Email alert
2024 Vol. 46, No. 4
Display Method:
2024, 46(4): 1-4.
Abstract:
2024, 46(4): 1147-1156.
doi: 10.11999/JEIT230570
Abstract:
A large-scale dataset of drone radio frequency signals, namely DroneRFa, is constructed to research and develop anti-drone detection and recognition technologies. This dataset uses a software-defined radio device to monitor communication signals between drones and their controllers, including 9 types of flying drone signals in an outdoor environment, 15 types of drone signals in an indoor environment, and 1 type of background signal as a reference. Each type of data has no less than 12 segments, each containing more than 100 million sampling points. The data acquisition covered three Industrial Scientific Medical (ISM) radio bands, and recorded the multifrequency communication activity of drones. The dataset has detailed flying distance and communication frequency band labeling, which are represented with prefix characters and binary codes to facilitate easy access to specific data required by users. Furthermore, this paper proposes two drone identification schemes based on spectral and visual statistical features and deep learning representation to verify the reliability and validity of the dataset.
A large-scale dataset of drone radio frequency signals, namely DroneRFa, is constructed to research and develop anti-drone detection and recognition technologies. This dataset uses a software-defined radio device to monitor communication signals between drones and their controllers, including 9 types of flying drone signals in an outdoor environment, 15 types of drone signals in an indoor environment, and 1 type of background signal as a reference. Each type of data has no less than 12 segments, each containing more than 100 million sampling points. The data acquisition covered three Industrial Scientific Medical (ISM) radio bands, and recorded the multifrequency communication activity of drones. The dataset has detailed flying distance and communication frequency band labeling, which are represented with prefix characters and binary codes to facilitate easy access to specific data required by users. Furthermore, this paper proposes two drone identification schemes based on spectral and visual statistical features and deep learning representation to verify the reliability and validity of the dataset.
2024, 46(4): 1157-1172.
doi: 10.11999/JEIT230544
Abstract:
UAV (Unmanned Aerial Vehicle) swarm, with its application advantages and development prospects, has become one of the current hot spots of interest for researchers in the field of artificial intelligence. The UAV swarm confrontation technology under incomplete information has become one of the research directions with the highest requirements for swarm cooperativeness and intelligence due to the high dynamics of swarm structure changes and the complex and variable environmental information that cannot be fully perceived. Its research achievements can promote the rapid development and wide application of intelligent unmanned systems. This paper comprehensively reviews of the recent progress in the research of UAV swarm confrontation under incomplete information environments. According to the Observe-Orient-Decide-Act (OODA) loop theory, the UAV swarm confrontation process is divided into four interlocking key components of situation assessment, intention inference, mission planning, and maneuver decision, and is further subdivided into eight sub-research objectives. By analyzing and comparing the relevant research works in recent years, the research focuses and difficulties of various tasks in the field of UAV swarm confrontation and the achieved research results are highlighted, and the challenges faced by UAV swarm confrontation technology are discussed, including the cooperative control of large-scale heterogeneous swarms, the handling of incomplete information, the modeling of complex decision-making processes, and the tackling of practical application tasks.
UAV (Unmanned Aerial Vehicle) swarm, with its application advantages and development prospects, has become one of the current hot spots of interest for researchers in the field of artificial intelligence. The UAV swarm confrontation technology under incomplete information has become one of the research directions with the highest requirements for swarm cooperativeness and intelligence due to the high dynamics of swarm structure changes and the complex and variable environmental information that cannot be fully perceived. Its research achievements can promote the rapid development and wide application of intelligent unmanned systems. This paper comprehensively reviews of the recent progress in the research of UAV swarm confrontation under incomplete information environments. According to the Observe-Orient-Decide-Act (OODA) loop theory, the UAV swarm confrontation process is divided into four interlocking key components of situation assessment, intention inference, mission planning, and maneuver decision, and is further subdivided into eight sub-research objectives. By analyzing and comparing the relevant research works in recent years, the research focuses and difficulties of various tasks in the field of UAV swarm confrontation and the achieved research results are highlighted, and the challenges faced by UAV swarm confrontation technology are discussed, including the cooperative control of large-scale heterogeneous swarms, the handling of incomplete information, the modeling of complex decision-making processes, and the tackling of practical application tasks.
2024, 46(4): 1173-1185.
doi: 10.11999/JEIT230847
Abstract:
Orbital Angular Momentum (OAM) technology provides a new degree of freedom to tackle the problems of the increasingly strained frequency resources, due to its theoretical orthogonal modes and non-interference with each other, which shows a good advantage in expanding the channel capacity. To face complex and diverse wireless communication scenarios, the design of reconfigurable OAM antennas is the physical layer basis for multi-mode multiplexing, intelligent information sensing, and artificial intelligence antennas. In this paper, the design method and characteristics of OAM reconfigurable antenna are studied according to the realization mechanism of reconfigurable antennas. Then, the research progress of reconfigurable OAM antennas is systematically reviewed. Finally, the future design of reconfigurable OAM antenna is discussed.
Orbital Angular Momentum (OAM) technology provides a new degree of freedom to tackle the problems of the increasingly strained frequency resources, due to its theoretical orthogonal modes and non-interference with each other, which shows a good advantage in expanding the channel capacity. To face complex and diverse wireless communication scenarios, the design of reconfigurable OAM antennas is the physical layer basis for multi-mode multiplexing, intelligent information sensing, and artificial intelligence antennas. In this paper, the design method and characteristics of OAM reconfigurable antenna are studied according to the realization mechanism of reconfigurable antennas. Then, the research progress of reconfigurable OAM antennas is systematically reviewed. Finally, the future design of reconfigurable OAM antenna is discussed.
2024, 46(4): 1186-1195.
doi: 10.11999/JEIT230406
Abstract:
In order to integrate signals of the Visible Light Positioning (VLP) and Visible Light Communication (VLC), a Frequency Division Multiplexing (FDM) based signal structure and a robust power allocation scheme for the multi-LED integrated Visible Light Position and Communication (VLPC) system are proposed. Firstly, an FDM-based VLPC signal structure is designed to carry two signals simultaneously with independent spectrum resource allocation, which can reduce transmission delay and improve real-time positioning. Then, the channel estimation based on positioning results is investigated, and the coupling relationship and statistical feature between the channel estimation error and positioning error are revealed. Furthermore, a VLPC robust power allocation problem is proposed to minimize the Cramér-Rao Lower Bound(CRLB) of the VLP under the power constraints and outage chance constraint of the communication rate. This nonconvex problem is transformed into a series of iterative convex semidefinite programming subproblems through semidefinite relaxation, worst-case conditional value-at-risk, and successive convex approximation. Finally, from the simulation results, it is verified that the proposed scheme can simultaneously achieve robust communication and effective positioning. The robust achievable rate exceeds 350 Mbit/s, and the centimeter-level positioning can still be achieved when the minimum rate requirement is 200 Mbit/s, and the maximum tolerable outage probability is 0.01.
In order to integrate signals of the Visible Light Positioning (VLP) and Visible Light Communication (VLC), a Frequency Division Multiplexing (FDM) based signal structure and a robust power allocation scheme for the multi-LED integrated Visible Light Position and Communication (VLPC) system are proposed. Firstly, an FDM-based VLPC signal structure is designed to carry two signals simultaneously with independent spectrum resource allocation, which can reduce transmission delay and improve real-time positioning. Then, the channel estimation based on positioning results is investigated, and the coupling relationship and statistical feature between the channel estimation error and positioning error are revealed. Furthermore, a VLPC robust power allocation problem is proposed to minimize the Cramér-Rao Lower Bound(CRLB) of the VLP under the power constraints and outage chance constraint of the communication rate. This nonconvex problem is transformed into a series of iterative convex semidefinite programming subproblems through semidefinite relaxation, worst-case conditional value-at-risk, and successive convex approximation. Finally, from the simulation results, it is verified that the proposed scheme can simultaneously achieve robust communication and effective positioning. The robust achievable rate exceeds 350 Mbit/s, and the centimeter-level positioning can still be achieved when the minimum rate requirement is 200 Mbit/s, and the maximum tolerable outage probability is 0.01.
2024, 46(4): 1196-1202.
doi: 10.11999/JEIT230475
Abstract:
Due to the very strict requirements of the system for linear components, the current Orthogonal Frequency Division Multiplexing(OFDM) system has the problem of high Peak to Average Power Ratio (PAPR), which causes signal distortion of OFDM and OFDM-based radar integrated system, and affects the system performance. To solve this problem, a Tone Reservation(TR) method based on the weighted least squares method is proposed in this paper. Firstly, the subcarriers in the OFDM design scheme are divided into data subcarriers and reserved blank subcarriers, data subcarriers modulate data subcarriers, and blank subcarriers modulate blank data. Then, the original data is used to obtain the best peak clipping coefficient and peak clipping data through the weighted least squares method, and the peak clipping data is modulated on the blank subcarrier. Finally, the peak shaving data is superimposed on the original data to complete the suppression of PAPR. Simulation results show that the TR algorithm based on weighted least squares method can achieve good PAPR suppression effect in 1 to 3 iterations, and its convergence speed is significantly improved compared with the traditional algorithm.
Due to the very strict requirements of the system for linear components, the current Orthogonal Frequency Division Multiplexing(OFDM) system has the problem of high Peak to Average Power Ratio (PAPR), which causes signal distortion of OFDM and OFDM-based radar integrated system, and affects the system performance. To solve this problem, a Tone Reservation(TR) method based on the weighted least squares method is proposed in this paper. Firstly, the subcarriers in the OFDM design scheme are divided into data subcarriers and reserved blank subcarriers, data subcarriers modulate data subcarriers, and blank subcarriers modulate blank data. Then, the original data is used to obtain the best peak clipping coefficient and peak clipping data through the weighted least squares method, and the peak clipping data is modulated on the blank subcarrier. Finally, the peak shaving data is superimposed on the original data to complete the suppression of PAPR. Simulation results show that the TR algorithm based on weighted least squares method can achieve good PAPR suppression effect in 1 to 3 iterations, and its convergence speed is significantly improved compared with the traditional algorithm.
