Current Issue

25 August 2023, Volume 44 Issue 8

Previous Issue    Next Issue

Papers

Adversarial sample generation algorithm for vertical federated learning

Xiaolin CHEN, Daoguang ZAN, Bingchao WU, Bei GUAN, Yongji WANG

2023, 44(8):  1-13.  doi:10.11959/j.issn.1000-436x.2023149

Asbtract ( 284 )   HTML ( 80)   PDF (2945KB) ( 439 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

To adapt to the scenario characteristics of vertical federated learning (VFL) applications regarding high communication cost, fast model iteration, and decentralized data storage, a generalized adversarial sample generation algorithm named VFL-GASG was proposed.Specifically, an adversarial sample generation framework was constructed for the VFL architecture.A white-box adversarial attack in the VFL was implemented by extending the centralized machine learning adversarial sample generation algorithm with different policies such as L-BFGS, FGSM, and C＆amp;W.By introducing deep convolutional generative adversarial network (DCGAN), an adversarial sample generation algorithm named VFL-GASG was designed to address the problem of universality in the generation of adversarial perturbations.Hidden layer vectors were utilized as local prior knowledge to train the adversarial perturbation generation model, and through a series of convolution-deconvolution network layers, finely crafted adversarial perturbations were produced.Experiments show that VFL-GASG can maintain a high attack success while achieving a higher generation efficiency, robustness, and generalization ability than the baseline algorithm, and further verify the impact of relevant settings for adversarial attacks.

Multi-camera video collaborative analysis method based on edge computing

Zhibo QI, Lei DU, Ru HUO, Fan YANG, Tao HUANG

2023, 44(8):  14-26.  doi:10.11959/j.issn.1000-436x.2023150

Asbtract ( 190 )   HTML ( 46)   PDF (4564KB) ( 390 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

In order to reduce the processing volume of multi-camera real-time video data in smart city scenarios, a video collaborative analysis method based on machine learning algorithms at the edge was proposed.Firstly, for the important objects detected by each camera, different key windows were designed to filter the region of interest (RoI) in the video, reduce the video data volume and extract its features.Then, based on the extracted data features, the same objects in the videos from different cameras were annotated, and a strategy for calculating the association degree value between cameras was designed for further reducing the video data volume.Finally, the GC-ReID algorithm based on graph convolutional network (GCN) and re-identification (ReID) was proposed, aiming at achieving the collaborative analysis of multi-camera videos.The experimental results show that proposed method can effectively reduce the system latency and improve the video compression rate while ensuring the high accuracy, compared with the existing video analysis methods.

Adaptive tensor train learning algorithm based on single-aspect streaming model

Baoze MA, Guojun LI, Long XING, Changrong YE

2023, 44(8):  27-36.  doi:10.11959/j.issn.1000-436x.2023154

Asbtract ( 85 )   HTML ( 17)   PDF (1914KB) ( 112 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

An adaptive tensor train (TT) learning algorithm for the online decomposition problem of high-order tensors in single-aspect streaming model was investigated.Firstly, it was deduced that single-aspect streaming increment only changes the dimension of temporal TT core.Secondly, the forgetting factor and regularization item were introduced to construct the objective function of exponentially weighted least-squares.Finally, the block-coordinate descent learning strategy was used to estimate the temporal and non-temporal TT core tensors respectively.Simulation results demonstrate that the proposed algorithm is validated in terms of increment size, TT-rank, noise and time-varying intensity, the average relative error and operation time are smaller than that of the comparison algorithms.The tensor slice reconstruction ability is superior than that of the comparison algorithms in the video adaptive analysis.

Communication-efficient distributed precoding design for Massive MIMO

Mian LI, Yang LI, Zonghui ZHANG, Qingjiang SHI

2023, 44(8):  37-48.  doi:10.11959/j.issn.1000-436x.2023147

Asbtract ( 172 )   HTML ( 27)   PDF (885KB) ( 205 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

A communication-efficient distributed precoding scheme was proposed for multi-baseband processing unit (BBU) baseband processing architecture, aiming to reduce fronthaul data exchange and computational complexity between BBUs.Firstly, a distributed framework based on R-WMMSE algorithm was proposed, which utilized the subspace property of the optimal solution to compress the interactive data losslessly, thereby reducing data exchange.Furthermore, two learnable compression modules based on matrix multiplication were designed, using optimized computing structures and matrix parameters to reduce the parameters and computations while maintaining function expressiveness.Finally, the learnable modules and the distributed precoding framework were jointly optimized with achievable rate as the optimization objective to obtain the final model.The proposed scheme can achieve guaranteed precoding performance under lower requirements on data interaction and computational complexity

