Pub Date : 2024-11-04DOI: 10.1109/TSP.2024.3484582
Yu-Hang Xiao;David Ramírez;Lei Huang;Xiao Peng Li;Hing Cheung So
One-bit sampling has emerged as a promising technique in multiple-input multiple-output (MIMO) radar systems due to its ability to significantly reduce data volume, hardware complexity, and power consumption. Nevertheless, current detection methods have not adequately addressed the impact of colored noise, which is frequently encountered in real scenarios. In this paper, we present a novel detection method that accounts for colored noise in MIMO radar systems. Specifically, we derive Rao's test by computing the derivative of the likelihood function with respect to the target reflectivity parameter and the Fisher information matrix, resulting in a detector that takes the form of a weighted matched filter. To ensure constant false alarm rate (CFAR), we also consider noise covariance uncertainty and examine its effect on the probability of false alarm. The detection probability is also studied analytically. Simulation results demonstrate that the proposed detector provides considerable performance gains in the presence of colored noise.
在多输入多输出（MIMO）雷达系统中，单比特采样已成为一种很有前途的技术，因为它能显著减少数据量、硬件复杂性和功耗。然而，目前的检测方法还没有充分解决实际场景中经常遇到的彩色噪声的影响。在本文中，我们提出了一种在 MIMO 雷达系统中考虑彩色噪声的新型检测方法。具体来说，我们通过计算与目标反射率参数和费舍尔信息矩阵相关的似然函数导数来推导 Rao 检验，从而得到一种加权匹配滤波器形式的检测器。为了确保恒定的误报率（CFAR），我们还考虑了噪声协方差的不确定性，并研究了其对误报概率的影响。我们还对检测概率进行了分析研究。仿真结果表明，在存在彩色噪声的情况下，所提出的检测器能提供相当大的性能提升。
{"title":"One-Bit Target Detection in Colocated MIMO Radar With Colored Background Noise","authors":"Yu-Hang Xiao;David Ramírez;Lei Huang;Xiao Peng Li;Hing Cheung So","doi":"10.1109/TSP.2024.3484582","DOIUrl":"10.1109/TSP.2024.3484582","url":null,"abstract":"One-bit sampling has emerged as a promising technique in multiple-input multiple-output (MIMO) radar systems due to its ability to significantly reduce data volume, hardware complexity, and power consumption. Nevertheless, current detection methods have not adequately addressed the impact of colored noise, which is frequently encountered in real scenarios. In this paper, we present a novel detection method that accounts for colored noise in MIMO radar systems. Specifically, we derive Rao's test by computing the derivative of the likelihood function with respect to the target reflectivity parameter and the Fisher information matrix, resulting in a detector that takes the form of a weighted matched filter. To ensure constant false alarm rate (CFAR), we also consider noise covariance uncertainty and examine its effect on the probability of false alarm. The detection probability is also studied analytically. Simulation results demonstrate that the proposed detector provides considerable performance gains in the presence of colored noise.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"72 ","pages":"5274-5290"},"PeriodicalIF":4.6,"publicationDate":"2024-11-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142580308","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, channel parameter estimation and location sensing under phase noise (PN) are achieved based on nested tensor decomposition. The PN has two effects on the received signal, i.e., common phase error (CPE) and inter-carrier interference (ICI). Using the multi-dimensionality of millimeter wave channels, the received training signal is formulated as a nested parallel factor (PARAFAC) tensor model. Resorting to the compression and line search, CPE and compound channel are iteratively estimated by fitting the outer PARAFAC model in the first stage. In the second stage, a closed-form algorithm and an iterative-form algorithm are respectively developed to fit the inner PARAFAC model. Specifically, the closed-form one leverages the spatial smoothing and forward-backward, and the iterative-form one utilizes the unitary transformation. Channel parameter estimation and location sensing of mobile station and scatterers are achieved in the third stage. The Cram $acute{text{e}}$ r-Rao bounds (CRBs) of CPE and channel parameters are also derived to provide benchmarks. Compared with existing algorithms, the proposed algorithms exhibit performance close to CRBs, and show improved performance with low computational complexity. Besides, the proposed algorithms can cope with more challenging cases where line-of-sight (LOS) path does not exist and non-LOS paths are spatially correlated even with significant ICI.