2024, 46(4): 1203-1211.
doi: 10.11999/JEIT230426
Abstract:
To resolve the channel uncertainties, wireless information leakage and low communication quality caused by obstacles, a robust secure resource allocation algorithm for a Reconfigurable Intelligent Surface (RIS) aided Multiple Input Single Output (MISO) Symbiotic Radio (SR) network is proposed. Considering the constraint of the secure rate of the primary user, the constraint of the minimum rate of the secondary user and the constraint of the minimum harvested energy of the RIS, a resource allocation problem based on the bounded channel uncertainties is formulated by jointly optimizing the active beamforming vector and the passive beamforming vector. Then, the parameter perturbation included non-convex problem is transformed into a deterministic convex optimization problem via the semidefinite relaxation, S-procedure and the variable substitution methods, and a semidefinite relaxation based robust resource allocation algorithm is proposed. Numerical results verify that the proposed algorithm has better convergence and robustness compared with existing algorithms.
To resolve the channel uncertainties, wireless information leakage and low communication quality caused by obstacles, a robust secure resource allocation algorithm for a Reconfigurable Intelligent Surface (RIS) aided Multiple Input Single Output (MISO) Symbiotic Radio (SR) network is proposed. Considering the constraint of the secure rate of the primary user, the constraint of the minimum rate of the secondary user and the constraint of the minimum harvested energy of the RIS, a resource allocation problem based on the bounded channel uncertainties is formulated by jointly optimizing the active beamforming vector and the passive beamforming vector. Then, the parameter perturbation included non-convex problem is transformed into a deterministic convex optimization problem via the semidefinite relaxation, S-procedure and the variable substitution methods, and a semidefinite relaxation based robust resource allocation algorithm is proposed. Numerical results verify that the proposed algorithm has better convergence and robustness compared with existing algorithms.
2024, 46(4): 1212-1221.
doi: 10.11999/JEIT230518
Abstract:
Considering the low classification accuracy of spreading and conventional modulated signals under low signal-to-noise ratio conditions, a multimodal attention mechanism signal modulation recognition method based on Generative Adversarial Network (GAN) and Convolutional Neural Networks (CNN) with Long Short-Term Memory (LSTM) network is proposed. Firstly, the Time-Frequency Images (TFIs) of the to-be-recognized signals are generated and the noise reduction process of TFIs is realized by using GAN; Secondly, the In-phase and Quadrature data (I/Q data) of the signals with TFIs are used as model inputs, and the CNN-based TFIs recognition branch and the LSTM-based I/Q data recognition branch are built; Finally, an attentional mechanism is added to the model to enhance the role of important features in I/Q data and TFIs in the determination of classification results. The experimental results show that the proposed method effectively improves the overall classification accuracy by 2% to 7% compared with the unimodal recognition model and other baseline models, and possesses stronger feature expression capability and robustness under low signal-to-noise ratio conditions.
Considering the low classification accuracy of spreading and conventional modulated signals under low signal-to-noise ratio conditions, a multimodal attention mechanism signal modulation recognition method based on Generative Adversarial Network (GAN) and Convolutional Neural Networks (CNN) with Long Short-Term Memory (LSTM) network is proposed. Firstly, the Time-Frequency Images (TFIs) of the to-be-recognized signals are generated and the noise reduction process of TFIs is realized by using GAN; Secondly, the In-phase and Quadrature data (I/Q data) of the signals with TFIs are used as model inputs, and the CNN-based TFIs recognition branch and the LSTM-based I/Q data recognition branch are built; Finally, an attentional mechanism is added to the model to enhance the role of important features in I/Q data and TFIs in the determination of classification results. The experimental results show that the proposed method effectively improves the overall classification accuracy by 2% to 7% compared with the unimodal recognition model and other baseline models, and possesses stronger feature expression capability and robustness under low signal-to-noise ratio conditions.
2024, 46(4): 1222-1230.
doi: 10.11999/JEIT230458
Abstract:
Due to the limited transmitting power of sensors in the Wireless Sensor Network (WSN) and high probability of large distance between sensors and their associated Base Station(BS), the sensor data may not be received in time. This will reduce the data freshness of sensor data and affect the quality of decision for delay sensitive service. Therefore, the use of Unmanned Aerial Vehicles (UAVs) to assist in collecting sensor data has become an effective solution to decrease the data freshness, measured by Age of Information (AoI), in wireless sensor networks. A UAV trajectory optimization algorithm based on the Multi-Agent Proximal Policy Optimization (MAPPO) method is developed in this paper, which employs a centralized-training and distributed-execution framework. By jointly optimizing the flight trajectories of all UAVs, the average AoI of all ground nodes is minimized. The simulation results verify the effectiveness of our proposed UAV trajectory optimization algorithm on minimizing the AoI in the WSN.
Due to the limited transmitting power of sensors in the Wireless Sensor Network (WSN) and high probability of large distance between sensors and their associated Base Station(BS), the sensor data may not be received in time. This will reduce the data freshness of sensor data and affect the quality of decision for delay sensitive service. Therefore, the use of Unmanned Aerial Vehicles (UAVs) to assist in collecting sensor data has become an effective solution to decrease the data freshness, measured by Age of Information (AoI), in wireless sensor networks. A UAV trajectory optimization algorithm based on the Multi-Agent Proximal Policy Optimization (MAPPO) method is developed in this paper, which employs a centralized-training and distributed-execution framework. By jointly optimizing the flight trajectories of all UAVs, the average AoI of all ground nodes is minimized. The simulation results verify the effectiveness of our proposed UAV trajectory optimization algorithm on minimizing the AoI in the WSN.
2024, 46(4): 1231-1239.
doi: 10.11999/JEIT230377
Abstract:
Wireless spectrum status sensing is one of the prerequisites for achieving efficient utilization of spectrum resources and harmonious coexistence among systems. A spectrum sensing scheme based on interpolation and Few-Shot Learning(FSL) classification is proposed to address the sparsity of spectrum data, unstable distribution of data categories, and severe shortage of labeled data in complex wireless propagation environments. Firstly, the sparsely distributed observation data is interpolated and a spectral status map is constructed as the input data to the spectral status classifier. Then, for the cases where the distributions of data categories are unstable and the amount of data is severely insufficient, a few-shot learning-based classification algorithm is proposed, incorporating the embedding modules and measurement modules to realize fast and accurate spectrum status classification. Specifically, the embedding module is used to map spectral data to the embedding space and extract hidden image features from the spectral data. In the measurement module, two category representation methods, prototype-based and sample-based, are proposed to determine the category of the samples by calculating the similarity between the samples and the categories. Finally, an A-way B-shot task training model is set to ensure that the classification model will not cause overfitting problems due to the small number of test samples. Simulation results show that compared with traditional machine learning methods, the proposed model can achieve accurate classification under low signal-to-noise ratio conditions. In addition, it can quickly distinguish the categories of radiation source activity scenarios even when the number of samples in the test set is small or when new classes that have never been seen in the training set appear in the test set.
Wireless spectrum status sensing is one of the prerequisites for achieving efficient utilization of spectrum resources and harmonious coexistence among systems. A spectrum sensing scheme based on interpolation and Few-Shot Learning(FSL) classification is proposed to address the sparsity of spectrum data, unstable distribution of data categories, and severe shortage of labeled data in complex wireless propagation environments. Firstly, the sparsely distributed observation data is interpolated and a spectral status map is constructed as the input data to the spectral status classifier. Then, for the cases where the distributions of data categories are unstable and the amount of data is severely insufficient, a few-shot learning-based classification algorithm is proposed, incorporating the embedding modules and measurement modules to realize fast and accurate spectrum status classification. Specifically, the embedding module is used to map spectral data to the embedding space and extract hidden image features from the spectral data. In the measurement module, two category representation methods, prototype-based and sample-based, are proposed to determine the category of the samples by calculating the similarity between the samples and the categories. Finally, an A-way B-shot task training model is set to ensure that the classification model will not cause overfitting problems due to the small number of test samples. Simulation results show that compared with traditional machine learning methods, the proposed model can achieve accurate classification under low signal-to-noise ratio conditions. In addition, it can quickly distinguish the categories of radiation source activity scenarios even when the number of samples in the test set is small or when new classes that have never been seen in the training set appear in the test set.
2024, 46(4): 1240-1246.
doi: 10.11999/JEIT230401
Abstract:
For the effects of spatial-wideband effects during millimeter wave positioning, a novel 3-Dimensional (3D) positioning estimation method based on Reconfigurable Intelligent Surface (RIS) and Single Input Single Output (SISO) millimeter wave system is proposed. First, by designing the RIS phase profiles, the channel parameters of the direct Line-of-Sight (LoS) delay, the RIS-aided Virtual Line-of-Sight (VLoS) delay, and the Angle-of-Departure (AoD) between the RIS and the user are coarsely estimated based on the Inverse Fast Fourier Transform (IFFT). Then, the quasi-Newton method is used to refine these parameters and to estimate the location of the user. Simulations are conducted to compare the positioning performance of the proposed spatial-wideband estimation method with that of the traditional narrowband estimation method. The results show that by taking into account the spatial-wideband effects, the positioning accuracy can be improved by approximately 10% for a bandwidth of 480 MHz and by more than 20% as the bandwidth increases beyond 800 MHz.
For the effects of spatial-wideband effects during millimeter wave positioning, a novel 3-Dimensional (3D) positioning estimation method based on Reconfigurable Intelligent Surface (RIS) and Single Input Single Output (SISO) millimeter wave system is proposed. First, by designing the RIS phase profiles, the channel parameters of the direct Line-of-Sight (LoS) delay, the RIS-aided Virtual Line-of-Sight (VLoS) delay, and the Angle-of-Departure (AoD) between the RIS and the user are coarsely estimated based on the Inverse Fast Fourier Transform (IFFT). Then, the quasi-Newton method is used to refine these parameters and to estimate the location of the user. Simulations are conducted to compare the positioning performance of the proposed spatial-wideband estimation method with that of the traditional narrowband estimation method. The results show that by taking into account the spatial-wideband effects, the positioning accuracy can be improved by approximately 10% for a bandwidth of 480 MHz and by more than 20% as the bandwidth increases beyond 800 MHz.