Extensible hierarchical codec semantic communication system

Yuyuan ZHANG, Haitao ZHAO, Jibo WEI, Kuo CAO, Yichi ZHANG, Peng LUO, Yueling LIU, Kai MEI

2023, 44(8):  49-60.  doi:10.11959/j.issn.1000-436x.2023157

Asbtract ( 151 )   HTML ( 38)   PDF (4230KB) ( 336 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Aiming at the problem that most current researches on text semantic communication mainly rely on simulation system for theoretical verification, an extensible hierarchical semantic communication system was proposed by taking advantage of the separation of hierarchical encoding and decoding architecture in semantic level and grammatical level.The system was compatible with the reliable communication technology under the framework of Shannon information through the mode of semantic and syntactic separation, and realized the nested combination of semantic communication and traditional communication.Furthermore, a universal and extensible verification system was built based on the software radio platform to verify the proposed semantic communication system architecture.The verification system took semantic communication software platform as the driving core of hardware drive and algorithm call, integrated the whole process of signal generation, information transmission, data acquisition, decoding and evaluation at the receiving end, and could be further extended for semantic and syntactic level.Finally, the text semantic communication was tested based on this verification system, which verified that it had higher validity and reliability than the traditional communication mode.

Efficient certificateless searchable encryption scheme with verifiability

Xinhua CUI, Youliang TIAN, Qijia ZHANG

2023, 44(8):  61-77.  doi:10.11959/j.issn.1000-436x.2023156

Asbtract ( 122 )   HTML ( 28)   PDF (1765KB) ( 263 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Searchable encryption offers an effective way to achieve data privacy protection and keyword search in cloud computing environments.Currently, the existing schemes not only lack dynamic update and efficient verification mechanism, but also suffer from the certificate management burden and key escrow issue.To address these issues, a verifiable multi-keyword searchable encryption scheme based on improved Merkle-Tree had been proposed recently.However, through cryptoanalysis, that scheme could not achieve the indistinguishability.With improvement, an efficient able certificateless searchable encryption scheme with verifiability was proposed.Rigorous analysis show that the proposed scheme not only supports the indistinguishability and the unforgeability, but also enjoys higher computing efficiency and lower communication cost, which is more suitable for terminal devices with limited resources.

Design of beamforming for IRS multi-partition-aided THz multi-subarray

Zufan ZHANG, Rui TANG

2023, 44(8):  78-88.  doi:10.11959/j.issn.1000-436x.2023158

Asbtract ( 80 )   HTML ( 16)   PDF (1674KB) ( 255 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

In the IRS-aided THz communication system, in order to break the channel sparsity limitation on the system spatial multiplexing gain, a hybrid beam forming architecture with wide-spaced multi-subarrays at the transceiver/transmitter side was proposed and the transmission scheme of IRS multi-partition-aided THz multi-subarrays was designed.Firstly, a non-convex objective function with multivariate coupling and non-convex constraints was constructed based on the principle of maximizing spectrum efficiency.Then, the optimization problem was decoupled into two easy-to-solve sub-problems, namely, the reflection coefficient matrix design of IRS and the hybrid beamforming matrix design at the transceiver/receiver.Finally, the Riemannian manifold optimization algorithm was used to calculate the reflection coefficient matrix of IRS, and the closed solution of the hybrid beamforming matrix design at the transceiver/receiver was obtained through mathematical derivation.Simulation results show that compared with the baseline scheme, the proposed scheme can achieve better spectrum efficiency.

Speaker verification method based on cross-domain attentive feature fusion

Zhen YANG, Tianlang WANG, Haiyan GUO, Tingting WANG

2023, 44(8):  89-98.  doi:10.11959/j.issn.1000-436x.2023142

Asbtract ( 141 )   HTML ( 23)   PDF (2338KB) ( 164 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Aiming at the problem that the lack of structure information among speech signal sample in the front-end acoustic features of speaker verification system, a speaker verification method based on cross-domain attentive feature fusion was proposed.Firstly, a feature extraction method based on the graph signal processing (GSP) was proposed to extract the structural information of speech signals, each sample point in a speech signal frame was regarded as a graph node to construct the speech graph signal and the graph frequency information of the speech signal was extracted through the graph Fourier transform and filter banks.Then, an attentive feature fusion network with the residual neural network and the squeeze-and- excitation block was proposed to fuse the features in the traditional time-frequency domain and those in the graph frequency domain to promote the speaker verification system performance.Finally, the experiment was carried out on the VoxCeleb, SITW, and CN-Celeb datasets.The experimental results show that the proposed method performs better than the baseline ECAPA-TDNN model in terms of equal error rate (EER) and minimum detection cost function (min-DCF).

