Pub Date : 2024-08-28DOI: 10.1109/OJSP.2024.3451167
Thomas Kropfreiter;Florian Meyer;David F. Crouse;Stefano Coraluppi;Franz Hlawatsch;Peter Willett
Joint probabilistic data association (JPDA) filter methods and multiple hypothesis tracking (MHT) methods are widely used for multitarget tracking (MTT). However, they are known to exhibit undesirable behavior in tracking scenarios with targets in close proximity: JPDA filter methods suffer from the track coalescence effect, i.e., the estimated tracks of targets in close proximity tend to merge and can become indistinguishable, while MHT methods suffer from an opposite effect known as track repulsion, i.e., the estimated tracks of targets in close proximity tend to repel each other in the sense that their separation is larger than the actual distance between the targets. In this paper, we review the JPDA filter and MHT methods and discuss the track coalescence and track repulsion effects. We also consider a more recent methodology for MTT that is based on the belief propagation (BP) algorithm. We argue that BP-based MTT does not exhibit track repulsion because it is not based on maximum a posteriori estimation, and that it exhibits significantly reduced track coalescence because certain properties of the BP messages related to data association encourage separation of target state estimates. Our theoretical arguments are confirmed by numerical results for four representative simulation scenarios.
联合概率数据关联（JPDA）滤波方法和多重假设跟踪（MHT）方法被广泛用于多目标跟踪（MTT）。然而，众所周知，在目标距离很近的跟踪场景中，它们会表现出不理想的行为：JPDA 滤波方法会受到轨迹凝聚效应的影响，即近距离目标的估计轨迹趋于合并，变得难以区分；而 MHT 方法则会受到称为轨迹排斥的相反效应的影响，即近距离目标的估计轨迹趋于相互排斥，它们之间的距离大于目标之间的实际距离。本文回顾了 JPDA 滤波和 MHT 方法，并讨论了轨迹凝聚和轨迹排斥效应。我们还考虑了一种基于信念传播（BP）算法的最新 MTT 方法。我们认为，基于 BP 算法的 MTT 不会出现轨迹排斥现象，因为它不是基于最大后验估计，而且由于与数据关联相关的 BP 信息的某些特性会鼓励目标状态估计的分离，因此它能显著减少轨迹凝聚。我们的理论论点得到了四种代表性模拟场景的数值结果的证实。
{"title":"Track Coalescence and Repulsion in Multitarget Tracking: An Analysis of MHT, JPDA, and Belief Propagation Methods","authors":"Thomas Kropfreiter;Florian Meyer;David F. Crouse;Stefano Coraluppi;Franz Hlawatsch;Peter Willett","doi":"10.1109/OJSP.2024.3451167","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3451167","url":null,"abstract":"Joint probabilistic data association (JPDA) filter methods and multiple hypothesis tracking (MHT) methods are widely used for multitarget tracking (MTT). However, they are known to exhibit undesirable behavior in tracking scenarios with targets in close proximity: JPDA filter methods suffer from the track coalescence effect, i.e., the estimated tracks of targets in close proximity tend to merge and can become indistinguishable, while MHT methods suffer from an opposite effect known as track repulsion, i.e., the estimated tracks of targets in close proximity tend to repel each other in the sense that their separation is larger than the actual distance between the targets. In this paper, we review the JPDA filter and MHT methods and discuss the track coalescence and track repulsion effects. We also consider a more recent methodology for MTT that is based on the belief propagation (BP) algorithm. We argue that BP-based MTT does not exhibit track repulsion because it is not based on maximum a posteriori estimation, and that it exhibits significantly reduced track coalescence because certain properties of the BP messages related to data association encourage separation of target state estimates. Our theoretical arguments are confirmed by numerical results for four representative simulation scenarios.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"1089-1106"},"PeriodicalIF":2.9,"publicationDate":"2024-08-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10654568","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142595863","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The affinity propagation clustering (APC) algorithm is popular for fingerprint database clustering because it can cluster without pre-defining the number of clusters. However, the clustering performance of the APC algorithm heavily depends on the chosen fingerprint similarity metric, with distance-based metrics being the most commonly used. Despite its popularity, the APC algorithm lacks comprehensive research on how distance-based metrics affect clustering performance. This emphasizes the need for a better understanding of how these metrics influence its clustering performance, particularly in fingerprint databases. This paper investigates the impact of various distance-based fingerprint similarity metrics on the clustering performance of the APC