2024, 46(4): 1247-1256.
doi: 10.11999/JEIT230733
Abstract:
Media Access Control (MAC) plays a pivotal role in ensuring proper operation of Wireless Body Area Networks (WBAN). However, current solutions still cannot satisfy high performance requirements of low latency and energy consumption for emergency data reporting. A Multi-factor Emergency Scheduling Scheme (MESS) is proposed to meeting such a strict demand. First, a data classification method is designed to sort data as periodic data and emergency data, respectively. Unlike consistent data characteristics in other schemes, data heterogeneity is considered in our solution, which is more practical for different nodes. Second, a multi-factor priority division scheme is devised, according to the disease-related factor, critical degree factor, health severity factor and age of information factor. This is a more comprehensive consideration of the key characteristics of the node. In addition, a dynamic slot allocation and sequencing approach is designed, in which time slots of nodes are allocated based on the data classification and multi-factor priority-based ordering. This enhances low latency and guarantees energy efficiency of nodes. Theoretical and simulation results demonstrate the advantages of MESS in terms of delay and energy efficiency.
Media Access Control (MAC) plays a pivotal role in ensuring proper operation of Wireless Body Area Networks (WBAN). However, current solutions still cannot satisfy high performance requirements of low latency and energy consumption for emergency data reporting. A Multi-factor Emergency Scheduling Scheme (MESS) is proposed to meeting such a strict demand. First, a data classification method is designed to sort data as periodic data and emergency data, respectively. Unlike consistent data characteristics in other schemes, data heterogeneity is considered in our solution, which is more practical for different nodes. Second, a multi-factor priority division scheme is devised, according to the disease-related factor, critical degree factor, health severity factor and age of information factor. This is a more comprehensive consideration of the key characteristics of the node. In addition, a dynamic slot allocation and sequencing approach is designed, in which time slots of nodes are allocated based on the data classification and multi-factor priority-based ordering. This enhances low latency and guarantees energy efficiency of nodes. Theoretical and simulation results demonstrate the advantages of MESS in terms of delay and energy efficiency.
2024, 46(4): 1257-1265.
doi: 10.11999/JEIT230630
Abstract:
For a Reconfigurable Intelligent Surface (RIS)-assisted Multiple Input Multiple Output (MIMO) Non-Orthogonal Multiple Access (NOMA) downlink system, the transmit covariances matrix at the base station and the phase-shifting matrix at the RIS are jointly designed based on statistical Channel State Information (CSI). First, in the spatial correlated Rician channel, a deterministic large-system approximation for the ergodic sum rate is obtained for an RIS-assisted MIMO-NOMA system by resorting to the large-dimensional random matrix theory. Then, by maximizing the approximated sum rate, the transmit covariances matrix for the strong and weak user as well as the phase-shifting matrix at the RIS are designed based on statistical CSI under the constraints of total transmit power and the rates threshold for weak user. The simulations validate the high accuracy of our approximations and our proposed transmit covariances and phase-shifting matrix can improve the system performances significantly.
For a Reconfigurable Intelligent Surface (RIS)-assisted Multiple Input Multiple Output (MIMO) Non-Orthogonal Multiple Access (NOMA) downlink system, the transmit covariances matrix at the base station and the phase-shifting matrix at the RIS are jointly designed based on statistical Channel State Information (CSI). First, in the spatial correlated Rician channel, a deterministic large-system approximation for the ergodic sum rate is obtained for an RIS-assisted MIMO-NOMA system by resorting to the large-dimensional random matrix theory. Then, by maximizing the approximated sum rate, the transmit covariances matrix for the strong and weak user as well as the phase-shifting matrix at the RIS are designed based on statistical CSI under the constraints of total transmit power and the rates threshold for weak user. The simulations validate the high accuracy of our approximations and our proposed transmit covariances and phase-shifting matrix can improve the system performances significantly.
2024, 46(4): 1266-1275.
doi: 10.11999/JEIT230515
Abstract:
Benefited from the characteristics of high mobility, low cost and convenient deployment, by deploying communication and sensing equipment and supporting efficient resource sharing of communication and sensing technologies, Unmanned Aerial Vehicles (UAVs) are expected to act as high-performance aerial platforms which integrate communication and sensing technologies. In this paper, Multiple-Input-Multiple-Output (MIMO) UAV-enabled joint communication and sensing scenario is examined, the constraints of the flight energy of the UAV, multi-antenna transmission and user service requirement are jointly considered, the problem of UAV communication, sensing precoding and flight trajectory is formulated as a multi-objective optimization problem which jointly maximizes the minimum data rate of communication users and the minimum discovery probability of targets. Since the minimum rate maximization problem of communication users is a non-convex optimization problem, which is difficult to solve directly, the original optimization problem is decomposed into communication precoding design subproblem and UAV trajectory design subproblem. The two subproblems are solved successively by applying alternate iteration method until the algorithm reaches convergence. Specifically, a Zero-Forcing (ZF) algorithm is put forward for solving the communication precoding design subproblem. A Successive Convex Approximation (SCA) algorithm is applied to determine the optimal trajectory of the UAV. Based on the optimal trajectory of UAV, the sensing location selection problem is modeled as a weighted distance minimization problem, and then the extensive search algorithm is applied to obtain the optimal locations. Finally, a ZF algorithm-based joint communication and sensing precoding is presented. The effectiveness of the proposed algorithm is verified by simulations.
Benefited from the characteristics of high mobility, low cost and convenient deployment, by deploying communication and sensing equipment and supporting efficient resource sharing of communication and sensing technologies, Unmanned Aerial Vehicles (UAVs) are expected to act as high-performance aerial platforms which integrate communication and sensing technologies. In this paper, Multiple-Input-Multiple-Output (MIMO) UAV-enabled joint communication and sensing scenario is examined, the constraints of the flight energy of the UAV, multi-antenna transmission and user service requirement are jointly considered, the problem of UAV communication, sensing precoding and flight trajectory is formulated as a multi-objective optimization problem which jointly maximizes the minimum data rate of communication users and the minimum discovery probability of targets. Since the minimum rate maximization problem of communication users is a non-convex optimization problem, which is difficult to solve directly, the original optimization problem is decomposed into communication precoding design subproblem and UAV trajectory design subproblem. The two subproblems are solved successively by applying alternate iteration method until the algorithm reaches convergence. Specifically, a Zero-Forcing (ZF) algorithm is put forward for solving the communication precoding design subproblem. A Successive Convex Approximation (SCA) algorithm is applied to determine the optimal trajectory of the UAV. Based on the optimal trajectory of UAV, the sensing location selection problem is modeled as a weighted distance minimization problem, and then the extensive search algorithm is applied to obtain the optimal locations. Finally, a ZF algorithm-based joint communication and sensing precoding is presented. The effectiveness of the proposed algorithm is verified by simulations.
2024, 46(4): 1276-1285.
doi: 10.11999/JEIT230578
Abstract:
To investigate transmission characteristics of in-vivo Mmultiple Input Multiple Output (MIMO) communication systems at TeraHertz (THz) frequency, a precise skin-fat model is constructed at 0.8～1.2 THz. Full wave electromagnetic simulations are conducted on the vertical and horizontal links in the skin-fat model, the characteristics of terahertz in-vivo channel are analyzed, and a terahertz in-vivo path loss model is established. Firstly, the skin-fat model is constructed based on the dielectric properties of human tissue at terahertz frequency and the anatomical structure of human skin. Secondly, the path loss and shadow fading of the three links are compared and analyzed, and a terahertz in-vivo path loss model with equivalent absorption factor is proposed. Finally, the Rice K factor, the root mean square delay extension, and the MIMO capacity of the three links are analyzed. The results demonstrate that the terahertz in-vivo path loss model with equivalent absorption factor can more accurately describe the path loss of extended distance in vertical link 2, on-body transmitter can provide enhancement in MIMO capacity. This work can give an insight into the design and optimization of THz in-vivo communication systems.
To investigate transmission characteristics of in-vivo Mmultiple Input Multiple Output (MIMO) communication systems at TeraHertz (THz) frequency, a precise skin-fat model is constructed at 0.8～1.2 THz. Full wave electromagnetic simulations are conducted on the vertical and horizontal links in the skin-fat model, the characteristics of terahertz in-vivo channel are analyzed, and a terahertz in-vivo path loss model is established. Firstly, the skin-fat model is constructed based on the dielectric properties of human tissue at terahertz frequency and the anatomical structure of human skin. Secondly, the path loss and shadow fading of the three links are compared and analyzed, and a terahertz in-vivo path loss model with equivalent absorption factor is proposed. Finally, the Rice K factor, the root mean square delay extension, and the MIMO capacity of the three links are analyzed. The results demonstrate that the terahertz in-vivo path loss model with equivalent absorption factor can more accurately describe the path loss of extended distance in vertical link 2, on-body transmitter can provide enhancement in MIMO capacity. This work can give an insight into the design and optimization of THz in-vivo communication systems.
2024, 46(4): 1286-1295.
doi: 10.11999/JEIT230315
Abstract:
The paper proposes a hopping signal sorting method based on radio polarization features to address the problem of poor classification and recognition performance of hopping network stations under the condition of similar characteristics in the “space-time-frequency-energy” domain of user and agile frequency hopping parameters using the existing methods. Radio dual-polarization features are introduced into hopping reconnaissance, fully utilizing the differences in cross-polarization discrimination of each user and achieving accurate hopping network station sorting. To address the issue that the cross-polarization discrimination parameter of users of the same type is susceptible to noise pollution, a dual-channel dual-polarization receiving system is constructed to suppress signal noise and ensure the accuracy of polarization feature extraction. Based on the spectral clustering idea, a soft classification decision of polarization characteristics is made, which further improves the network station selection effect and achieves accurate identification of hopping signals. Simulation experiments show that under a 5 dB signal-to-noise ratio, the proposed algorithm can accurately identify multiple hopping network stations under synchronous orthogonal and non-orthogonal networking modes, with a success rate of over 99% in identification and classification, verifying the effectiveness of the new method.