{"title":"Channel Parameter Estimation and Location Sensing in mmWave Systems Under Phase Noise via Nested PARAFAC Analysis","authors":"Meng Han;Jianhe Du;Yuanzhi Chen;Libiao Jin;Feifei Gao","doi":"10.1109/TSP.2024.3488781","DOIUrl":"10.1109/TSP.2024.3488781","url":null,"abstract":"In this paper, channel parameter estimation and location sensing under phase noise (PN) are achieved based on nested tensor decomposition. The PN has two effects on the received signal, i.e., common phase error (CPE) and inter-carrier interference (ICI). Using the multi-dimensionality of millimeter wave channels, the received training signal is formulated as a nested parallel factor (PARAFAC) tensor model. Resorting to the compression and line search, CPE and compound channel are iteratively estimated by fitting the outer PARAFAC model in the first stage. In the second stage, a closed-form algorithm and an iterative-form algorithm are respectively developed to fit the inner PARAFAC model. Specifically, the closed-form one leverages the spatial smoothing and forward-backward, and the iterative-form one utilizes the unitary transformation. Channel parameter estimation and location sensing of mobile station and scatterers are achieved in the third stage. The Cram\u0000<inline-formula><tex-math>$acute{text{e}}$</tex-math></inline-formula>\u0000r-Rao bounds (CRBs) of CPE and channel parameters are also derived to provide benchmarks. Compared with existing algorithms, the proposed algorithms exhibit performance close to CRBs, and show improved performance with low computational complexity. Besides, the proposed algorithms can cope with more challenging cases where line-of-sight (LOS) path does not exist and non-LOS paths are spatially correlated even with significant ICI.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"72 ","pages":"5422-5438"},"PeriodicalIF":4.6,"publicationDate":"2024-10-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142561935","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 : 2024-10-30DOI: 10.1109/TSP.2024.3488554
Carlos Alejandro Lopez;Jaume Riba
The purpose of this article is to present the Parametric Minimum Error Entropy (PMEE) principle and to show a case study of the proposed criterion in a blind sensor fusion and regression problem. This case study consists on the estimation of a temporal series with a certain temporal invariance, which is measured from multiple independent sensors with unknown variances and unknown mutual correlations of the measurement errors. In this setting, we show that a particular case of the PMEE criterion is obtained from the Conditional Maximum Likelihood (CML) principle of the measurement model, leading to a semi-data-driven solution. Despite the fact that Information Theoretic Criteria (ITC) are inherently robust, they often result in difficult non-convex optimization problems. Our proposal is to address the non-convexity by means of a Majorization-Minimization (MM) based algorithm. We prove the conditions in which the resulting solution of the proposed algorithm reaches a stationary point of the original problem. In fact, the aforementioned global convergence of the proposed algorithm is possible thanks to a reformulation of the original cost function in terms of a variable constrained in the Grassmann manifold. As shown in this paper, the latter procedure is possible thanks to a homogeneity property of the PMEE cost function.