algorithm. It identifies the best fingerprint similarity metric for optimal clustering performance for a given fingerprint database. The analysis is conducted across five experimentally generated online fingerprint databases, utilizing seven distance-based metrics: Euclidean, squared Euclidean, Manhattan, Spearman, cosine, Canberra, and Chebyshev distances. Using the silhouette score as the performance metric, the simulation results indicate that structural characteristics of the fingerprint database, such as the distribution of fingerprint vectors, play a key role in selecting the best fingerprint similarity metric. However, Euclidean and Manhattan distances are generally the preferable choices for use as fingerprint similarity metrics for the APC algorithm across most fingerprint databases, regardless of their structural characteristics. It is recommended that other factors, such as computational intensity and the presence or absence of outliers, be considered alongside the structural characteristics of the fingerprint database when choosing the appropriate fingerprint similarity metric for maximum clustering performance.
{"title":"Impact of Varying Distance-Based Fingerprint Similarity Metrics on Affinity Propagation Clustering Performance in Received Signal Strength-Based Fingerprint Databases","authors":"Abdulmalik Shehu Yaro;Filip Maly;Karel Maly;Pavel Prazak","doi":"10.1109/OJSP.2024.3449816","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3449816","url":null,"abstract":"The affinity propagation clustering (APC) algorithm is popular for fingerprint database clustering because it can cluster without pre-defining the number of clusters. However, the clustering performance of the APC algorithm heavily depends on the chosen fingerprint similarity metric, with distance-based metrics being the most commonly used. Despite its popularity, the APC algorithm lacks comprehensive research on how distance-based metrics affect clustering performance. This emphasizes the need for a better understanding of how these metrics influence its clustering performance, particularly in fingerprint databases. This paper investigates the impact of various distance-based fingerprint similarity metrics on the clustering performance of the APC algorithm. It identifies the best fingerprint similarity metric for optimal clustering performance for a given fingerprint database. The analysis is conducted across five experimentally generated online fingerprint databases, utilizing seven distance-based metrics: Euclidean, squared Euclidean, Manhattan, Spearman, cosine, Canberra, and Chebyshev distances. Using the silhouette score as the performance metric, the simulation results indicate that structural characteristics of the fingerprint database, such as the distribution of fingerprint vectors, play a key role in selecting the best fingerprint similarity metric. However, Euclidean and Manhattan distances are generally the preferable choices for use as fingerprint similarity metrics for the APC algorithm across most fingerprint databases, regardless of their structural characteristics. It is recommended that other factors, such as computational intensity and the presence or absence of outliers, be considered alongside the structural characteristics of the fingerprint database when choosing the appropriate fingerprint similarity metric for maximum clustering performance.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"1005-1014"},"PeriodicalIF":2.9,"publicationDate":"2024-08-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10646489","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142276495","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-15DOI: 10.1109/OJSP.2024.3444717
Simon Anuk;Tamir Bendory;Amichai Painsky
This paper studies the optimal solution of the classical problem of detecting the location of multiple image occurrences in a two-dimensional, noisy measurement. Assuming the image occurrences do not overlap, we formulate this task as a constrained maximum likelihood optimization problem. We show that the maximum likelihood estimator is equivalent to an instance of the winner determination problem from the field of combinatorial auction and that the solution can be obtained by searching over a binary tree. We then design a pruning mechanism that significantly accelerates the runtime of the search. We demonstrate on simulations and electron microscopy data sets that the proposed algorithm provides accurate detection in challenging regimes of high noise levels and densely packed image occurrences.