The paper proposes a hopping signal sorting method based on radio polarization features to address the problem of poor classification and recognition performance of hopping network stations under the condition of similar characteristics in the “space-time-frequency-energy” domain of user and agile frequency hopping parameters using the existing methods. Radio dual-polarization features are introduced into hopping reconnaissance, fully utilizing the differences in cross-polarization discrimination of each user and achieving accurate hopping network station sorting. To address the issue that the cross-polarization discrimination parameter of users of the same type is susceptible to noise pollution, a dual-channel dual-polarization receiving system is constructed to suppress signal noise and ensure the accuracy of polarization feature extraction. Based on the spectral clustering idea, a soft classification decision of polarization characteristics is made, which further improves the network station selection effect and achieves accurate identification of hopping signals. Simulation experiments show that under a 5 dB signal-to-noise ratio, the proposed algorithm can accurately identify multiple hopping network stations under synchronous orthogonal and non-orthogonal networking modes, with a success rate of over 99% in identification and classification, verifying the effectiveness of the new method.
2024, 46(4): 1296-1305.
doi: 10.11999/JEIT230317
Abstract:
To address the low efficiency of task collaboration computation caused by limited resources of Industrial Internet of Things (IIoT) devices and dynamic changes of edge server resources, a Digital Twin (DT)-assisted task offloading algorithm is proposed for IIoT. First, the cloud-edge-end three-layer digital twin-assisted task offloading framework is constructed by the algorithm, and the approximate optimal task offloading strategy is generated in the created digital twin layer. Second, under the constraints of task computation time and energy, the joint optimization problem of user association and task partition in the computation offloading process is studied from the perspective of delay. An optimization model is established with the goal of minimizing the task offloading time and service failure penalty. Finally, a user association and task partition algorithm based on Deep Multi-Agent Parameterized Q-Network (DMAPQN) is proposed. The approximate optimal user association and task partition strategy is obtained by each intelligent agent through continuous exploration and learning, and it is issued to the physical entity network for execution. Simulation results show that the proposed task offloading algorithm effectively reduces the task collaboration computation time and provides approximate optimal offloading strategies for each computational task.
To address the low efficiency of task collaboration computation caused by limited resources of Industrial Internet of Things (IIoT) devices and dynamic changes of edge server resources, a Digital Twin (DT)-assisted task offloading algorithm is proposed for IIoT. First, the cloud-edge-end three-layer digital twin-assisted task offloading framework is constructed by the algorithm, and the approximate optimal task offloading strategy is generated in the created digital twin layer. Second, under the constraints of task computation time and energy, the joint optimization problem of user association and task partition in the computation offloading process is studied from the perspective of delay. An optimization model is established with the goal of minimizing the task offloading time and service failure penalty. Finally, a user association and task partition algorithm based on Deep Multi-Agent Parameterized Q-Network (DMAPQN) is proposed. The approximate optimal user association and task partition strategy is obtained by each intelligent agent through continuous exploration and learning, and it is issued to the physical entity network for execution. Simulation results show that the proposed task offloading algorithm effectively reduces the task collaboration computation time and provides approximate optimal offloading strategies for each computational task.
2024, 46(4): 1306-1313.
doi: 10.11999/JEIT230623
Abstract:
In the train control system of urban rail transit, Train-to-Train (T2T) communication, a new train communication mode, use direct communication between trains to reduce communication delay and improve train operation efficiency. In the scenario of the coexistence of T2T communication and Train to Ground (T2G) communication, an improved Advantage Actor-Critic-ac (A2C-ac) resource allocation algorithm based on Multi-Agent Deep Reinforcement Learning (MADRL) is proposed to solve the interference problem caused by multiplexing T2G links, and under the premise of ensuring the quality of user communication. Firstly, taking the system throughput as the optimization goal and the T2T communication transmitter as the agent, the policy network adopts a hierarchical output structure to guide the agent in selecting the spectrum resources and power level to be reused. Then the agent makes corresponding actions and interacts with the communication environment to obtain the throughput of T2G users and T2T users in the time slot. The value network evaluates the two separately and uses the weight factor\begin{document}$ \beta $\end{document}
In the train control system of urban rail transit, Train-to-Train (T2T) communication, a new train communication mode, use direct communication between trains to reduce communication delay and improve train operation efficiency. In the scenario of the coexistence of T2T communication and Train to Ground (T2G) communication, an improved Advantage Actor-Critic-ac (A2C-ac) resource allocation algorithm based on Multi-Agent Deep Reinforcement Learning (MADRL) is proposed to solve the interference problem caused by multiplexing T2G links, and under the premise of ensuring the quality of user communication. Firstly, taking the system throughput as the optimization goal and the T2T communication transmitter as the agent, the policy network adopts a hierarchical output structure to guide the agent in selecting the spectrum resources and power level to be reused. Then the agent makes corresponding actions and interacts with the communication environment to obtain the throughput of T2G users and T2T users in the time slot. The value network evaluates the two separately and uses the weight factor
2024, 46(4): 1314-1320.
doi: 10.11999/JEIT230414
Abstract:
The Intelligent-Reflecting-Surface (IRS) / Reconfigurable-Intelligent-Surface (RIS) is challenged by the placement and direction of its adhering panel when it is used to assist an actual wireless communication system to improve performance. As a mathematic programming problem, RIS placement not only depends on objective design but also is subjective to the distribution of buildings surrounded and the effective reflection area of walls to be hung with the RIS. The in-door deployment of a planar RIS is much more complex than the out-door counterpart in free and open space. The focus of this paper is on the in-door deployment of RIS adhering to environmental walls. A multi-terminal access optimization is modelled by site planning and a simplified equivalent expression is presented. A degenerated case for a single terminal is analyzed in order to transform the non-linear problem to be tractable. The function of Cassini oval is deduced from the objective and feasible solutions are narrowed to the common projection area of terminals and base-state. A heuristic and efficient algorithm is then developed based on a binary searching scheme. Numerical simulations by two in-door cases with complex constructions have verified that the proposed algorithm is benefit to speed-up computing, and extensible for multiple-RIS network planning.
The Intelligent-Reflecting-Surface (IRS) / Reconfigurable-Intelligent-Surface (RIS) is challenged by the placement and direction of its adhering panel when it is used to assist an actual wireless communication system to improve performance. As a mathematic programming problem, RIS placement not only depends on objective design but also is subjective to the distribution of buildings surrounded and the effective reflection area of walls to be hung with the RIS. The in-door deployment of a planar RIS is much more complex than the out-door counterpart in free and open space. The focus of this paper is on the in-door deployment of RIS adhering to environmental walls. A multi-terminal access optimization is modelled by site planning and a simplified equivalent expression is presented. A degenerated case for a single terminal is analyzed in order to transform the non-linear problem to be tractable. The function of Cassini oval is deduced from the objective and feasible solutions are narrowed to the common projection area of terminals and base-state. A heuristic and efficient algorithm is then developed based on a binary searching scheme. Numerical simulations by two in-door cases with complex constructions have verified that the proposed algorithm is benefit to speed-up computing, and extensible for multiple-RIS network planning.
2024, 46(4): 1321-1332.
doi: 10.11999/JEIT230445
Abstract:
In Mobile Edge Computing (MEC) intensive deployment scenarios, the uncertainty of edge server load can easily cause edge server overload, leading to a significant increase in delay and energy consumption during the computation offloading process. In response to this issue, a multi-user computation offloading optimization model and algorithm based on Deep Deterministic Policy Gradient (DDPG) is proposed. Firstly, considering the balance optimization of delay and energy consumption, a utility function is established to maximize system utility as the optimization objective, and the computational offloading problem is transformed into a mixed integer nonlinear programming problem. Then, in response to the problem of large state space and coexistence of discrete and continuous variables in the action space, the DDPG deep reinforcement learning algorithm is discretized and improved. Based on this, a multi-user computation offloading optimization method is proposed. Finally, this method is used to solve nonlinear programming problems. The simulation experimental results show that compared with existing algorithms, the proposed method can effectively reduce the probability of edge server overload and has good stability.
In Mobile Edge Computing (MEC) intensive deployment scenarios, the uncertainty of edge server load can easily cause edge server overload, leading to a significant increase in delay and energy consumption during the computation offloading process. In response to this issue, a multi-user computation offloading optimization model and algorithm based on Deep Deterministic Policy Gradient (DDPG) is proposed. Firstly, considering the balance optimization of delay and energy consumption, a utility function is established to maximize system utility as the optimization objective, and the computational offloading problem is transformed into a mixed integer nonlinear programming problem. Then, in response to the problem of large state space and coexistence of discrete and continuous variables in the action space, the DDPG deep reinforcement learning algorithm is discretized and improved. Based on this, a multi-user computation offloading optimization method is proposed. Finally, this method is used to solve nonlinear programming problems. The simulation experimental results show that compared with existing algorithms, the proposed method can effectively reduce the probability of edge server overload and has good stability.
2024, 46(4): 1333-1342.
doi: 10.11999/JEIT230394
Abstract:
As an important component of smart city construction, intelligent vehicle networking system has received increasing attention from academia and industry in recent years. In the intelligent vehicle networking system, the vehicle communication module transmits real-time data through the wireless sensor networks, improving the driving safety and travel efficiency of intelligent vehicles. The intelligent vehicle networking system is prone to data transmission interception in an open environment, causing sensitive information leakage, and even malicious attackers can anonymously forge the real identity of the intelligent vehicle to disrupt traffic order. Therefore, the intelligent vehicle networking system needs to achieve anonymous authentication and negotiate the correct session key to ensure the confidentiality of sensitive information. In this paper, an anonymous authentication and key agreement protocol is proposed for distributed intelligent vehicle networking system architecture. The protocol protects the authentication identifier based on the secret sharing technology of the Chinese Remainder Theorem. The intelligent vehicle can recover the corresponding identifier in different regions with linear computing overhead. The identifier can be used safely for a long time, and the intelligent vehicle can complete security authentication without using tamper proof devices. The roadside communication base station can check the anonymity and integrity of information, and negotiate the session key for subsequent secure communication with the intelligent vehicle, while achieving bidirectional authentication. In addition, the protocol can expand practical functions such as batch anonymous authentication, domain key update, vehicle to vehicle anonymous authentication, and anonymous identity traceability in complex distributed intelligent vehicle networking system. The security and performance analysis shows that the protocol can be safely and efficiently deployed in a distributed intelligent vehicle networking environment.
