning is proposed. Fast Fourier transform is used to process the vibration signal of rolling bearing under different loads to construct frequency domain amplitude sequence data sets, from which a certain or some load data set is selected as the source domain, and other load data set is selected as the target domain. Secondly, an improved Broad Learning System (BLS) network is constructed by improving the way of building enhanced nodes windows of BLS in a cyclic extended way and introducing the Maxout activation function into the enhancement layer. At the same time, genetic algorithm is introduced to optimize the node structure of the improved BLS network. Then a pre-trained model based on source domain data is built. Finally, the network parameters, the weight parameters in feature layer and enhancement layer of the pre-trained model are transfered to target domain network, and a small amount target domain samples are used to fine-tune the network to build the state classification model. The experimental results show that the average training time of the proposed method is 32.6 s, and the average test accuracy is 98.9%. Compared with other methods, it could build a classification model in a shorter time and obtain good classification accuracy.
For the training time of deep learning network is long, as well as larger distribution difference between source domain and target domain data of rolling bearings under different loads, a fast classification method of rolling bearing state based on improved broad model transfer learning is proposed. Fast Fourier transform is used to process the vibration signal of rolling bearing under different loads to construct frequency domain amplitude sequence data sets, from which a certain or some load data set is selected as the source domain, and other load data set is selected as the target domain. Secondly, an improved Broad Learning System (BLS) network is constructed by improving the way of building enhanced nodes windows of BLS in a cyclic extended way and introducing the Maxout activation function into the enhancement layer. At the same time, genetic algorithm is introduced to optimize the node structure of the improved BLS network. Then a pre-trained model based on source domain data is built. Finally, the network parameters, the weight parameters in feature layer and enhancement layer of the pre-trained model are transfered to target domain network, and a small amount target domain samples are used to fine-tune the network to build the state classification model. The experimental results show that the average training time of the proposed method is 32.6 s, and the average test accuracy is 98.9%. Compared with other methods, it could build a classification model in a shorter time and obtain good classification accuracy.
2023, 45(5): 1833-1841.
doi: 10.11999/JEIT220444
Abstract:
Due to its relatively higher accuracy, the Anchor base algorithm has become a research hotspot for pedestrian detection in crowded scenes. However, the algorithm needs to design manually anchor boxes, which limits its generality. At the same time, a single Non-Maximum Suppression (NMS) screening threshold acting on crowd areas with different densities will lead to a certain degree of missed detection or false detection. To this end, a dual-head detection algorithm combining Anchor free and Anchor base detectors is proposed. Specifically, the Anchor free detector is used to perform rough detection on the image, and the coarse detection results are automatically clustered to generate anchor frames and then fed back to the Region Proposal Network (RPN) module, instead of manually designing the anchor frames in the RPN stage. Meanwhile, the density information of the population in different regions can be obtained through the statistics of the rough detection result information. A pedestrian head-whole body mutual supervision detection framework is designed, and the head detection results and the whole body detection results supervise each other, so as to reduce effectively the suppressed and missed target instances. A novel NMS method is proposed, which can adaptively select appropriate screening thresholds for crowd regions of different densities, thereby minimizing false detections caused by NMS process. The proposed detector is experimentally validated on the CrowdHuman dataset and the CityPersons dataset, achieving comparable performance to current state-of-the-art pedestrian detection methods.
Due to its relatively higher accuracy, the Anchor base algorithm has become a research hotspot for pedestrian detection in crowded scenes. However, the algorithm needs to design manually anchor boxes, which limits its generality. At the same time, a single Non-Maximum Suppression (NMS) screening threshold acting on crowd areas with different densities will lead to a certain degree of missed detection or false detection. To this end, a dual-head detection algorithm combining Anchor free and Anchor base detectors is proposed. Specifically, the Anchor free detector is used to perform rough detection on the image, and the coarse detection results are automatically clustered to generate anchor frames and then fed back to the Region Proposal Network (RPN) module, instead of manually designing the anchor frames in the RPN stage. Meanwhile, the density information of the population in different regions can be obtained through the statistics of the rough detection result information. A pedestrian head-whole body mutual supervision detection framework is designed, and the head detection results and the whole body detection results supervise each other, so as to reduce effectively the suppressed and missed target instances. A novel NMS method is proposed, which can adaptively select appropriate screening thresholds for crowd regions of different densities, thereby minimizing false detections caused by NMS process. The proposed detector is experimentally validated on the CrowdHuman dataset and the CityPersons dataset, achieving comparable performance to current state-of-the-art pedestrian detection methods.
2023, 45(5): 1842-1851.
doi: 10.11999/JEIT220358
Abstract:
In the face verification task of the face recognition model, traditional adversarial attack methods can not quickly generate real and natural adversarial examples, and the adversarial examples generated for one model under the white-box setting perform worse when transferred to other models. A GAN-based method TAdvFace is proposed for transferable adversarial example generation. TAdvFace uses an attention generator to improve the extraction of facial features. A Gaussian filtering operation is used to improve the smoothness of the adversarial samples. An automatic adjustment strategy is used to adjust the loss weight of identity discrimination, which can quickly generate high-quality migratable adversarial samples based on different face images. Experimental results show that through the white box training of a single model, the adversarial examples generated by the TAdvFace can achieve great attack results and transferability in a variety of face recognition models and commercial API models.