{"title":"Parametric Minimum Error Entropy Criterion: A Case Study in Blind Sensor Fusion and Regression Problems","authors":"Carlos Alejandro Lopez;Jaume Riba","doi":"10.1109/TSP.2024.3488554","DOIUrl":"10.1109/TSP.2024.3488554","url":null,"abstract":"The purpose of this article is to present the Parametric Minimum Error Entropy (PMEE) principle and to show a case study of the proposed criterion in a blind sensor fusion and regression problem. This case study consists on the estimation of a temporal series with a certain temporal invariance, which is measured from multiple independent sensors with unknown variances and unknown mutual correlations of the measurement errors. In this setting, we show that a particular case of the PMEE criterion is obtained from the Conditional Maximum Likelihood (CML) principle of the measurement model, leading to a \u0000<italic>semi-data-driven</i>\u0000 solution. Despite the fact that Information Theoretic Criteria (ITC) are inherently robust, they often result in difficult non-convex optimization problems. Our proposal is to address the non-convexity by means of a Majorization-Minimization (MM) based algorithm. We prove the conditions in which the resulting solution of the proposed algorithm reaches a stationary point of the original problem. In fact, the aforementioned global convergence of the proposed algorithm is possible thanks to a reformulation of the original cost function in terms of a variable constrained in the Grassmann manifold. As shown in this paper, the latter procedure is possible thanks to a homogeneity property of the PMEE cost function.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"72 ","pages":"5091-5106"},"PeriodicalIF":4.6,"publicationDate":"2024-10-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142555958","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 : 2024-10-28DOI: 10.1109/TSP.2024.3487256
Liang Liu;Zhan Zhang;Xinyun Zhang;Ping Wei;Jiancheng An;Hongbin Li
As the demand for wireless communication continues to surge, spectrum congestion becomes more severe. Compressive spectrum sensing with joint frequency and Direction Of Arrival (DOA) estimation is instrumental to enable efficient spectrum utilization in ultra-wideband scenarios. To reduce hardware complexity, this paper proposes a resource-efficient array undersampling structure which is distinctive in that each array element only connects to one branch of the Modulated Wideband Converter (MWC), and the modulated signals in different branches have identical periods but different waveforms. The proposed structure integrates spatial sampling by array elements and temporal undersampling by the MWC. A signal model is developed for the proposed sampling structure, which can deal with more general scenarios, involving multiple subband signals, and cross-band signals. Joint spectrum sensing algorithms are proposed based on compressed sensing and subspace decomposition. Additionally, multi-resolution grid optimization strategy is designed to eliminate grid effect with low computational complexity. We also analyze the impact of structural parameters on algorithm performance, which reveals that the number of array elements determines the maximum number of signals that can be estimated within a subband, while the equivalent total number of channels of the reception system determines the maximum number of signals that the system can estimate. Our analysis shows that the proposed sampling structure offers a greater flexibility in structural parameter selection and system design. Finally, simulations show that under the condition of the same or similar average sampling rate, the proposed structure and corresponding methods can achieve higher spectrum and DOA estimation accuracy.
{"title":"Joint Spectrum Sensing and DOA Estimation Based on a Resource-Efficient Sub-Nyquist Array Receiver","authors":"Liang Liu;Zhan Zhang;Xinyun Zhang;Ping Wei;Jiancheng An;Hongbin Li","doi":"10.1109/TSP.2024.3487256","DOIUrl":"10.1109/TSP.2024.3487256","url":null,"abstract":"As the demand for wireless communication continues to surge, spectrum congestion becomes more severe. Compressive spectrum sensing with joint frequency and Direction Of Arrival (DOA) estimation is instrumental to enable efficient spectrum utilization in ultra-wideband scenarios. To reduce hardware complexity, this paper proposes a resource-efficient array undersampling structure which is distinctive in that each array element only connects to one branch of the Modulated Wideband Converter (MWC), and the modulated signals in different branches have identical periods but different waveforms. The proposed structure integrates spatial sampling by array elements and temporal undersampling by the MWC. A signal model is developed for the proposed sampling structure, which can deal with more