As an important component of smart city construction, intelligent vehicle networking system has received increasing attention from academia and industry in recent years. In the intelligent vehicle networking system, the vehicle communication module transmits real-time data through the wireless sensor networks, improving the driving safety and travel efficiency of intelligent vehicles. The intelligent vehicle networking system is prone to data transmission interception in an open environment, causing sensitive information leakage, and even malicious attackers can anonymously forge the real identity of the intelligent vehicle to disrupt traffic order. Therefore, the intelligent vehicle networking system needs to achieve anonymous authentication and negotiate the correct session key to ensure the confidentiality of sensitive information. In this paper, an anonymous authentication and key agreement protocol is proposed for distributed intelligent vehicle networking system architecture. The protocol protects the authentication identifier based on the secret sharing technology of the Chinese Remainder Theorem. The intelligent vehicle can recover the corresponding identifier in different regions with linear computing overhead. The identifier can be used safely for a long time, and the intelligent vehicle can complete security authentication without using tamper proof devices. The roadside communication base station can check the anonymity and integrity of information, and negotiate the session key for subsequent secure communication with the intelligent vehicle, while achieving bidirectional authentication. In addition, the protocol can expand practical functions such as batch anonymous authentication, domain key update, vehicle to vehicle anonymous authentication, and anonymous identity traceability in complex distributed intelligent vehicle networking system. The security and performance analysis shows that the protocol can be safely and efficiently deployed in a distributed intelligent vehicle networking environment.
2024, 46(4): 1343-1352.
doi: 10.11999/JEIT230421
Abstract:
In intelligent driving based on federated learning, the resource constraints of Intelligent Connected Vehicle (ICV) and possible device failures will lead to the decrease of the precision of federated learning training and the increase of delay and energy consumption. Therefore, an optimization scheme of edge selection and resource allocation in digital twin-assisted federated learning is proposed. Firstly, a digital twin-assisted federated learning mechanism is proposed, allowing ICV to choose to participate in federated learning locally or through its digital twin. Secondly, by constructing a computational and communication model for digital twin-assisted federated learning, an edge selection and computing resource allocation joint optimization problem is established with the objective of minimizing cumulative training delay and energy consumption, and is transformed into a partially observable Markov decision process. Finally, an edge selection and resource allocation algorithm based on Multi-agent Parametrized Deep Q-Networks (MPDQN) is proposed to learn approximately optimal edge selection and resource allocation strategies to minimize federated learning cumulative delay and energy consumption. Simulation results show that the proposed algorithm can effectively reduce cumulative training delay and energy consumption of federated learning training while ensuring model accuracy.
In intelligent driving based on federated learning, the resource constraints of Intelligent Connected Vehicle (ICV) and possible device failures will lead to the decrease of the precision of federated learning training and the increase of delay and energy consumption. Therefore, an optimization scheme of edge selection and resource allocation in digital twin-assisted federated learning is proposed. Firstly, a digital twin-assisted federated learning mechanism is proposed, allowing ICV to choose to participate in federated learning locally or through its digital twin. Secondly, by constructing a computational and communication model for digital twin-assisted federated learning, an edge selection and computing resource allocation joint optimization problem is established with the objective of minimizing cumulative training delay and energy consumption, and is transformed into a partially observable Markov decision process. Finally, an edge selection and resource allocation algorithm based on Multi-agent Parametrized Deep Q-Networks (MPDQN) is proposed to learn approximately optimal edge selection and resource allocation strategies to minimize federated learning cumulative delay and energy consumption. Simulation results show that the proposed algorithm can effectively reduce cumulative training delay and energy consumption of federated learning training while ensuring model accuracy.
2024, 46(4): 1353-1362.
doi: 10.11999/JEIT230669
Abstract:
Focusing on the problem of large channel estimation transmission overhead in Intelligent Reflective Surface IRS) assisted multi-user communication system in high Doppler scenario, an IRS-OTFS communication system is constructed based on the characteristics of Orthogonal Time-Frequency Space (OTFS) modulation, which gives full play to the performance advantages of IRS and OTFS, and on this basis, a Model-Agnostic Meta-Learning (MAML) algorithm with adaptive learning rate is proposed. The IRS-OTFS multi-user channel estimation task is trained offline, the learning rate is adaptively adjusted according to the convergence speed of each task to prevent training imbalance, and the correlation between channels and the few samples and generalization characteristics of MAML algorithm are used to obtain global models and adaptive models, so as to quickly learn the transmission characteristics of new user channels, reduce transmission overhead, and improve the accuracy of channel estimation. Theoretical analysis and simulation results show that the algorithm reduces the transmission overhead by about 50% under the same channel transmission conditions, and has a performance improvement of about 4.8 dB compared with the benchmark algorithm.
Focusing on the problem of large channel estimation transmission overhead in Intelligent Reflective Surface IRS) assisted multi-user communication system in high Doppler scenario, an IRS-OTFS communication system is constructed based on the characteristics of Orthogonal Time-Frequency Space (OTFS) modulation, which gives full play to the performance advantages of IRS and OTFS, and on this basis, a Model-Agnostic Meta-Learning (MAML) algorithm with adaptive learning rate is proposed. The IRS-OTFS multi-user channel estimation task is trained offline, the learning rate is adaptively adjusted according to the convergence speed of each task to prevent training imbalance, and the correlation between channels and the few samples and generalization characteristics of MAML algorithm are used to obtain global models and adaptive models, so as to quickly learn the transmission characteristics of new user channels, reduce transmission overhead, and improve the accuracy of channel estimation. Theoretical analysis and simulation results show that the algorithm reduces the transmission overhead by about 50% under the same channel transmission conditions, and has a performance improvement of about 4.8 dB compared with the benchmark algorithm.
2024, 46(4): 1363-1372.
doi: 10.11999/JEIT230632
Abstract:
A delay-optimized Service Function Chain (SFC) deployment approach is proposed by combining deep reinforcement learning with the delay-based Dijkstra pathfinding algorithm for the problem of resource-constrained edge networks and low end-to-end delay tolerance for service flows. Firstly, an attention mechanism-based Sequence to Sequence (Seq2Seq) agent network and a delay-based Dijkstra pathfinding algorithm are designed for generating Virtual Network Function(VNF) deployments and link mapping for SFC, while the constraint problem of the delay optimization model is considered and incorporated into the reinforcement learning objective function using Lagrangian relaxation techniques; Secondly, to assist the network agent in converging quickly, a baseline evaluator network is used to assess the expected reward value of the deployment strategy; Finally, in the testing phase, the deployment strategy of the agent is improved by reducing the probability of convergence of the network to a local optimum through greedy search and sampling techniques. Comparison experiments show that the method reduces the latency by about 10% and 86.3% than the First-Fit algorithm and TabuSearch algorithm, respectively, and is about 74.2% and 84.4% more stable than these two algorithms in the case of limited network resources. This method provides a more stable end-to-end service with lower latency, enabling a better experience for latency-sensitive services.
A delay-optimized Service Function Chain (SFC) deployment approach is proposed by combining deep reinforcement learning with the delay-based Dijkstra pathfinding algorithm for the problem of resource-constrained edge networks and low end-to-end delay tolerance for service flows. Firstly, an attention mechanism-based Sequence to Sequence (Seq2Seq) agent network and a delay-based Dijkstra pathfinding algorithm are designed for generating Virtual Network Function(VNF) deployments and link mapping for SFC, while the constraint problem of the delay optimization model is considered and incorporated into the reinforcement learning objective function using Lagrangian relaxation techniques; Secondly, to assist the network agent in converging quickly, a baseline evaluator network is used to assess the expected reward value of the deployment strategy; Finally, in the testing phase, the deployment strategy of the agent is improved by reducing the probability of convergence of the network to a local optimum through greedy search and sampling techniques. Comparison experiments show that the method reduces the latency by about 10% and 86.3% than the First-Fit algorithm and TabuSearch algorithm, respectively, and is about 74.2% and 84.4% more stable than these two algorithms in the case of limited network resources. This method provides a more stable end-to-end service with lower latency, enabling a better experience for latency-sensitive services.
2024, 46(4): 1373-1381.
doi: 10.11999/JEIT230462
Abstract:
An elevation-dependent stochastic localization algorithm is proposed to address the problem of different satellites contributing differently to the localization error in Global Navigation Satellite System (GNSS)-based passive radar. Herein, the Cramer-Rao Lower Bound (CRLB) and statistical characteristics of the target position estimator are theoretically analyzed, and the contributions of satellite position and ground-station position errors to passive localization error are calculated. Simulation results show that the proposed algorithm can reasonably distribute the error from the pseudo-range measurements of multiple GNSS satellites with different directions and reflective paths. The localization performance reaches the CRLB but will not deteriorate considerably due to a change in the star selection scheme. Analysis of the ground-station position and satellite position errors show that their contributions to the total positioning error can be ignored if the standard deviation of the ground-station position is less than 10 cm and that of the satellite position is less than 1 km.
An elevation-dependent stochastic localization algorithm is proposed to address the problem of different satellites contributing differently to the localization error in Global Navigation Satellite System (GNSS)-based passive radar. Herein, the Cramer-Rao Lower Bound (CRLB) and statistical characteristics of the target position estimator are theoretically analyzed, and the contributions of satellite position and ground-station position errors to passive localization error are calculated. Simulation results show that the proposed algorithm can reasonably distribute the error from the pseudo-range measurements of multiple GNSS satellites with different directions and reflective paths. The localization performance reaches the CRLB but will not deteriorate considerably due to a change in the star selection scheme. Analysis of the ground-station position and satellite position errors show that their contributions to the total positioning error can be ignored if the standard deviation of the ground-station position is less than 10 cm and that of the satellite position is less than 1 km.
2024, 46(4): 1382-1390.
doi: 10.11999/JEIT230665
Abstract:
Radio Frequency Interference (RFI) will pollute the echo signal of Synthetic Aperture Radar (SAR) and increase the difficulty of SAR image interpretation. Pulsed Direct Wave Interference (PDWI), as a typical RFI, covers SAR echo information with bright fringes in the original echo domain, which has a serious impact on SAR imaging quality. Among the existing interference suppression methods, the traditional Eigen Subspace Projection (ESP) method suppresses the whole pulse containing interference, which leads to the loss of useful signals in the non-interference positions in the pulse. In order to protect useful signals, an improved ESP SAR pulsed direct wave interference suppression method is proposed. Firstly, the specific position of PDWI in the time domain is obtained by twice detecting interference. Secondly, ESP operation is used to separate the useful signal from the interference signal only for the detected interference position data. Finally, the interference data of ESP reconstruction is subtracted from the original data to achieve interference suppression. Simulation and measured data processing show that, compared with the existing methods, this method can effectively avoid the loss of useful signals in SAR raw data and suppress the pulsed direct wave interference.