In the face verification task of the face recognition model, traditional adversarial attack methods can not quickly generate real and natural adversarial examples, and the adversarial examples generated for one model under the white-box setting perform worse when transferred to other models. A GAN-based method TAdvFace is proposed for transferable adversarial example generation. TAdvFace uses an attention generator to improve the extraction of facial features. A Gaussian filtering operation is used to improve the smoothness of the adversarial samples. An automatic adjustment strategy is used to adjust the loss weight of identity discrimination, which can quickly generate high-quality migratable adversarial samples based on different face images. Experimental results show that through the white box training of a single model, the adversarial examples generated by the TAdvFace can achieve great attack results and transferability in a variety of face recognition models and commercial API models.
2023, 45(5): 1852-1858.
doi: 10.11999/JEIT220360
Abstract:
A multiple valued current mode comparator is introduced to control the threshold of current mode CMOS circuits. Compared with binary logic circuits, a single wire of multiple valued logic circuits allows more information transmission. Compared with voltage signal, current signal is easy to realize arithmetic operations, such as addition and subtraction, which is more convenient in the design of multiple valued logic. At the same time, the design method of quaternary valued basic unit based on comparator is proposed, and the designs of quaternary valued max, min and inverter are realized. On this basis, full adder and subtractor are designed and realized. The design method is also applicable to binary, ternary and n-valued logic. The experimental results show that the designed circuit has correct logic function, lower power consumption and fewer transistors than the current mode CMOS full adder in the relevant literature.
A multiple valued current mode comparator is introduced to control the threshold of current mode CMOS circuits. Compared with binary logic circuits, a single wire of multiple valued logic circuits allows more information transmission. Compared with voltage signal, current signal is easy to realize arithmetic operations, such as addition and subtraction, which is more convenient in the design of multiple valued logic. At the same time, the design method of quaternary valued basic unit based on comparator is proposed, and the designs of quaternary valued max, min and inverter are realized. On this basis, full adder and subtractor are designed and realized. The design method is also applicable to binary, ternary and n-valued logic. The experimental results show that the designed circuit has correct logic function, lower power consumption and fewer transistors than the current mode CMOS full adder in the relevant literature.
2023, 45(5): 1859-1872.
doi: 10.11999/JEIT220664
Abstract:
With the rapid development of new technologies, such as the Internet and big data, more and more distributed data need to be processed by multiple parties. Therefore, privacy protection technology is facing greater challenges. Secure multi-party computation is an important privacy protection technology, which can provide solutions for the secure and efficient sharing of data. As an important branch of secure multi-party computation, Private Set Intersection (PSI) technology can calculate the intersection of private data sets of two or more participants under the premise of protecting the data privacy of participants. It can be divided into two-party PSI and multi-party PSI according to the number of participants. With the expansion of private data sharing scale, application scenarios with more than two participants are more and more common. Multi party PSI has the same technical basis as the two party PSI, but has essential differences. Firstly, the research progress of the two-party PSI is discussed. Then the development processes of multi-party PSI are analyzed in detail. The multi-party PSI is divided into traditional multi-party PSI and threshold multi-party PSI according to the different scenarios. At the same time, protocols in different scenarios are divided more carefully according to the different basic cryptographic protocols they used and their different functions. The typical protocols are analyzed, and the cryptographic protocols, security model, computation and communication complexity of the protocols are discussed. Finally, the research hotspots and future development directions of multi-party PSI are pointed out.
With the rapid development of new technologies, such as the Internet and big data, more and more distributed data need to be processed by multiple parties. Therefore, privacy protection technology is facing greater challenges. Secure multi-party computation is an important privacy protection technology, which can provide solutions for the secure and efficient sharing of data. As an important branch of secure multi-party computation, Private Set Intersection (PSI) technology can calculate the intersection of private data sets of two or more participants under the premise of protecting the data privacy of participants. It can be divided into two-party PSI and multi-party PSI according to the number of participants. With the expansion of private data sharing scale, application scenarios with more than two participants are more and more common. Multi party PSI has the same technical basis as the two party PSI, but has essential differences. Firstly, the research progress of the two-party PSI is discussed. Then the development processes of multi-party PSI are analyzed in detail. The multi-party PSI is divided into traditional multi-party PSI and threshold multi-party PSI according to the different scenarios. At the same time, protocols in different scenarios are divided more carefully according to the different basic cryptographic protocols they used and their different functions. The typical protocols are analyzed, and the cryptographic protocols, security model, computation and communication complexity of the protocols are discussed. Finally, the research hotspots and future development directions of multi-party PSI are pointed out.
