Pub Date : 2024-08-20DOI: 10.1109/TSP.2024.3446572
Ziyu Guo;Tao Yang;Peng Chen;Jun Han;Xiaoyang Zeng;Bo Hu
In this paper, we consider the problem of estimating the angular parameters, i.e., the nominal angle-of-arrivals (AoAs) and angular spreads, of incoherently distributed sources using the phased-array equipped with a single RF chain. We first derive the approximate Fourier series of the received power. The coefficients can be expressed in closed form with the angular parameters. In the case of single source, this finding directly suggests the design of the low-complexity algorithm that performs spatial sampling and discrete Fourier transform to estimate the Fourier series coefficients, from which the nominal AoA and angular spread can be obtained successively. In the case of multiple sources, we focus on one source at one time, and the multiples sources are handled one by one. Based on the Fourier series expression, the power fitting approach is proposed to build the nonlinear least-squares problem. Then, the semi-exhaustive search algorithm is developed to find the solution, which gives the angular parameters of the target source. Additionally, the approximate Cramer-Rao bound is derived as benchmark. The numerical results demonstrate that in certain cases, the proposed methods can even outperform the existing method that uses fully-digital array.
{"title":"Angular Parameter Estimation for Incoherently Distributed Sources With Single RF Chain","authors":"Ziyu Guo;Tao Yang;Peng Chen;Jun Han;Xiaoyang Zeng;Bo Hu","doi":"10.1109/TSP.2024.3446572","DOIUrl":"10.1109/TSP.2024.3446572","url":null,"abstract":"In this paper, we consider the problem of estimating the angular parameters, i.e., the nominal angle-of-arrivals (AoAs) and angular spreads, of incoherently distributed sources using the phased-array equipped with a single RF chain. We first derive the approximate Fourier series of the received power. The coefficients can be expressed in closed form with the angular parameters. In the case of single source, this finding directly suggests the design of the low-complexity algorithm that performs spatial sampling and discrete Fourier transform to estimate the Fourier series coefficients, from which the nominal AoA and angular spread can be obtained successively. In the case of multiple sources, we focus on one source at one time, and the multiples sources are handled one by one. Based on the Fourier series expression, the power fitting approach is proposed to build the nonlinear least-squares problem. Then, the semi-exhaustive search algorithm is developed to find the solution, which gives the angular parameters of the target source. Additionally, the approximate Cramer-Rao bound is derived as benchmark. The numerical results demonstrate that in certain cases, the proposed methods can even outperform the existing method that uses fully-digital array.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"72 ","pages":"5244-5257"},"PeriodicalIF":4.6,"publicationDate":"2024-08-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142022102","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In this paper, we address the problem of detecting a Noise-Like Jammer (NLJ) that does not quickly transmit all the available power but it gradually increases the transmitted power. This control strategy would prevent conventional electronic counter-countermeasures from revealing the presence of a noise power discontinuity in the window under test. As a consequence, the radar system under attack becomes blind requiring a reaction by an expert operator. In order to face such a situation, we devise two innovative NLJ detection architectures by assuming at the design stage specific models for the NLJ power variation. The first model is based on a linear variation law over the observation time, whereas in the second model, the NLJ power experiences unconstrained fluctuations in the window under test. Under these hypotheses, we resort to ad hoc modifications of the generalized likelihood ratio test, where the unknown parameters are replaced by suitable estimates obtained through iterative procedures. The performance analysis, carried out using synthetic data, shows the effectiveness and superiority of the proposed detectors over the conventional approach.