Radio Frequency Interference (RFI) will pollute the echo signal of Synthetic Aperture Radar (SAR) and increase the difficulty of SAR image interpretation. Pulsed Direct Wave Interference (PDWI), as a typical RFI, covers SAR echo information with bright fringes in the original echo domain, which has a serious impact on SAR imaging quality. Among the existing interference suppression methods, the traditional Eigen Subspace Projection (ESP) method suppresses the whole pulse containing interference, which leads to the loss of useful signals in the non-interference positions in the pulse. In order to protect useful signals, an improved ESP SAR pulsed direct wave interference suppression method is proposed. Firstly, the specific position of PDWI in the time domain is obtained by twice detecting interference. Secondly, ESP operation is used to separate the useful signal from the interference signal only for the detected interference position data. Finally, the interference data of ESP reconstruction is subtracted from the original data to achieve interference suppression. Simulation and measured data processing show that, compared with the existing methods, this method can effectively avoid the loss of useful signals in SAR raw data and suppress the pulsed direct wave interference.
2024, 46(4): 1391-1398.
doi: 10.11999/JEIT230622
Abstract:
To solve the problems of pixel points overlap and low processing efficiency in existing end-to-end radar signal deinterleaving methods based on image segmentation, an end-to-end sorting method using a point cloud segmentation network is proposed in this paper. Firstly, the Pulse Description Words (PWD) of radar pulse stream are mapped to point clouds. Then, the PointNet++ is used to segment each point according to its radiation source. Finally, the points with the same label are clustered to form pulse sets, and the radiation sources within each pulse set are then extracted to form corresponding emitter description words. The simulation results demonstrate that the proposed method can effectively separate unknown radar signals while maintaining reliability and stability, even in scenarios with pulse loss and false pulse interference. Additionally, the implementation efficiency of this method is higher because of the model with lightweight characteristics.
To solve the problems of pixel points overlap and low processing efficiency in existing end-to-end radar signal deinterleaving methods based on image segmentation, an end-to-end sorting method using a point cloud segmentation network is proposed in this paper. Firstly, the Pulse Description Words (PWD) of radar pulse stream are mapped to point clouds. Then, the PointNet++ is used to segment each point according to its radiation source. Finally, the points with the same label are clustered to form pulse sets, and the radiation sources within each pulse set are then extracted to form corresponding emitter description words. The simulation results demonstrate that the proposed method can effectively separate unknown radar signals while maintaining reliability and stability, even in scenarios with pulse loss and false pulse interference. Additionally, the implementation efficiency of this method is higher because of the model with lightweight characteristics.
2024, 46(4): 1399-1407.
doi: 10.11999/JEIT230650
Abstract:
In order to solve the problem that the estimation accuracy of traditional methods for estimating the shape parameters of radar K-distributed sea clutter is seriously degraded when there are outliers, a method for estimating the shape parameters of radar K-distributed sea clutter based on Support Vector Regression (SVR) and sample percentile ratio is proposed in this paper. Firstly, the clutter parameters and the percentile ranks are given, the sample percentile ratio and its logarithm are calculated according to the cumulative distribution function of the K distribution, and then an SVR model with the logarithm of the sample quantile ratio as input and the shape parameters to be estimated as output is established. The hyperparameters of SVR model are determined by cross-validation, and finally the SVR model is trained to estimate the shape parameter of K-distributed sea clutter robustly and accurately. The simulated and measured radar data show that the estimation error of the proposed method is lower than that of the conventional moment-based methods, and its estimation performance is similar to that of the percentile-based methods. Moreover, compared with the existing percentile-based methods, the hyperparameters of the proposed method are easy to determine, and it does not depend on table lookup.
In order to solve the problem that the estimation accuracy of traditional methods for estimating the shape parameters of radar K-distributed sea clutter is seriously degraded when there are outliers, a method for estimating the shape parameters of radar K-distributed sea clutter based on Support Vector Regression (SVR) and sample percentile ratio is proposed in this paper. Firstly, the clutter parameters and the percentile ranks are given, the sample percentile ratio and its logarithm are calculated according to the cumulative distribution function of the K distribution, and then an SVR model with the logarithm of the sample quantile ratio as input and the shape parameters to be estimated as output is established. The hyperparameters of SVR model are determined by cross-validation, and finally the SVR model is trained to estimate the shape parameter of K-distributed sea clutter robustly and accurately. The simulated and measured radar data show that the estimation error of the proposed method is lower than that of the conventional moment-based methods, and its estimation performance is similar to that of the percentile-based methods. Moreover, compared with the existing percentile-based methods, the hyperparameters of the proposed method are easy to determine, and it does not depend on table lookup.
2024, 46(4): 1408-1417.
doi: 10.11999/JEIT230413
Abstract:
To reduce the mapping basepoint offset and Gaussian truncation errors caused by spatiotemporal inconsistency in angle-only maneuvering target tracking systems, mapping representation and\begin{document}$ {\ell _1} $\end{document} \begin{document}$ {\ell _{2,1}} $\end{document}
To reduce the mapping basepoint offset and Gaussian truncation errors caused by spatiotemporal inconsistency in angle-only maneuvering target tracking systems, mapping representation and
2024, 46(4): 1418-1427.
doi: 10.11999/JEIT230464
Abstract:
It is difficult to retrieve the ocean wind direction using the specular reflection signal owing to its insensitivity to the sea surface wind direction. The sensitivity of the scattered Global Navigation Satellite System-Reflectometry (GNSS-R) signal from the non-specular geometry to wind direction is first investigated in this paper. The sub-satellite non-specular observation mode is proposed, and the observable quanitity sensitive to wind direction in this mode is defined. Based on this mode, the wind vector retrieval algorithm using spaceborne GNSS-R based on the Maximum Likelihood Estimation (MLE) is presented, in which the sub-satellite non-specular scattering signals from two and more navigation satellites are used to retrieve ocean wind vector. A simulator is developed to demonstrate and test the proposed algorithm. The results show that due to the ambiguity of the observation geometry and the symmetry of the ocean spectrum, there are four uncertain solutions in the retrieved wind directions. The ambiguity of the measurement geometry can be eliminated by using the multi-satellite observation, and only two possible solutions of wind directions still are remained. When the scattered signals from three satellites are used, and the Signal-to-Noise Ratio (SNR) is over 11 dB, the Root Mean Square Error (RMSE) of the retrieved wind speed and direction are less 2 m/s and 15° with a wind speed rang of 2～25 m/s.
It is difficult to retrieve the ocean wind direction using the specular reflection signal owing to its insensitivity to the sea surface wind direction. The sensitivity of the scattered Global Navigation Satellite System-Reflectometry (GNSS-R) signal from the non-specular geometry to wind direction is first investigated in this paper. The sub-satellite non-specular observation mode is proposed, and the observable quanitity sensitive to wind direction in this mode is defined. Based on this mode, the wind vector retrieval algorithm using spaceborne GNSS-R based on the Maximum Likelihood Estimation (MLE) is presented, in which the sub-satellite non-specular scattering signals from two and more navigation satellites are used to retrieve ocean wind vector. A simulator is developed to demonstrate and test the proposed algorithm. The results show that due to the ambiguity of the observation geometry and the symmetry of the ocean spectrum, there are four uncertain solutions in the retrieved wind directions. The ambiguity of the measurement geometry can be eliminated by using the multi-satellite observation, and only two possible solutions of wind directions still are remained. When the scattered signals from three satellites are used, and the Signal-to-Noise Ratio (SNR) is over 11 dB, the Root Mean Square Error (RMSE) of the retrieved wind speed and direction are less 2 m/s and 15° with a wind speed rang of 2～25 m/s.
2024, 46(4): 1428-1437.
doi: 10.11999/JEIT230573
Abstract:
The image-based water flow measurement technology based on the Direct Linear Transformation (DLT) method relies on ground control points, and has problems such as low efficiency, high risk, and difficult operation in natural rivers with wide sections. A camera attitude angle calibration method based on Cross-Section Waterline Orientation (CSWO) is proposed. In this method, the calibration process is divided into two steps: laboratory calibration and field calibration, and the latter is divided into camera positioning and orientation. In the orientation step, the optical axis is required to be aligned with the cross-section when the camera is installed, so that the azimuth angle is set to zero. The roll angle is calculated by the slope of the straight waterline marked manually in the undistorted image. Then the sub-pixel image coordinate of the intersection point between the waterline and the vertical axis of the image is calculated. Finally, the pitch angle is calculated according to the perspective projection imaging model combined with the object coordinates obtained by the water level and the interpolated elevation of cross-section. This method has been applied to Space-Time Image Velocimetry (STIV) to measure the image-free surface velocity of a river with a width of 200 m. The results show that the maximum absolute error of starting distance is 0.59 m, the maximum relative error is 0.45%, and the maximum relative error of surface velocity is less than 6.3%.
The image-based water flow measurement technology based on the Direct Linear Transformation (DLT) method relies on ground control points, and has problems such as low efficiency, high risk, and difficult operation in natural rivers with wide sections. A camera attitude angle calibration method based on Cross-Section Waterline Orientation (CSWO) is proposed. In this method, the calibration process is divided into two steps: laboratory calibration and field calibration, and the latter is divided into camera positioning and orientation. In the orientation step, the optical axis is required to be aligned with the cross-section when the camera is installed, so that the azimuth angle is set to zero. The roll angle is calculated by the slope of the straight waterline marked manually in the undistorted image. Then the sub-pixel image coordinate of the intersection point between the waterline and the vertical axis of the image is calculated. Finally, the pitch angle is calculated according to the perspective projection imaging model combined with the object coordinates obtained by the water level and the interpolated elevation of cross-section. This method has been applied to Space-Time Image Velocimetry (STIV) to measure the image-free surface velocity of a river with a width of 200 m. The results show that the maximum absolute error of starting distance is 0.59 m, the maximum relative error is 0.45%, and the maximum relative error of surface velocity is less than 6.3%.