2023, 45(5): 1873-1887.
doi: 10.11999/JEIT220242
Abstract:
Compared with the Fifth Generation mobile network (5G), the Sixth Generation mobile network (6G) is expected to introduce new performance indicators and application scenarios. Global coverage, high spectrum/energy/cost efficiency, high level of intelligence and security are leading features in 6G era. Different from traditional Base Station (BS)-centric network, the User-Centric Network (UCN) emerges as a key enabler for 6G by combining emerging technologies from information industries. In this novel framework, a comprehensive overview of physical layer, network layer and link layer is provided. As a starting point, the concepts and general architecture of the UCNs are surveied and discussed. Then, the survey is classified as: channel estimation and prediction; performance analysis with diverse performance metrics; different types of RRM (Radio Resource Management). Finally, based on extensive discussions, open issues are provided to guide future scholarly research directions. It is anticipated that this survey will provide a quick and comprehensive understanding of the current state of the arts for UCNs which attracting more researchers into this area.
Compared with the Fifth Generation mobile network (5G), the Sixth Generation mobile network (6G) is expected to introduce new performance indicators and application scenarios. Global coverage, high spectrum/energy/cost efficiency, high level of intelligence and security are leading features in 6G era. Different from traditional Base Station (BS)-centric network, the User-Centric Network (UCN) emerges as a key enabler for 6G by combining emerging technologies from information industries. In this novel framework, a comprehensive overview of physical layer, network layer and link layer is provided. As a starting point, the concepts and general architecture of the UCNs are surveied and discussed. Then, the survey is classified as: channel estimation and prediction; performance analysis with diverse performance metrics; different types of RRM (Radio Resource Management). Finally, based on extensive discussions, open issues are provided to guide future scholarly research directions. It is anticipated that this survey will provide a quick and comprehensive understanding of the current state of the arts for UCNs which attracting more researchers into this area.
2023, 45(5): 1888-1898.
doi: 10.11999/JEIT220420
Abstract:
Memory wall has become one of the key challenges in Von Neumann architecture, memory-centric computing architectures, such as In-Memory Computing (IMC) and Near-Memory Computing (NMC) are expected to break the Von-Neumann bottleneck, improving computing performance and energy efficiency. The progress of memory-centric computing technology, as well as the principles, advantages and problems based on a variety of memory media, such as traditional memories (e.g., DRAM, SRAM and Flash) and emerging non-volatile memories (e.g., ReRAM, PCM, MRAM and FeFET) are introduced in this paper. Then, the circuit structure and main applications with IMC chips are highlighted, taking Witmem's product WTM2101 as an example. Finally, the future development prospects and challenges of the all-in-one chip are also analysed.
Memory wall has become one of the key challenges in Von Neumann architecture, memory-centric computing architectures, such as In-Memory Computing (IMC) and Near-Memory Computing (NMC) are expected to break the Von-Neumann bottleneck, improving computing performance and energy efficiency. The progress of memory-centric computing technology, as well as the principles, advantages and problems based on a variety of memory media, such as traditional memories (e.g., DRAM, SRAM and Flash) and emerging non-volatile memories (e.g., ReRAM, PCM, MRAM and FeFET) are introduced in this paper. Then, the circuit structure and main applications with IMC chips are highlighted, taking Witmem's product WTM2101 as an example. Finally, the future development prospects and challenges of the all-in-one chip are also analysed.
2023, 45(5): 1899-1910.
doi: 10.11999/JEIT220418
Abstract:
Software-Defined Networking (SDN) is the key technique of the next-generation network. Recently, SDN has become a hot spot in both academia and industry. Wide Area Network (WAN) is one of the primary application scenarios in the industry for SDN, which is known as Software-Defined WAN (SD-WAN). In SD-WAN, flexible traffic scheduling and network performance improvement are realized by the flow path programmability, which is enabled by the SDN controller to change dynamically the paths of flows traversing SDN switches. However, controller failure is a common phenomenon. When the controller fails, the switches controlled by the failed controller become offline, and the flows traversing the offline switches become offline too. In this way, the path programmability can not be guaranteed, and thus flexible flow control becomes invalid, leading to severe network performance degradation. This survey is presented to introduce the research works on maintaining path programmability in SD-WAN. First, the path programmability and the important feature for maintaining the path programmability in SD-WAN are introduced. Second, different types of existing solutions for coping with the controller failure in SD-WAN are proposed. Finally, potential improvements and future directions on this research topic are proposed．
Software-Defined Networking (SDN) is the key technique of the next-generation network. Recently, SDN has become a hot spot in both academia and industry. Wide Area Network (WAN) is one of the primary application scenarios in the industry for SDN, which is known as Software-Defined WAN (SD-WAN). In SD-WAN, flexible traffic scheduling and network performance improvement are realized by the flow path programmability, which is enabled by the SDN controller to change dynamically the paths of flows traversing SDN switches. However, controller failure is a common phenomenon. When the controller fails, the switches controlled by the failed controller become offline, and the flows traversing the offline switches become offline too. In this way, the path programmability can not be guaranteed, and thus flexible flow control becomes invalid, leading to severe network performance degradation. This survey is presented to introduce the research works on maintaining path programmability in SD-WAN. First, the path programmability and the important feature for maintaining the path programmability in SD-WAN are introduced. Second, different types of existing solutions for coping with the controller failure in SD-WAN are proposed. Finally, potential improvements and future directions on this research topic are proposed．