{"title":"ECCM Strategies for Radar Systems Against Smart Noise-Like Jammers","authors":"Dario Benvenuti;Pia Addabbo;Gaetano Giunta;Goffredo Foglia;Danilo Orlando","doi":"10.1109/TSP.2024.3445530","DOIUrl":"10.1109/TSP.2024.3445530","url":null,"abstract":"In this paper, we address the problem of detecting a Noise-Like Jammer (NLJ) that does not quickly transmit all the available power but it gradually increases the transmitted power. This control strategy would prevent conventional electronic counter-countermeasures from revealing the presence of a noise power discontinuity in the window under test. As a consequence, the radar system under attack becomes blind requiring a reaction by an expert operator. In order to face such a situation, we devise two innovative NLJ detection architectures by assuming at the design stage specific models for the NLJ power variation. The first model is based on a linear variation law over the observation time, whereas in the second model, the NLJ power experiences unconstrained fluctuations in the window under test. Under these hypotheses, we resort to ad hoc modifications of the generalized likelihood ratio test, where the unknown parameters are replaced by suitable estimates obtained through iterative procedures. The performance analysis, carried out using synthetic data, shows the effectiveness and superiority of the proposed detectors over the conventional approach.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"72 ","pages":"3912-3926"},"PeriodicalIF":4.6,"publicationDate":"2024-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142007417","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-15DOI: 10.1109/TSP.2024.3444194
Xiaomeng Liu;Timothy N. Davidson
When multiple devices seek to offload computational tasks to their access point, the nature of the multiple access scheme plays a critical role in the system performance. For a system with heterogeneous tasks, we adopt a time-slotted signaling architecture in which different numbers of devices transmit in each slot, subject to individual power constraints. We consider the problem of jointly selecting the devices that will offload, along with optimizing their communication resources (their powers and rates in each time slot, and the time slot lengths) so as to minimize the a weighted sum of the energy expended by the devices. We employ a customized tree search algorithm for the offloading decisions in which a resource allocation problem is solved at each node. For time-division multiple access (TDMA) and “rate optimal” multiple access, we obtain reduced-dimension convex formulations of the resource allocation problem. For non-orthogonal multiple access (NOMA) with independent decoding (ID) or fixed-order sequential decoding (FOSD) we show that the resource allocation problem has a difference-of-convex structure and we develop a successive convex approximation algorithm with feasible point pursuit. Furthermore, for the FOSD scheme we obtain a closed-form expression that provides the optimal decoding order when it is feasible, and efficient algorithms for finding a good decoding order when it is not. Our results capture the inherent tradeoffs between the complexity of a multiple access scheme (and its resource allocation algorithm), and its performance in the computation offloading application.
{"title":"Multiple-Time-Slot Multiple Access Binary Computation Offloading in the $K$-User Case","authors":"Xiaomeng Liu;Timothy N. Davidson","doi":"10.1109/TSP.2024.3444194","DOIUrl":"10.1109/TSP.2024.3444194","url":null,"abstract":"When multiple devices seek to offload computational tasks to their access point, the nature of the multiple access scheme plays a critical role in the system performance. For a system with heterogeneous tasks, we adopt a time-slotted signaling architecture in which different numbers of devices transmit in each slot, subject to individual power constraints. We consider the problem of jointly selecting the devices that will offload, along with optimizing their communication resources (their powers and rates in each time slot, and the time slot lengths) so as to minimize the a weighted sum of the energy expended by the devices. We employ a customized tree search algorithm for the offloading decisions in which a resource allocation problem is solved at each node. For time-division multiple access (TDMA) and “rate optimal” multiple access, we obtain reduced-dimension convex formulations of the resource allocation problem. For non-orthogonal multiple access (NOMA) with independent decoding (ID) or fixed-order sequential decoding (FOSD) we show that the resource allocation problem has a difference-of-convex structure and we develop a successive convex approximation algorithm with feasible point pursuit. Furthermore, for the FOSD scheme we obtain a closed-form expression that provides the optimal decoding order when it is feasible, and efficient algorithms for finding a good decoding order when it is not. Our results capture the inherent tradeoffs between the complexity of a multiple access scheme (and its resource allocation algorithm), and its performance in the computation offloading application.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"72 ","pages":"3927-3944"},"PeriodicalIF":4.6,"publicationDate":"2024-08-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141991939","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-14DOI: 10.1109/TSP.2024.3443875
Guy Revach;Timur Locher;Nir Shlezinger;Ruud J. G. van Sloun;Rik Vullings
Electrocardiography (ECG) signals play a pivotal role in many healthcare applications, especially in at-home monitoring of vital signs. Wearable technologies, which these applications often depend upon, frequently produce low-quality ECG signals. While several methods exist for ECG denoising to enhance signal quality and aid clinical interpretation, they often underperform with ECG data from wearable technology due to limited noise tolerance or inadequate flexibility in capturing ECG dynamics. This paper introduces HKF, a hierarchical and adaptive Kalman filter, which uses a proprietary state space model to effectively capture both intra- and inter-heartbeat dynamics for ECG signal denoising. HKF learns a patient-specific structured prior for the ECG signal's intra-heartbeat dynamics in an online manner, resulting in a filter that adapts to the specific ECG signal characteristics of each patient. In an empirical study, HKF demonstrated superior denoising performance (reduced Mean-Squared Error) while preserving the unique properties of the waveform. In a comparative analysis, HKF outperformed previously proposed methods for ECG denoising, such as the model-based Kalman filter and data-driven autoencoders. This makes it a suitable candidate for applications in extramural healthcare settings.