2024, 46(4): 1438-1447.
doi: 10.11999/JEIT230405
Abstract:
At present, view-based 3D model classification has the problems of insufficient visual information for single view and redundant information for multiple views, and treating all views equally will ignore the differences between different projection angles. To solve the above problems, a 3D model classification method based on Shannon entropy representative feature and voting mechanism is proposed. Firstly, multiple angle groups are set uniformly around 3D model, and multiple view sets representing the model are obtained. In order to extract effectively deep features from view, channel attention mechanism is introduced into the feature extraction network. Secondly, based on view discriminative features output from Softmax function, Shannon entropy is used to select representative feature for avoiding redundant feature of multiple views. Finally, based on representative features from multiple angle groups, voting mechanism is used to classify 3D model. Experiments show that the classification accuracy of the proposed method on 3D model dataset ModelNet10 reaches 96.48%, and classification performance is outstanding.
At present, view-based 3D model classification has the problems of insufficient visual information for single view and redundant information for multiple views, and treating all views equally will ignore the differences between different projection angles. To solve the above problems, a 3D model classification method based on Shannon entropy representative feature and voting mechanism is proposed. Firstly, multiple angle groups are set uniformly around 3D model, and multiple view sets representing the model are obtained. In order to extract effectively deep features from view, channel attention mechanism is introduced into the feature extraction network. Secondly, based on view discriminative features output from Softmax function, Shannon entropy is used to select representative feature for avoiding redundant feature of multiple views. Finally, based on representative features from multiple angle groups, voting mechanism is used to classify 3D model. Experiments show that the classification accuracy of the proposed method on 3D model dataset ModelNet10 reaches 96.48%, and classification performance is outstanding.
2024, 46(4): 1448-1457.
doi: 10.11999/JEIT230651
Abstract:
HyperSpectral Image (HSI) classification methods based on limited labeled samples have made significant progress in recent years. However, due to the specificity of hyperspectral images, redundant information and limited labeled samples pose great challenges for extracting highly discriminative features. In addition, owing to the uneven distribution of pixels in each category, how to strengthen the role of central pixels and attenuate the negative impact of surrounding pixels with different categories is also the key to improve the classification performance. To overcome the above limitations, an HSI classification method based on Multi-Scale Asymmetric Dense Network (MS-ADNet) is proposed. Firstly, a multi-scale sample construction module is proposed, which extracts multiple scale patches around each pixel and performs deconvolution and stitching to construct multiscale input samples that contain both detailed structural regions and large homogeneous regions. Next, an asymmetric densely connected structure is proposed to achieve kernel skeleton enhancement in joint spatial and spectral feature extraction, i.e., enhancement of features extracted from the central cross-skeleton portion of a square convolutional kernel, which effectively facilitates feature reuse. Moreover, to improve the discriminability of spectral features, a streamlined element spectral attention mechanism is proposed and placed at the front and back ends of the densely connected network. With only five samples per class used for network training, the proposed method achieves competitive classification results with overall accuracies of 77.66%, 84.54%, and 92.39% on the Indiana Pines, Pavia University, and Salinas datasets, respectively.
HyperSpectral Image (HSI) classification methods based on limited labeled samples have made significant progress in recent years. However, due to the specificity of hyperspectral images, redundant information and limited labeled samples pose great challenges for extracting highly discriminative features. In addition, owing to the uneven distribution of pixels in each category, how to strengthen the role of central pixels and attenuate the negative impact of surrounding pixels with different categories is also the key to improve the classification performance. To overcome the above limitations, an HSI classification method based on Multi-Scale Asymmetric Dense Network (MS-ADNet) is proposed. Firstly, a multi-scale sample construction module is proposed, which extracts multiple scale patches around each pixel and performs deconvolution and stitching to construct multiscale input samples that contain both detailed structural regions and large homogeneous regions. Next, an asymmetric densely connected structure is proposed to achieve kernel skeleton enhancement in joint spatial and spectral feature extraction, i.e., enhancement of features extracted from the central cross-skeleton portion of a square convolutional kernel, which effectively facilitates feature reuse. Moreover, to improve the discriminability of spectral features, a streamlined element spectral attention mechanism is proposed and placed at the front and back ends of the densely connected network. With only five samples per class used for network training, the proposed method achieves competitive classification results with overall accuracies of 77.66%, 84.54%, and 92.39% on the Indiana Pines, Pavia University, and Salinas datasets, respectively.
2024, 46(4): 1458-1467.
doi: 10.11999/JEIT230496
Abstract:
In long-term visual tracking, most of the target loss discriminative methods require artificially determined thresholds, and the selection of optimal thresholds is usually difficult, resulting in weak generalization ability of long-term tracking algorithms. A target drift Discriminative Network (DNet) that does not require artificially selected thresholds is proposed. The network adopts Siamese structure and uses both static and dynamic templates to determine whether the tracking results are lost or not. Among them, the introduction of dynamic templates effectively improves the algorithm’s ability to adapt to changes in target appearance. In order to train the proposed target drift discriminative network, a sample-rich dataset is established. To verify the effectiveness of the proposed network, a complete long-term tracking algorithm is constructed in this paper by combining this network with the base tracker and the re-detection module. It is tested on classical visual tracking datasets such as UAV20L, LaSOT, VOT2018-LT and VOT2020-LT. The experimental results show that compared with the base tracker, the tracking accuracy and success rate are improved by 10.4% and 7.5% on UAV20L dataset, respectively.
In long-term visual tracking, most of the target loss discriminative methods require artificially determined thresholds, and the selection of optimal thresholds is usually difficult, resulting in weak generalization ability of long-term tracking algorithms. A target drift Discriminative Network (DNet) that does not require artificially selected thresholds is proposed. The network adopts Siamese structure and uses both static and dynamic templates to determine whether the tracking results are lost or not. Among them, the introduction of dynamic templates effectively improves the algorithm’s ability to adapt to changes in target appearance. In order to train the proposed target drift discriminative network, a sample-rich dataset is established. To verify the effectiveness of the proposed network, a complete long-term tracking algorithm is constructed in this paper by combining this network with the base tracker and the re-detection module. It is tested on classical visual tracking datasets such as UAV20L, LaSOT, VOT2018-LT and VOT2020-LT. The experimental results show that compared with the base tracker, the tracking accuracy and success rate are improved by 10.4% and 7.5% on UAV20L dataset, respectively.
2024, 46(4): 1468-1478.
doi: 10.11999/JEIT230291
Abstract:
Transcranial Magnetic Stimulation (TMS) has been widely used in clinical neuroregulation fields such as improving brain cognitive function because of its ability to non-invasibly detect and modulate the excitability and function of the cerebral cortex. The functional realization of working memory requires the synchronous activity of multiple brain regions. In this paper, it is of great significance to combine behavioral and electrophysiology, establish causal network connections across brain regions, and explore the regulatory mechanism of different modes of TMS on brain cognitive function from the perspective of memory-related brain regions. Firstly, repetitive Transcranial Magnetic Stimulation (rTMS) and intermittent Theta Burst Stimulation (iTBS) are performed in aged Wistar rats, a blank control group is established at the same time, and the Local Field Potentials (LFPs) are collected by an in vivo multichannel microelectrode array during a working memory task. Then, the LFPs brain causal network is constructed based on the directed transfer function. Finally, by comparing the behavioral results and the causal network parameters of each brain region, the effects of TMS on the working memory behavior and the information synergy between brain regions are explored. The results show that the average number of days that rats in the rTMS and iTBS groups performed the correct working memory task is reduced, and the average correct rate is higher than that in the blank control group. After stimulation, the bidirectional network connectivity between prefrontal lobe and hippocampus in the rTMS group and the iTBS group is significantly enhanced, and the strength of information flow and causal streamability are significantly improved (P < 0.05). Therefore, both rTMS and iTBS can promote the communication between the hippocampus and prefrontal cortex, thereby improving the working memory ability of aged rats.
Transcranial Magnetic Stimulation (TMS) has been widely used in clinical neuroregulation fields such as improving brain cognitive function because of its ability to non-invasibly detect and modulate the excitability and function of the cerebral cortex. The functional realization of working memory requires the synchronous activity of multiple brain regions. In this paper, it is of great significance to combine behavioral and electrophysiology, establish causal network connections across brain regions, and explore the regulatory mechanism of different modes of TMS on brain cognitive function from the perspective of memory-related brain regions. Firstly, repetitive Transcranial Magnetic Stimulation (rTMS) and intermittent Theta Burst Stimulation (iTBS) are performed in aged Wistar rats, a blank control group is established at the same time, and the Local Field Potentials (LFPs) are collected by an in vivo multichannel microelectrode array during a working memory task. Then, the LFPs brain causal network is constructed based on the directed transfer function. Finally, by comparing the behavioral results and the causal network parameters of each brain region, the effects of TMS on the working memory behavior and the information synergy between brain regions are explored. The results show that the average number of days that rats in the rTMS and iTBS groups performed the correct working memory task is reduced, and the average correct rate is higher than that in the blank control group. After stimulation, the bidirectional network connectivity between prefrontal lobe and hippocampus in the rTMS group and the iTBS group is significantly enhanced, and the strength of information flow and causal streamability are significantly improved (P < 0.05). Therefore, both rTMS and iTBS can promote the communication between the hippocampus and prefrontal cortex, thereby improving the working memory ability of aged rats.
2024, 46(4): 1479-1487.
doi: 10.11999/JEIT230181
Abstract:
Due to the foreground occlusion of the reference texture and the difference in angle-of-views, many holes can be found in the synthesized images produced by depth image-based virtual view rendering. Prior disocclusion methods are time-consuming and need more texture consistency between hole-filled regions and the synthesized image. In this paper, depth maps are first pre-processed to reduce foreground penetration during hole filling. Then, for holes in the synthesized images after 3D warping, an image generation network based on the architecture of a Generative Adversarial Network (GAN) is designed to fill the holes. This network consists of two sub-networks. The first network generates the texture and structure information of hole regions, while the second network adopts an attention module combining contextual feature fusion to improve the quality of the hole-filled regions. The proposed network can effectively solve the problem of the hole-filling areas being prone to producing artifacts when fast motion exist in the foreground objects. Experimental results on multi-view video plus depth sequences show that the proposed method is superior to the existing methods in both subjective and objective quality.