{"title":"Image Detection Using Combinatorial Auction","authors":"Simon Anuk;Tamir Bendory;Amichai Painsky","doi":"10.1109/OJSP.2024.3444717","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3444717","url":null,"abstract":"This paper studies the optimal solution of the classical problem of detecting the location of multiple image occurrences in a two-dimensional, noisy measurement. Assuming the image occurrences do not overlap, we formulate this task as a constrained maximum likelihood optimization problem. We show that the maximum likelihood estimator is equivalent to an instance of the winner determination problem from the field of combinatorial auction and that the solution can be obtained by searching over a binary tree. We then design a pruning mechanism that significantly accelerates the runtime of the search. We demonstrate on simulations and electron microscopy data sets that the proposed algorithm provides accurate detection in challenging regimes of high noise levels and densely packed image occurrences.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"1015-1022"},"PeriodicalIF":2.9,"publicationDate":"2024-08-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10637697","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142320494","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-15DOI: 10.1109/OJSP.2024.3444707
Duc Thien Nguyen;Konstantinos Slavakis
This paper introduces a novel kernel regression framework for data imputation, coined multilinear kernel regression and imputation via the manifold assumption (MultiL-KRIM). Motivated by manifold learning, MultiL-KRIM models data features as a point-cloud located in or close to a user-unknown smooth manifold embedded in a reproducing kernel Hilbert space. Unlike typical manifold-learning routes, which seek low-dimensional patterns via regularizers based on graph-Laplacian matrices, MultiL-KRIM builds instead on the intuitive concept of tangent spaces to manifolds and incorporates collaboration among point-cloud neighbors (regressors) directly into the data-modeling term of the loss function. Multiple kernel functions are allowed to offer robustness and rich approximation properties, while multiple matrix factors offer low-rank modeling, dimensionality reduction and streamlined computations, with no need of training data. Two important application domains showcase the functionality of MultiL-KRIM: time-varying-graph-signal (TVGS) recovery, and reconstruction of highly accelerated dynamic-magnetic-resonance-imaging (dMRI) data. Extensive numerical tests on real TVGS and synthetic dMRI data demonstrate that the “shallow” MultiL-KRIM offers remarkable speedups over its predecessors and outperforms other “shallow” state-of-the-art techniques, with a more intuitive and explainable pipeline than deep-image-prior methods.
{"title":"Multilinear Kernel Regression and Imputation via Manifold Learning","authors":"Duc Thien Nguyen;Konstantinos Slavakis","doi":"10.1109/OJSP.2024.3444707","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3444707","url":null,"abstract":"This paper introduces a novel kernel regression framework for data imputation, coined multilinear kernel regression and imputation via the manifold assumption (MultiL-KRIM). Motivated by manifold learning, MultiL-KRIM models data features as a point-cloud located in or close to a user-unknown smooth manifold embedded in a reproducing kernel Hilbert space. Unlike typical manifold-learning routes, which seek low-dimensional patterns via regularizers based on graph-Laplacian matrices, MultiL-KRIM builds instead on the intuitive concept of tangent spaces to manifolds and incorporates collaboration among point-cloud neighbors (regressors) directly into the data-modeling term of the loss function. Multiple kernel functions are allowed to offer robustness and rich approximation properties, while multiple matrix factors offer low-rank modeling, dimensionality reduction and streamlined computations, with no need of training data. Two important application domains showcase the functionality of MultiL-KRIM: time-varying-graph-signal (TVGS) recovery, and reconstruction of highly accelerated dynamic-magnetic-resonance-imaging (dMRI) data. Extensive numerical tests on real TVGS and synthetic dMRI data demonstrate that the “shallow” MultiL-KRIM offers remarkable speedups over its predecessors and outperforms other “shallow” state-of-the-art techniques, with a more intuitive and explainable pipeline than deep-image-prior methods.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"1073-1088"},"PeriodicalIF":2.9,"publicationDate":"2024-08-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10637724","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142587638","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-15DOI: 10.1109/OJSP.2024.3444709