NLOS location enhancement algorithm based on depth-first multipath parameter estimation

Xiaofeng LU, Ye DONG, Yuejie LI

2023, 44(8):  99-110.  doi:10.11959/j.issn.1000-436x.2023144

Asbtract ( 113 )   HTML ( 29)   PDF (2324KB) ( 174 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

In order to improve the positioning accuracy of millimeter wave system with non-line-of-sight (NLOS) paths, a depth-first multipath parameter estimation algorithm was proposed based on distributed compressed sensing.According to the evaluated multipath parameters, NLOS path could be identified, so that the localization performance was enhanced.Firstly, depth-first algorithm was applied to reduce the unnecessary path searching, and the more accurate multipath parameters were obtained.Secondly, under the reverse positioning distance residual method, NLOS path recognition could be carried out.Then, the scatterers in the NLOS path were matched, and the position of which were regarded as virtual anchor nodes.Combining the information of base stations and virtual anchor nodes, positioning enhancement was realized.Finally, localization performance of the proposed algorithm was simulated, compared with the distance weighted least square (LS) and maximum discrimination transformation (MDT) algorithms, the performance of the proposed algorithm is improved by 17% and 8% respectively.

Capacity analysis for MISO-UWOC systems over GGD weak turbulence with zero boresight pointing error

Yueheng LI, Yining XU, Meiyan JU, Ping HUANG

2023, 44(8):  111-124.  doi:10.11959/j.issn.1000-436x.2023152

Asbtract ( 33 )   HTML ( 2)   PDF (1943KB) ( 85 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Generalized Gamma distribution (GGD) was chosen to describe the weak oceanic turbulence, and a new hybrid fading channel model that integrated the GGD weak turbulence, the zero boresight pointing error, and the implicit path loss plus multipath propagation characterized by the fading free impulse response (FFIR) was proposed.Subsequently, mathematical expressions for the ergodic capacity and outage capacity of the multiple-input single-output underwater wireless optical communication (MISO-UWOC) systems were derived through the Meijer-G function under a selective transmission (ST) diversity scheme especially while inter-symbol interference (ISI) effects were considered or not.Finally, the correctness of the theoretical formulas derived above was verified by some numerical results.The simulation results show that with the introduction of the ST diversity, the ergodic capacity of the MISO-UWOC systems, taking the transmission ports N=2 as an example, is at least 1.3 times better than that of the conventional point-to-point (P2P) transmission under the same channel condition and system parameters, while the maximum outage capacity decreases is also more than 60% compared with the conventional P2P one.However, the introduction of the ISI will severely reduce this performance improvement.

Resource allocation strategy of low earth orbit satellite oriented to user transmission difference

Fatang CHEN, Miao HUANG, Yufeng JIN

2023, 44(8):  125-133.  doi:10.11959/j.issn.1000-436x.2023153

Asbtract ( 62 )   HTML ( 14)   PDF (1336KB) ( 170 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

In the low earth orbit (LEO) uplink communication scenario, aiming at the different needs of different users for high capacity transmission (HCT) and high reliability transmission (HRT), the spectrum of two users with different needs was shared, a resource allocation model for maximizing the total capacity of HCT users under the constraint of HRT users was established, and the user power and channel resources were optimized.Based on the statistical characteristics of channel fading, the power allocation was carried out to deal with the uncertainty challenge caused by the randomness of channel fading.On the basis of power allocation, the Hungarian algorithm was used to pair users to share the same channel.In order to improve the fairness of HCT users, the maximization of minimum ergodic capacity was included in the optimization goal, and a balance matrix was introduced to solve the problem based on the existing algorithms.Simulation results show that the total capacity of HCT users of the proposed algorithm is higher than that of other algorithms under the same HRT user outage probability, and the improved algorithm also has a significant effect on improving the fairness and robustness of HCT users.