Due to the foreground occlusion of the reference texture and the difference in angle-of-views, many holes can be found in the synthesized images produced by depth image-based virtual view rendering. Prior disocclusion methods are time-consuming and need more texture consistency between hole-filled regions and the synthesized image. In this paper, depth maps are first pre-processed to reduce foreground penetration during hole filling. Then, for holes in the synthesized images after 3D warping, an image generation network based on the architecture of a Generative Adversarial Network (GAN) is designed to fill the holes. This network consists of two sub-networks. The first network generates the texture and structure information of hole regions, while the second network adopts an attention module combining contextual feature fusion to improve the quality of the hole-filled regions. The proposed network can effectively solve the problem of the hole-filling areas being prone to producing artifacts when fast motion exist in the foreground objects. Experimental results on multi-view video plus depth sequences show that the proposed method is superior to the existing methods in both subjective and objective quality.
2024, 46(4): 1488-1498.
doi: 10.11999/JEIT230558
Abstract:
In order to improve the success rate and reconstruction accuracy of the compressed sensing reconstruction algorithm, the Look Ahead and Regular Backtracking Orthogonal Matching Pursuit based on Dice coefficient (DLARBOMP) is proposed. In this algorithm, from the perspective of matching criteria and atom selection in the pre-selection stage, the Dice coefficient is used to replace the atomic inner product to calculate the correlation value and preserve the characteristics of the original signal, to select the atom that best matches the residual and improve the reconstruction accuracy. At the same time, to reduce backtracking time in the reconstruction process, regularization is used to select multiple atoms instead of a single atom in each iteration, achieving a balance between reconstruction accuracy and time. Finally, the experimental results of sparse one-dimensional signal and two-dimensional image signal reconstruction show that the proposed DLARBOMP algorithm considers both performance and efficiency when reconstructing one-dimensional signal, and enhances the Peak Signal-to-Noise Ratio (PSNR) when reconstructing two-dimensional compressed image signal, as compared to Orthogonal Matching Pursuit (OMP) and the state-of-the-art greedy algorithms.
In order to improve the success rate and reconstruction accuracy of the compressed sensing reconstruction algorithm, the Look Ahead and Regular Backtracking Orthogonal Matching Pursuit based on Dice coefficient (DLARBOMP) is proposed. In this algorithm, from the perspective of matching criteria and atom selection in the pre-selection stage, the Dice coefficient is used to replace the atomic inner product to calculate the correlation value and preserve the characteristics of the original signal, to select the atom that best matches the residual and improve the reconstruction accuracy. At the same time, to reduce backtracking time in the reconstruction process, regularization is used to select multiple atoms instead of a single atom in each iteration, achieving a balance between reconstruction accuracy and time. Finally, the experimental results of sparse one-dimensional signal and two-dimensional image signal reconstruction show that the proposed DLARBOMP algorithm considers both performance and efficiency when reconstructing one-dimensional signal, and enhances the Peak Signal-to-Noise Ratio (PSNR) when reconstructing two-dimensional compressed image signal, as compared to Orthogonal Matching Pursuit (OMP) and the state-of-the-art greedy algorithms.
2024, 46(4): 1499-1512.
doi: 10.11999/JEIT230509
Abstract:
With the development of Deep Learning, the number of parameters and computation of Convolutional Neural Network (CNN) increases dramatically, which greatly raises the cost of deploying CNN algorithms on edge devices. To reduce the difficulty of the deployment and decrease the inference latency and energy consumption of CNN on the edge side, a Reconfigurable CNN Co-Processor for edge computing is proposed. Based on the data flow pattern of channel-wise processing, the proposed two-level distributed storage scheme solves the problem of power consumption overhead and performance degradation caused by large data movement between PE units and large-scale migration of intermediate data on chip. To avoid the complex data interconnection network propagation mechanism in PE arrays and reduce the complexity of control, a flexible local access mechanism and a padding mechanism based on address translation are proposed, which can perform conventional convolution, deep separable convolution, pooling and fully connected operations with great flexibility. The proposed co-processor contains 256 Processing Elements (PEs) and 176 kB on-chip SRAM. Synthesized and post-layout with 55-nm TT Corner CMOS process (25 °C, 1.2 V), the CNN co-processor achieves a maximum clock frequency of 328 MHz and an area of 4.41 mm2. The peak performance of the co-processor is 163.8 GOPs at 320 MHz and the area efficiency is 37.14 GOPs/mm2, the energy efficiency of LeNet-5 and MobileNet are 210.7 GOPs/W and 340.08 GOPs/W, respectively, which is able to meet the energy-efficiency and performance requirements of edge intelligent computing scenarios.
With the development of Deep Learning, the number of parameters and computation of Convolutional Neural Network (CNN) increases dramatically, which greatly raises the cost of deploying CNN algorithms on edge devices. To reduce the difficulty of the deployment and decrease the inference latency and energy consumption of CNN on the edge side, a Reconfigurable CNN Co-Processor for edge computing is proposed. Based on the data flow pattern of channel-wise processing, the proposed two-level distributed storage scheme solves the problem of power consumption overhead and performance degradation caused by large data movement between PE units and large-scale migration of intermediate data on chip. To avoid the complex data interconnection network propagation mechanism in PE arrays and reduce the complexity of control, a flexible local access mechanism and a padding mechanism based on address translation are proposed, which can perform conventional convolution, deep separable convolution, pooling and fully connected operations with great flexibility. The proposed co-processor contains 256 Processing Elements (PEs) and 176 kB on-chip SRAM. Synthesized and post-layout with 55-nm TT Corner CMOS process (25 °C, 1.2 V), the CNN co-processor achieves a maximum clock frequency of 328 MHz and an area of 4.41 mm2. The peak performance of the co-processor is 163.8 GOPs at 320 MHz and the area efficiency is 37.14 GOPs/mm2, the energy efficiency of LeNet-5 and MobileNet are 210.7 GOPs/W and 340.08 GOPs/W, respectively, which is able to meet the energy-efficiency and performance requirements of edge intelligent computing scenarios.
2024, 46(4): 1513-1520.
doi: 10.11999/JEIT230361
Abstract:
A three-dimensional electric field sensing chip equipped with a single shielding electrode and piezoelectric actuation is proposed. This design achieves high-sensitivity detection of the three-dimensional electric fields, simultaneously reducing excitation voltage and crosstalk noise is proposed. The sensing structure comprises one group of shielding electrodes and four sets of symmetrically distributed sensing electrodes. In response to piezoelectric actuation, the shielding electrodes undergo vertical vibrations, while the four sensing electrode sets generate induced currents when subjected to external electric fields. A differential decoupling method can be used to calculate the signals corresponding to the electric field components along the x-, y-, and z-axes. Finite element simulation was conducted to design the structure of the three-dimensional electric field sensing chip, analyze the feasibility of its measurement, and optimize key structural parameters. The fabrication process for the sensing chip was designed and implemented. Experimental results reveal that the output sensitivities are 0.2214 mV/(kV/m) for the x-axis, 0.3580 mV/(kV/m) for the y-axis, and 2.1768 mV/(kV/m) for the z-axis. The maximum measurement error for the three-dimensional electric fields remains < 5.3%.
A three-dimensional electric field sensing chip equipped with a single shielding electrode and piezoelectric actuation is proposed. This design achieves high-sensitivity detection of the three-dimensional electric fields, simultaneously reducing excitation voltage and crosstalk noise is proposed. The sensing structure comprises one group of shielding electrodes and four sets of symmetrically distributed sensing electrodes. In response to piezoelectric actuation, the shielding electrodes undergo vertical vibrations, while the four sensing electrode sets generate induced currents when subjected to external electric fields. A differential decoupling method can be used to calculate the signals corresponding to the electric field components along the x-, y-, and z-axes. Finite element simulation was conducted to design the structure of the three-dimensional electric field sensing chip, analyze the feasibility of its measurement, and optimize key structural parameters. The fabrication process for the sensing chip was designed and implemented. Experimental results reveal that the output sensitivities are 0.2214 mV/(kV/m) for the x-axis, 0.3580 mV/(kV/m) for the y-axis, and 2.1768 mV/(kV/m) for the z-axis. The maximum measurement error for the three-dimensional electric fields remains < 5.3%.
2024, 46(4): 1521-1528.
doi: 10.11999/JEIT230601
Abstract:
Considering the widespread application of magnetic sensors in the Internet of Things (IoT), a Hall sensor readout interface circuit with low offset voltage and low-temperature drift characteristics based on a 180 nm CMOS process is designed in this work. In response to the temperature drift characteristic of the Hall sensor sensitivity, a temperature sensing circuit that is combined with the table lookup method to adjust the gain of the Programmable Gain Amplifier (PGA) is designed, which effectively reduces the Temperature Coefficient (TC) of the Hall sensor. On this basis, the offset voltage of the Hall sensor is greatly eliminated by the use of Correlated Double Sampling (CDS) technology in the main signal channel. The simulation results show that the TC of the Hall sensor is decreased from 966.4 ppm/°C to 58.1 ppm/°C in the temperature range of –40°C～125°C. The chip measurement results of the main signal channel show that the offset voltage of the Hall sensor is reduced from about 25 mV to about 4 mV and the nonlinear error of the Hall sensor is 0.50%, which occupies an active area of 0.69 mm2.
Considering the widespread application of magnetic sensors in the Internet of Things (IoT), a Hall sensor readout interface circuit with low offset voltage and low-temperature drift characteristics based on a 180 nm CMOS process is designed in this work. In response to the temperature drift characteristic of the Hall sensor sensitivity, a temperature sensing circuit that is combined with the table lookup method to adjust the gain of the Programmable Gain Amplifier (PGA) is designed, which effectively reduces the Temperature Coefficient (TC) of the Hall sensor. On this basis, the offset voltage of the Hall sensor is greatly eliminated by the use of Correlated Double Sampling (CDS) technology in the main signal channel. The simulation results show that the TC of the Hall sensor is decreased from 966.4 ppm/°C to 58.1 ppm/°C in the temperature range of –40°C～125°C. The chip measurement results of the main signal channel show that the offset voltage of the Hall sensor is reduced from about 25 mV to about 4 mV and the nonlinear error of the Hall sensor is 0.50%, which occupies an active area of 0.69 mm2.