Ahmed Murtada;Bhavani Shankar Mysore Rama Rao;Moein Ahmadi;Udo Schroeder
The emergence of Multiple-Input Multiple-Output (MIMO) millimeter-wave (mmWave) radar sensors has prompted interest in indoor sensing applications, including human detection, vital signs monitoring, and real-time tracking in crowded environments. These sensors, equipped with multiple antenna elements, offer high angular resolution, often referred to as imaging radars for their capability to detect high-resolution point clouds. Employing radar systems with high-angular resolution in occlusion-prone scenarios often results in sparse signal returns in range profiles. In extreme cases, only one target return may be observed, as the resolution grid size becomes significantly smaller than the targets, causing portions of the targets to consistently occupy the full area of a test cell. Leveraging this structure, we propose two detectors to enhance the detection of non-occluded targets in such scenarios, thereby providing accurate high-resolution point clouds. The first method employs multiple hypothesis testing over each range profile where the range cells within are considered mutually occluding. The second is formulated based on binary hypothesis testing for each cell, considering the distribution of the signal in the other cells within the same range profile. Numerical analysis demonstrates the superior performance of the latter method over both the classic detection and the former method, especially in low Signal-to-Noise Ratio (SNR) scenarios. Our work showcases the potential of occlusion-informed detection in imaging radars to improve the detection probability of non-occluded targets and reduce false alarms in challenging indoor environments.
{"title":"Occlusion-Informed Radar Detection for Millimeter-Wave Indoor Sensing","authors":"Ahmed Murtada;Bhavani Shankar Mysore Rama Rao;Moein Ahmadi;Udo Schroeder","doi":"10.1109/OJSP.2024.3444709","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3444709","url":null,"abstract":"The emergence of Multiple-Input Multiple-Output (MIMO) millimeter-wave (mmWave) radar sensors has prompted interest in indoor sensing applications, including human detection, vital signs monitoring, and real-time tracking in crowded environments. These sensors, equipped with multiple antenna elements, offer high angular resolution, often referred to as imaging radars for their capability to detect high-resolution point clouds. Employing radar systems with high-angular resolution in occlusion-prone scenarios often results in sparse signal returns in range profiles. In extreme cases, only one target return may be observed, as the resolution grid size becomes significantly smaller than the targets, causing portions of the targets to consistently occupy the full area of a test cell. Leveraging this structure, we propose two detectors to enhance the detection of non-occluded targets in such scenarios, thereby providing accurate high-resolution point clouds. The first method employs multiple hypothesis testing over each range profile where the range cells within are considered mutually occluding. The second is formulated based on binary hypothesis testing for each cell, considering the distribution of the signal in the other cells within the same range profile. Numerical analysis demonstrates the superior performance of the latter method over both the classic detection and the former method, especially in low Signal-to-Noise Ratio (SNR) scenarios. Our work showcases the potential of occlusion-informed detection in imaging radars to improve the detection probability of non-occluded targets and reduce false alarms in challenging indoor environments.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"976-990"},"PeriodicalIF":2.9,"publicationDate":"2024-08-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10637692","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142246512","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-14DOI: 10.1109/OJSP.2023.3347994
{"title":"IEEE Signal Processing Society Information","authors":"","doi":"10.1109/OJSP.2023.3347994","DOIUrl":"https://doi.org/10.1109/OJSP.2023.3347994","url":null,"abstract":"","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"C2-C2"},"PeriodicalIF":2.9,"publicationDate":"2024-08-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10636085","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141985880","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-07-30DOI: 10.1109/OJSP.2024.3435673
Ting Yu;Kele Xu;Xu Wang;Bo Ding;Dawei Feng