{"title":"HKF: Hierarchical Kalman Filtering With Online Learned Evolution Priors for Adaptive ECG Denoising","authors":"Guy Revach;Timur Locher;Nir Shlezinger;Ruud J. G. van Sloun;Rik Vullings","doi":"10.1109/TSP.2024.3443875","DOIUrl":"10.1109/TSP.2024.3443875","url":null,"abstract":"Electrocardiography (ECG) signals play a pivotal role in many healthcare applications, especially in at-home monitoring of vital signs. Wearable technologies, which these applications often depend upon, frequently produce low-quality ECG signals. While several methods exist for ECG denoising to enhance signal quality and aid clinical interpretation, they often underperform with ECG data from wearable technology due to limited noise tolerance or inadequate flexibility in capturing ECG dynamics. This paper introduces HKF, a hierarchical and adaptive Kalman filter, which uses a proprietary state space model to effectively capture both intra- and inter-heartbeat dynamics for ECG signal denoising. HKF learns a patient-specific structured prior for the ECG signal's intra-heartbeat dynamics in an online manner, resulting in a filter that adapts to the specific ECG signal characteristics of each patient. In an empirical study, HKF demonstrated superior denoising performance (reduced Mean-Squared Error) while preserving the unique properties of the waveform. In a comparative analysis, HKF outperformed previously proposed methods for ECG denoising, such as the model-based Kalman filter and data-driven autoencoders. This makes it a suitable candidate for applications in extramural healthcare settings.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"72 ","pages":"3990-4006"},"PeriodicalIF":4.6,"publicationDate":"2024-08-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141986281","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
A fundamental challenge in cognitive radio is the detection of primary users in a licensed spectrum. Cooperative sensing, which utilizes multiple receivers distributed across different locations, offers the advantage of utilizing multiple antennas and achieving spatial diversity gain. However, successful implementation of cooperative sensing relies on the ideal exchange of information among cooperating receivers, which may not always be feasible in real-world scenarios. In this paper, we consider the cooperative sensing problem in a non-ideal communication scenario, where the raw data broadcasted from a receiving node can be received by only a subset of the nearby nodes. Existing multiantenna detectors can not deal with such a scenario. To tackle this issue, we propose a novel cooperative sensing scheme, where each node sends only its local correlation coefficients to the fusion center. A detection mechanism based on factorizing the partially received sample covariance matrix is developed. To achieve fast convergence and avoid exhaustive step size tuning, a Bregman proximal method, based on an alternating minimization algorithm (with convergence guarantees), is also developed. The advantages of our proposed cooperative scheme is demonstrated through numerical simulations.