Time-slot allocation algorithm for LEO satellite beam hopping based on non-stationary MAB

Min LIN, Pengcheng KAN, Bai ZHAO, Ming CHENG, Lyuxi YANG

2023, 44(8):  134-143.  doi:10.11959/j.issn.1000-436x.2023160

Asbtract ( 96 )   HTML ( 23)   PDF (2933KB) ( 196 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Aiming at the problem of inadequate adaptation to dynamic changes in cell services in the resource allocation algorithm for hopping beams in low earth orbit (LEO) satellite systems, a time-slot allocation algorithm for LEO satellite beam hopping based on a non-stationary multi-armed bandit (MAB) was proposed.Firstly, the joint optimization problem of slot allocation and beam grading matching was established, with the minimization of the system’s second-order differential capacity as the optimization objective.Secondly, due to the non-convexity and difficulty in direct solution of this problem, a beam grading combination scheme generation algorithm was proposed based on the concept of effective cells and effective critical cells, which generated all possible beam grading combination schemes.Next, a dynamic slot allocation scheme based on the non-stationary MAB model was proposed, and joint optimization of slot allocation and beam grading matching was completed under the optimal beam grading combination scheme.Finally, the computer simulation results show that the average redundancy of the proposed algorithm is less than 20% in the case of multiple cell service distributions.In addition, compared with other schemes, the proposed algorithm can control the average beam revisit time to about 300 ms while maintaining high system throughput.

Secure and efficient group handover authentication protocol based on trajectory prediction in 5G-V2X

Yinghui ZHANG, Jiale QIAN, Jin CAO, Dong ZHENG

2023, 44(8):  144-154.  doi:10.11959/j.issn.1000-436x.2023136

Asbtract ( 109 )   HTML ( 18)   PDF (1482KB) ( 209 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Aiming at the efficiency issue of handover authentication for a large number of vehicles in the 5G-V2X scenario, a secure and efficient group handover authentication protocol based on trajectory prediction was proposed.Firstly, the effect of completing key agreement protocol in advance was achieved by predicting vehicle trajectories.Secondly, vehicles with mobility relevance were treated as the same group through user grouping algorithms, and then all vehicles within the group were batch verified using certificateless aggregation signature technology.In addition, to address the vulnerability of aggregated signature technology to DoS attacks, a binary search method was used to quickly locate malicious users and improve the efficiency of group handover authentication protocol.Finally, the security analysis of the protocol was conducted using the formal verification tool Scyther, and compared with the existing optimal protocol, the computational efficiency is improved by 30%.

Data poisoning attack detection approach for quality of service aware cloud API recommender system

Zhen CHEN, Wenchao QI, Taiyu BAO, Limin SHEN

2023, 44(8):  155-167.  doi:10.11959/j.issn.1000-436x.2023161

Asbtract ( 109 )   HTML ( 16)   PDF (5439KB) ( 205 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

To solve the problem that existing studies usually assumed that the QoS data of cloud API recommender system was reliable, ignoring the data poisoning attack on cloud API recommender system by malicious users in open network environment, a data poisoning attack detection approach based on multi-feature fusion was proposed.Firstly, a user connected network graph was constructed based on the designed similarity function, and users’ neighborhood features were captured using Node2vec.Secondly, sparse auto-encoder was used to mine user QoS deep feature, and user interpretation feature based on QoS data weighted average deviation was designed.Furthermore, a fake user detection model based on support vector machine was established by integrating user neighborhood feature, QoS deep feature, and interpretation feature, the model parameters were learned using grid search and alternating iterative optimization strategy to complete fake user detection.Finally, the effectiveness and superiority of the proposed approach were verified through extensive experiments, realizing the poison attack defense against QoS aware cloud API recommender system at the data side.

AEUR: authenticated encryption algorithm design based on uBlock round function

Yatao YANG, Hui DONG, Jiantao LIU, Yanshuo ZHANG

2023, 44(8):  168-178.  doi:10.11959/j.issn.1000-436x.2023159

Asbtract ( 107 )   HTML ( 18)   PDF (852KB) ( 207 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

In order to improve the efficiency of the implementation of the authenticated encryption algorithm without compromising the security of the algorithm, a new authenticated encryption algorithm AEUR was designed.Firstly, based on the uBlock round function, with resistance to internal collision attacks as the security objective, a mixed integer linear programming approach was used to search for generic iterative component R(t,s) to meet the security objective.Secondly, the authenticated encryption algorithm AEUR was designed by using this component.AEUR consisted of two parts: authenticated encryption and decrypted verification, both of which performed the same process without the need to design additional operational sessions, reducing the algorithm’s resource consumption.In addition, the correctness of the algorithm was verified by comparing the corresponding round state values, and the security of the algorithm was analyzed using various analysis methods such as linear attacks and sliding attacks.Finally, the algorithm was implemented in C language to prove the AEUR has good performance.The results show that the proposed algorithm has a better overall performance in terms of software runtime, with efficiency improvements of 3% and 46% compared to AEGIS and ALE, and 74% and 92% compared to AES-GCM and ACORN, respectively.