In recent years, the use of time series analysis has become widespread, prompting researchers to explore methods to improve classification. Time series self-supervised learning has emerged as a significant area of study, aiming to uncover patterns in unlabeled data for richer information. Contrastive self-supervised learning, particularly, has gained attention for time series classification. However, it introduces inductive bias by generating positive and negative samples. Another approach involves Masked Autoencoders (MAE), which are effective for various data types. However, due to their reliance on the Transformer architecture, they demand significant computational resources during the pre-training phase. Recently, inspired by the remarkable advancements achieved by convolutional networks in the domain of time series forecasting, we aspire to employ convolutional networks utilizing a strategy of mask recovery for pre-training time series models. This study introduces a novel model termed Hierarchical Sparse Convolutional Masked-Autoencoder, “HSC-MAE”, which seamlessly integrates convolutional operations with the MAE architecture to adeptly capture time series features across varying scales. Furthermore, the HSC-MAE model incorporates dedicated decoders that amalgamate global and local information, enhancing its capacity to comprehend intricate temporal patterns. To gauge the effectiveness of the proposed approach, an extensive array of experiments was conducted across nine distinct datasets. The experimental outcomes stand as a testament to the efficacy of HSC-MAE in effectively mitigating the aforementioned challenges.
{"title":"Self-Supervised Learning-Based Time Series Classification via Hierarchical Sparse Convolutional Masked-Autoencoder","authors":"Ting Yu;Kele Xu;Xu Wang;Bo Ding;Dawei Feng","doi":"10.1109/OJSP.2024.3435673","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3435673","url":null,"abstract":"In recent years, the use of time series analysis has become widespread, prompting researchers to explore methods to improve classification. Time series self-supervised learning has emerged as a significant area of study, aiming to uncover patterns in unlabeled data for richer information. Contrastive self-supervised learning, particularly, has gained attention for time series classification. However, it introduces inductive bias by generating positive and negative samples. Another approach involves Masked Autoencoders (MAE), which are effective for various data types. However, due to their reliance on the Transformer architecture, they demand significant computational resources during the pre-training phase. Recently, inspired by the remarkable advancements achieved by convolutional networks in the domain of time series forecasting, we aspire to employ convolutional networks utilizing a strategy of mask recovery for pre-training time series models. This study introduces a novel model termed Hierarchical Sparse Convolutional Masked-Autoencoder, “HSC-MAE”, which seamlessly integrates convolutional operations with the MAE architecture to adeptly capture time series features across varying scales. Furthermore, the HSC-MAE model incorporates dedicated decoders that amalgamate global and local information, enhancing its capacity to comprehend intricate temporal patterns. To gauge the effectiveness of the proposed approach, an extensive array of experiments was conducted across nine distinct datasets. The experimental outcomes stand as a testament to the efficacy of HSC-MAE in effectively mitigating the aforementioned challenges.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"964-975"},"PeriodicalIF":2.9,"publicationDate":"2024-07-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10614789","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142099791","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-07-30DOI: 10.1109/OJSP.2024.3435703
Fengyang Gu;Luxin Zhang;Shilian Zheng;Jie Chen;Keqiang Yue;Zhijin Zhao;Xiaoniu Yang
Radar is widely used in aviation, meteorology, and military fields, and radar pulse signal detection has become an indispensable and essential function of cognitive radio systems as well as electronic warfare systems. In this paper, we propose a deep learning-based radar signal detection method. Firstly, we propose a detection method based on raw in-phase and quadrature (IQ) input, which utilizes a convolutional neural network (CNN) to automatically learn the features of radar pulse signals and noises, to accomplish the detection task. To further reduce the computational complexity, we also propose a hybrid detection method that combines compressed sensing (CS) and deep learning, which reduces the length of the signal by compressed downsampling, and then feeds the compressed signal to the CNN for detection. Extensive simulation results show that our proposed IQ-based method outperforms the traditional short-time Fourier transform method as well as three existing deep learning-based detection methods in terms of probability of detection. Furthermore, our proposed IQ-CS-based method can achieve satisfactory detection performance with significantly reduced computational complexity.