{"title":"A Matrix-Factorization-Error-Ratio Approach to Cooperative Sensing in Non-Ideal Communication Environment","authors":"Rui Zhou;Wenqiang Pu;Licheng Zhao;Ming-Yi You;Qingjiang Shi;Sergios Theodoridis","doi":"10.1109/TSP.2024.3443291","DOIUrl":"10.1109/TSP.2024.3443291","url":null,"abstract":"A fundamental challenge in cognitive radio is the detection of primary users in a licensed spectrum. Cooperative sensing, which utilizes multiple receivers distributed across different locations, offers the advantage of utilizing multiple antennas and achieving spatial diversity gain. However, successful implementation of cooperative sensing relies on the ideal exchange of information among cooperating receivers, which may not always be feasible in real-world scenarios. In this paper, we consider the cooperative sensing problem in a non-ideal communication scenario, where the raw data broadcasted from a receiving node can be received by only a subset of the nearby nodes. Existing multiantenna detectors can not deal with such a scenario. To tackle this issue, we propose a novel cooperative sensing scheme, where each node sends only its local correlation coefficients to the fusion center. A detection mechanism based on factorizing the partially received sample covariance matrix is developed. To achieve fast convergence and avoid exhaustive step size tuning, a Bregman proximal method, based on an alternating minimization algorithm (with convergence guarantees), is also developed. The advantages of our proposed cooperative scheme is demonstrated through numerical simulations.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"72 ","pages":"3851-3864"},"PeriodicalIF":4.6,"publicationDate":"2024-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141980686","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-13DOI: 10.1109/TSP.2024.3443121
Youlin Fan;Bo Jiu;Wenqiang Pu;Ziniu Li;Kang Li;Hongwei Liu
This paper studies the problem of sensing mainlobe jamming strategy through interaction samples between a frequency agile radar and a transmit/receive time-sharing jammer. We model this interaction as an episodic Markov decision process, where the jammer's strategy is treated as the state transition probability that needs to be learned. To effectively learn the strategy, we employ two sensing criteria from the imitation learning perspective: Behavioral Cloning (BC) and Generative Adversarial Imitation Learning (GAIL). These criteria enable us to imitate the jammer's strategy based on collected interaction samples. Our theoretical analysis indicates that GAIL provides more accurate strategy sensing performance, while BC offers faster learning. Experimental results corroborate these findings. Additionally, empirical evidence shows that our trained anti-jamming strategies, informed by either BC or GAIL, significantly outperform existing intelligent anti-jamming strategy learning methods in terms of sample efficiency.
本文研究了通过频率敏捷雷达与发射/接收分时干扰器之间的交互样本来感知主波干扰策略的问题。我们将这种交互建模为一个偶发马尔可夫决策过程，其中干扰者的策略被视为需要学习的状态转换概率。为了有效地学习策略，我们从模仿学习的角度出发，采用了两种感知标准：行为克隆（BC）和生成对抗模仿学习（GAIL）。这些标准使我们能够根据收集到的交互样本模仿干扰者的策略。我们的理论分析表明，GAIL 能提供更准确的策略感知性能，而 BC 则能提供更快的学习速度。实验结果证实了这些结论。此外，经验证据表明，在 BC 或 GAIL 的指导下，我们训练的反干扰策略在样本效率方面明显优于现有的智能反干扰策略学习方法。
{"title":"Sensing Jamming Strategy From Limited Observations: An Imitation Learning Perspective","authors":"Youlin Fan;Bo Jiu;Wenqiang Pu;Ziniu Li;Kang Li;Hongwei Liu","doi":"10.1109/TSP.2024.3443121","DOIUrl":"10.1109/TSP.2024.3443121","url":null,"abstract":"This paper studies the problem of sensing mainlobe jamming strategy through interaction samples between a frequency agile radar and a transmit/receive time-sharing jammer. We model this interaction as an episodic Markov decision process, where the jammer's strategy is treated as the state transition probability that needs to be learned. To effectively learn the strategy, we employ two sensing criteria from the imitation learning perspective: Behavioral Cloning (BC) and Generative Adversarial Imitation Learning (GAIL). These criteria enable us to imitate the jammer's strategy based on collected interaction samples. Our theoretical analysis indicates that GAIL provides more accurate strategy sensing performance, while BC offers faster learning. Experimental results corroborate these findings. Additionally, empirical evidence shows that our trained anti-jamming strategies, informed by either BC or GAIL, significantly outperform existing intelligent anti-jamming strategy learning methods in terms of sample efficiency.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"72 ","pages":"4098-4114"},"PeriodicalIF":4.6,"publicationDate":"2024-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141980764","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
We present a robust framework to perform linear regression with missing entries in the features. By considering an elliptical data distribution, and specifically a multivariate normal model, we are able to conditionally formulate a distribution for the missing entries and present a robust framework, which minimizes the worst-case error caused by the uncertainty in the missing data. We show that the proposed formulation, which naturally takes into account the dependency between different variables, ultimately reduces to a convex program, for which we develop a customized and scalable solver. We analyze the consistency and structural behavior of the proposed framework in asymptotic regimes, and present technical discussions to estimate the required input parameters. We complement our analysis with experiments performed on synthetic, semi-synthetic, and real data, and show how the proposed formulation improves the prediction accuracy and robustness, and outperforms the competing techniques.