Comprehensive Reviews

Overview of research progress on blind separation methods for single channel communication signal

Wen DENG, Zhitao HUANG, Xiang WANG

2023, 44(8):  179-194.  doi:10.11959/j.issn.1000-436x.2023138

Asbtract ( 208 )   HTML ( 45)   PDF (1670KB) ( 406 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Significant progress has been made in the research of single channel blind signal separation (SCBSS) methods, as the theoretical and practical applications of SCBSS technology have been continuously improved.A new classification framework for blind signal separation (BSS) of communication signal problems was proposed based on an analysis of a large number of academic research results from both domestic and international sources.The division of SCBSS problems into four levels: multi-signal waveform separation, single-signal waveform separation, multi-signal symbol sequence separation, and single-signal symbol sequence separation was determined by taking into consideration the demand for output results of BSS systems in different application scenarios.Subsequently, the research status of SCBSS methods was systematically reviewed.Furthermore, the current research status of data-driven SCBSS was summarized, and the problems and solutions that needed to be addressed in data-driven SCBSS technology were explored.Finally, several potential research directions for SCBSS were analyzed and forecasted.The aim is to provide a reference for the research and application of SCBSS.

Correspondences

Resource allocation strategy based on deep reinforcement learning in 6G dense network

Fan YANG, Cheng YANG, Jie HUANG, Shilong ZHANG, Tao YU, Xun ZUO, Chuan YANG

2023, 44(8):  215-227.  doi:10.11959/j.issn.1000-436x.2023148

Asbtract ( 203 )   HTML ( 30)   PDF (3887KB) ( 293 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

In order to realize no overlapping interference between cells, 6G dense network (DN) adopting resource allocation is the important technology of enhancing network performance.However, limited resources and dense distribution of nodes make it difficult to solve the problem of resource allocation through traditional optimization methods.To tackle the problem, a point-line graph coloring based overlapping interference model was formulated and a Dueling deep Q-network (DQN) based resource allocation method was proposed, which combined deep reinforcement learning (DRL) and the overlapping interference model.Specifically, the proposed method adopted the overlapping interference model and resource reuse rate to design the immediate reward.Then, generating 6G DN resource allocation strategies were independently learned by using Dueling DQN to achieve the goal of realizing resource allocation without overlapping interference between cells.The performance evaluation results show that the proposed method can effectively increase both network throughput and resource reuse rate, as well as enhance network performance.

Parallel deep forest algorithm based on Spark and three-way interactive information

Yimin MAO, Zhan ZHOU, Zhigang CHEN

2023, 44(8):  228-240.  doi:10.11959/j.issn.1000-436x.2023143

Asbtract ( 67 )   HTML ( 10)   PDF (2260KB) ( 79 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

To address issues such as excessive redundancy and irrelevant features, long class vectors, slow model convergence, and low efficiency of parallel training in parallel deep forests, a parallel deep forest algorithm based on Spark and three-way interactive information was proposed.Firstly, a feature selection based on feature interaction (FSFI) strategy was proposed to filter the original features and eliminate irrelevant and redundant features.Secondly, a multi-granularity vector elimination (MGVE) strategy was proposed, which fused similar class vectors and shortened the class vector length.Subsequently, the cascade forest feature enhancement (CFFE) strategy was proposed to improve the utilization of information and accelerate the convergence speed of the model.Finally, a multi-level load balancing (MLB) strategy was proposed, combined with the Spark framework, to improve the parallelization efficiency through adaptive sub-forest division and heterogeneous skew data partitioning.Experimental results demonstrate that the proposed algorithm significantly improves the model classification effect and reduces the parallelization training time.

Novel video anomaly detection method based on global-local self-attention network

Jing YANG, Chengmao WU, Liuping ZHOU

2023, 44(8):  241-250.  doi:10.11959/j.issn.1000-436x.2023151

Asbtract ( 165 )   HTML ( 37)   PDF (1346KB) ( 270 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

In order to improve the accuracy of video anomaly detection, a novel video anomaly detection method based on global-local self-attention network was proposed.Firstly, the video sequence and the corresponding RGB sequence were fused to highlight the motion change of the object.Secondly, the temporal correlation of the video sequence in the local area was captured by the expansion convolution layer, along with the self-attention network was utilized to compute the global temporal dependencies of the video sequence.Meanwhile, by deepening the basic network U-Net and combining the relevant motion and representation constraints, the network model was trained end-to-end to improve the detection accuracy and robustness of the model.Finally, experiments were carried out on the public data sets UCSD Ped2, CUHK Avenue and ShanghaiTech, as well as the test results were visually analyzed.The experimental results show that the detection accuracy AUC of the proposed method reaches 97.4%, 86.8% and 73.2% respectively, which is obviously better than that of the compared methods.