雷达广泛应用于航空、气象和军事领域，雷达脉冲信号检测已成为认知无线电系统和电子战系统不可或缺的重要功能。本文提出了一种基于深度学习的雷达信号检测方法。首先，我们提出了一种基于原始同相和正交（IQ）输入的检测方法，利用卷积神经网络（CNN）自动学习雷达脉冲信号和噪声的特征，从而完成检测任务。为了进一步降低计算复杂度，我们还提出了一种结合压缩传感（CS）和深度学习的混合检测方法，即通过压缩降采样减少信号长度，然后将压缩后的信号输入 CNN 进行检测。广泛的仿真结果表明，我们提出的基于 IQ 的方法在检测概率方面优于传统的短时傅立叶变换方法以及现有的三种基于深度学习的检测方法。此外，我们提出的基于 IQ-CS 的方法可以在显著降低计算复杂度的情况下实现令人满意的检测性能。
{"title":"Detection of Radar Pulse Signals Based on Deep Learning","authors":"Fengyang Gu;Luxin Zhang;Shilian Zheng;Jie Chen;Keqiang Yue;Zhijin Zhao;Xiaoniu Yang","doi":"10.1109/OJSP.2024.3435703","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3435703","url":null,"abstract":"Radar is widely used in aviation, meteorology, and military fields, and radar pulse signal detection has become an indispensable and essential function of cognitive radio systems as well as electronic warfare systems. In this paper, we propose a deep learning-based radar signal detection method. Firstly, we propose a detection method based on raw in-phase and quadrature (IQ) input, which utilizes a convolutional neural network (CNN) to automatically learn the features of radar pulse signals and noises, to accomplish the detection task. To further reduce the computational complexity, we also propose a hybrid detection method that combines compressed sensing (CS) and deep learning, which reduces the length of the signal by compressed downsampling, and then feeds the compressed signal to the CNN for detection. Extensive simulation results show that our proposed IQ-based method outperforms the traditional short-time Fourier transform method as well as three existing deep learning-based detection methods in terms of probability of detection. Furthermore, our proposed IQ-CS-based method can achieve satisfactory detection performance with significantly reduced computational complexity.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"991-1004"},"PeriodicalIF":2.9,"publicationDate":"2024-07-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10614929","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142246552","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-07-09DOI: 10.1109/OJSP.2024.3425284
Yongsung Park;Peter Gerstoft
This paper introduces a DOA estimation method for sources beyond the aliasing frequency. The method utilizes multiple frequencies of sources to exploit the frequency difference between them, enabling processing at a frequency below the aliasing frequency. Gridless sparse processing with atomic norm minimization is derived for DOA using difference frequency (DF). This approach achieves higher DOA resolution than previous DF-DOA estimators by enforcing sparsity in the beamforming spectrum and estimating DOAs in the continuous angular domain. We consider one or more measurements in both time (snapshot) and frequency (DF). We also analyze approaches for considering multiple DFs: multi-DF and multi-DF spectral-averaging. Numerical simulations demonstrate the effective performance of the method compared to existing DF techniques.