{"title":"An Adversarially Robust Formulation of Linear Regression With Missing Data","authors":"Alireza Aghasi;Saeed Ghadimi;Yue Xing;Mohammadjavad Feizollahi","doi":"10.1109/TSP.2024.3442712","DOIUrl":"10.1109/TSP.2024.3442712","url":null,"abstract":"We present a robust framework to perform linear regression with missing entries in the features. By considering an elliptical data distribution, and specifically a multivariate normal model, we are able to conditionally formulate a distribution for the missing entries and present a robust framework, which minimizes the worst-case error caused by the uncertainty in the missing data. We show that the proposed formulation, which naturally takes into account the dependency between different variables, ultimately reduces to a convex program, for which we develop a customized and scalable solver. We analyze the consistency and structural behavior of the proposed framework in asymptotic regimes, and present technical discussions to estimate the required input parameters. We complement our analysis with experiments performed on synthetic, semi-synthetic, and real data, and show how the proposed formulation improves the prediction accuracy and robustness, and outperforms the competing techniques.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"72 ","pages":"4950-4966"},"PeriodicalIF":4.6,"publicationDate":"2024-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141980775","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-13DOI: 10.1109/TSP.2024.3440353
Cheng Du;Yi Jiang
In a massive multi-input multi-output (MIMO) cellular communication system, the conventional beam-sweeping scheme for common message broadcasting provides high beamforming gain but requires too many time slots due to the narrowness of the beams. To reduce the beam sweeping time while maintaining a sufficient beamforming gain, this paper focuses on designing broad beams with tunable beamwidths. First, by over-sampling a step-chirp analog signal, we construct a novel sequence family termed the generalized step-chirp (GSC) sequence with a simple closed-form expression, in which some parameters can be tuned to flexibly adjust the beamwidth and coarsen the phase resolution. The beamforming matrix of a uniform rectangular array (URA) of single–polarized antennas can be taken as the outer product of two GSC sequences. Second, by exploiting the full degree of freedom of URA of dual-polarized antennas, we further reduce the power variations in the passband by using a numerical algorithm. The algorithm can be drastically accelerated by exploiting the specific structure of the problem. Both schemes can be implemented using an analog phase shifter network (PSN) with finite resolution.