本文介绍了一种针对超出混叠频率的信号源的 DOA 估算方法。该方法利用信号源的多个频率来利用它们之间的频率差，从而实现低于混叠频率的处理。利用差分频率 (DF) 对 DOA 进行无网格稀疏处理，原子规范最小化。这种方法通过强制波束成形频谱的稀疏性和在连续角域中估计 DOA，实现了比以前的 DF-DOA 估计器更高的 DOA 分辨率。我们同时考虑了时间（快照）和频率（DF）方面的一个或多个测量。我们还分析了考虑多个 DF 的方法：多DF 和多DF 频谱平均。数值模拟证明，与现有的 DF 技术相比，该方法的性能非常有效。
{"title":"Difference Frequency Gridless Sparse Array Processing","authors":"Yongsung Park;Peter Gerstoft","doi":"10.1109/OJSP.2024.3425284","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3425284","url":null,"abstract":"This paper introduces a DOA estimation method for sources beyond the aliasing frequency. The method utilizes multiple frequencies of sources to exploit the frequency difference between them, enabling processing at a frequency below the aliasing frequency. Gridless sparse processing with atomic norm minimization is derived for DOA using difference frequency (DF). This approach achieves higher DOA resolution than previous DF-DOA estimators by enforcing sparsity in the beamforming spectrum and estimating DOAs in the continuous angular domain. We consider one or more measurements in both time (snapshot) and frequency (DF). We also analyze approaches for considering multiple DFs: multi-DF and multi-DF spectral-averaging. Numerical simulations demonstrate the effective performance of the method compared to existing DF techniques.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"914-925"},"PeriodicalIF":2.9,"publicationDate":"2024-07-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10591424","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141964707","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-07-09DOI: 10.1109/OJSP.2024.3425287
Elad Cohen;Hai Victor Habi;Reuven Peretz;Arnon Netzer
Deep neural networks have shown state-of-the-art results in audio source separation tasks in recent years. However, deploying such networks, especially on edge devices, is challenging due to memory and computation requirements. In this work, we focus on quantization, a leading approach for addressing these challenges. We start with a theoretical and empirical analysis of the signal-to-distortion ratio (SDR) in the presence of quantization noise, which presents a fundamental limitation in audio source separation tasks. These analyses show that quantization noise mainly affects performance when the model produces high SDRs. We empirically validate the theoretical insights and illustrate them on audio source separation models. In addition, the empirical analysis shows a high sensitivity to activations quantization, especially to the network's input and output signals. Following the analysis, we propose Fully Quantized Source Separation (FQSS), a quantization-aware training (QAT) method for audio source separation tasks. FQSS introduces a novel loss function based on knowledge distillation that considers quantization-sensitive samples during training and handles the quantization noise of the input and output signals. We validate the efficiency of our method in both time and frequency domains. Finally, we apply FQSS to several architectures (CNNs, LSTMs, and Transformers) and show negligible degradation compared to the full-precision baseline models.
{"title":"Fully Quantized Neural Networks for Audio Source Separation","authors":"Elad Cohen;Hai Victor Habi;Reuven Peretz;Arnon Netzer","doi":"10.1109/OJSP.2024.3425287","DOIUrl":"https://doi.org/10.1109/OJSP.2024.3425287","url":null,"abstract":"Deep neural networks have shown state-of-the-art results in audio source separation tasks in recent years. However, deploying such networks, especially on edge devices, is challenging due to memory and computation requirements. In this work, we focus on quantization, a leading approach for addressing these challenges. We start with a theoretical and empirical analysis of the signal-to-distortion ratio (SDR) in the presence of quantization noise, which presents a fundamental limitation in audio source separation tasks. These analyses show that quantization noise mainly affects performance when the model produces high SDRs. We empirically validate the theoretical insights and illustrate them on audio source separation models. In addition, the empirical analysis shows a high sensitivity to activations quantization, especially to the network's input and output signals. Following the analysis, we propose Fully Quantized Source Separation (FQSS), a quantization-aware training (QAT) method for audio source separation tasks. FQSS introduces a novel loss function based on knowledge distillation that considers quantization-sensitive samples during training and handles the quantization noise of the input and output signals. We validate the efficiency of our method in both time and frequency domains. Finally, we apply FQSS to several architectures (CNNs, LSTMs, and Transformers) and show negligible degradation compared to the full-precision baseline models.","PeriodicalId":73300,"journal":{"name":"IEEE open journal of signal processing","volume":"5 ","pages":"926-933"},"PeriodicalIF":2.9,"publicationDate":"2024-07-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10591369","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141966239","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：