在大规模多输入多输出（MIMO）蜂窝通信系统中，用于普通信息广播的传统波束扫频方案可提供较高的波束成形增益，但由于波束较窄，需要的时隙过多。为了在保持足够波束增益的同时缩短波束扫描时间，本文重点研究了设计波束宽度可调的宽波束。首先，通过对阶跃啁啾模拟信号进行过采样，我们构建了一个新的序列族，称为广义阶跃啁啾（GSC）序列，它具有简单的闭式表达式，其中一些参数可以调整，以灵活地调整波束宽度和粗化相位分辨率。单极化天线均匀矩形阵列（URA）的波束成形矩阵可视为两个 GSC 序列的外积。其次，通过利用双极化天线 URA 的全自由度，我们使用数值算法进一步减少了通带中的功率变化。通过利用问题的特定结构，可以大大加快算法的速度。这两种方案都可以使用具有有限分辨率的模拟移相器网络（PSN）来实现。
{"title":"Broad Beam Designs for Broadcast Channels","authors":"Cheng Du;Yi Jiang","doi":"10.1109/TSP.2024.3440353","DOIUrl":"10.1109/TSP.2024.3440353","url":null,"abstract":"In a massive multi-input multi-output (MIMO) cellular communication system, the conventional beam-sweeping scheme for common message broadcasting provides high beamforming gain but requires too many time slots due to the narrowness of the beams. To reduce the beam sweeping time while maintaining a sufficient beamforming gain, this paper focuses on designing broad beams with tunable beamwidths. First, by over-sampling a step-chirp analog signal, we construct a novel sequence family termed the generalized step-chirp (GSC) sequence with a simple closed-form expression, in which some parameters can be tuned to flexibly adjust the beamwidth and coarsen the phase resolution. The beamforming matrix of a uniform rectangular array (URA) of single–polarized antennas can be taken as the outer product of two GSC sequences. Second, by exploiting the full degree of freedom of URA of dual-polarized antennas, we further reduce the power variations in the passband by using a numerical algorithm. The algorithm can be drastically accelerated by exploiting the specific structure of the problem. Both schemes can be implemented using an analog phase shifter network (PSN) with finite resolution.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"72 ","pages":"3819-3833"},"PeriodicalIF":4.6,"publicationDate":"2024-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141980687","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-08-13DOI: 10.1109/TSP.2024.3442858
Wei Xu;An Liu;Yiting Zhang;Vincent Lau
Efficient learning and model compression algorithm for deep neural network (DNN) is a key workhorse behind the rise of deep learning (DL). In this work, we propose a message passing-based Bayesian deep learning algorithm called EM-TDAMP to avoid the drawbacks of traditional stochastic gradient descent (SGD)-based learning algorithms and regularization-based model compression methods. Specifically, we formulate the problem of DNN learning and compression as a sparse Bayesian inference problem, in which group sparse prior is employed to achieve structured model compression. Then, we propose an expectation maximization (EM) framework to estimate posterior distributions for parameters (E-step) and update hyperparameters (M-step), where the E-step is realized by a newly proposed turbo deep approximate message passing (TDAMP) algorithm. We further extend the EM-TDAMP and propose a novel Bayesian federated learning framework, in which the clients perform TDAMP to efficiently calculate the local posterior distributions based on the local data, and the central server first aggregates the local posterior distributions to update the global posterior distributions and then update hyperparameters based on EM to accelerate convergence. We detail the application of EM-TDAMP to Boston housing price prediction and handwriting recognition, and present extensive numerical results to demonstrate the advantages of EM-TDAMP.
{"title":"Bayesian Deep Learning via Expectation Maximization and Turbo Deep Approximate Message Passing","authors":"Wei Xu;An Liu;Yiting Zhang;Vincent Lau","doi":"10.1109/TSP.2024.3442858","DOIUrl":"10.1109/TSP.2024.3442858","url":null,"abstract":"Efficient learning and model compression algorithm for deep neural network (DNN) is a key workhorse behind the rise of deep learning (DL). In this work, we propose a message passing-based Bayesian deep learning algorithm called EM-TDAMP to avoid the drawbacks of traditional stochastic gradient descent (SGD)-based learning algorithms and regularization-based model compression methods. Specifically, we formulate the problem of DNN learning and compression as a sparse Bayesian inference problem, in which group sparse prior is employed to achieve structured model compression. Then, we propose an expectation maximization (EM) framework to estimate posterior distributions for parameters (E-step) and update hyperparameters (M-step), where the E-step is realized by a newly proposed turbo deep approximate message passing (TDAMP) algorithm. We further extend the EM-TDAMP and propose a novel Bayesian federated learning framework, in which the clients perform TDAMP to efficiently calculate the local posterior distributions based on the local data, and the central server first aggregates the local posterior distributions to update the global posterior distributions and then update hyperparameters based on EM to accelerate convergence. We detail the application of EM-TDAMP to Boston housing price prediction and handwriting recognition, and present extensive numerical results to demonstrate the advantages of EM-TDAMP.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"72 ","pages":"3865-3878"},"PeriodicalIF":4.6,"publicationDate":"2024-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141980685","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","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：