general scenarios, involving multiple subband signals, and cross-band signals. Joint spectrum sensing algorithms are proposed based on compressed sensing and subspace decomposition. Additionally, multi-resolution grid optimization strategy is designed to eliminate grid effect with low computational complexity. We also analyze the impact of structural parameters on algorithm performance, which reveals that the number of array elements determines the maximum number of signals that can be estimated within a subband, while the equivalent total number of channels of the reception system determines the maximum number of signals that the system can estimate. Our analysis shows that the proposed sampling structure offers a greater flexibility in structural parameter selection and system design. Finally, simulations show that under the condition of the same or similar average sampling rate, the proposed structure and corresponding methods can achieve higher spectrum and DOA estimation accuracy.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"72 ","pages":"5354-5370"},"PeriodicalIF":4.6,"publicationDate":"2024-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142536789","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}
This contribution investigates the problem of three-dimensional (3-D) angle-of-arrival (AOA) source localization (SL) in the presence of symmetric $alpha$ -stable ( $mathcal{Salpha S}$ ) impulsive noise for $alphain(0,2]$ . The azimuth and elevation angle measurements are initially rewritten into a pseudolinear form using spherical coordinate conversion, thereby making them more manageable. Subsequently, we adopt the maximum correntropy criterion with variable center (MCC-VC) to devise a robust 3-D AOA location estimator that functions effectively without the prior knowledge of parameters governing the impulsiveness and dispersion of $mathcal{Salpha S}$ noise distributions. While it gives rise to a straightforward alternating minimization algorithmic framework, our analysis reveals that solely embracing MCC-VC leads to bias issues stemming from the correlation between the measurement matrix and noise. Aiming at addressing such a challenge, we introduce instrumental variables (IVs) to develop a bias-reduced maximum correntropy criterion (MCC) estimator, termed MCC with IV (MCC-IV). Simulation results illustrate a considerable performance enhancement of MCC-IV compared to existing schemes for 3-D AOA SL, particularly in achieving mean square error much closer to the Cramér–Rao lower bound and mitigating bias substantially.
{"title":"Robust 3-D AOA Localization Based on Maximum Correntropy Criterion With Variable Center","authors":"Keyuan Hu;Wenxin Xiong;Zhi-Yong Wang;Hing Cheung So;Chi-Sing Leung","doi":"10.1109/TSP.2024.3486817","DOIUrl":"10.1109/TSP.2024.3486817","url":null,"abstract":"This contribution investigates the problem of three-dimensional (3-D) angle-of-arrival (AOA) source localization (SL) in the presence of symmetric \u0000<inline-formula><tex-math>$alpha$</tex-math></inline-formula>\u0000-stable (\u0000<inline-formula><tex-math>$mathcal{Salpha S}$</tex-math></inline-formula>\u0000) impulsive noise for \u0000<inline-formula><tex-math>$alphain(0,2]$</tex-math></inline-formula>\u0000. The azimuth and elevation angle measurements are initially rewritten into a pseudolinear form using spherical coordinate conversion, thereby making them more manageable. Subsequently, we adopt the maximum correntropy criterion with variable center (MCC-VC) to devise a robust 3-D AOA location estimator that functions effectively without the prior knowledge of parameters governing the impulsiveness and dispersion of \u0000<inline-formula><tex-math>$mathcal{Salpha S}$</tex-math></inline-formula>\u0000 noise distributions. While it gives rise to a straightforward alternating minimization algorithmic framework, our analysis reveals that solely embracing MCC-VC leads to bias issues stemming from the correlation between the measurement matrix and noise. Aiming at addressing such a challenge, we introduce instrumental variables (IVs) to develop a bias-reduced maximum correntropy criterion (MCC) estimator, termed MCC with IV (MCC-IV). Simulation results illustrate a considerable performance enhancement of MCC-IV compared to existing schemes for 3-D AOA SL, particularly in achieving mean square error much closer to the Cramér–Rao lower bound and mitigating bias substantially.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"72 ","pages":"5021-5035"},"PeriodicalIF":4.6,"publicationDate":"2024-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142536790","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 : 2024-10-25DOI: 10.1109/TSP.2024.3486240
Chenye Yang;Guanlin Liu;Lifeng Lai
In this paper, we investigate rewards attacks on stochastic multi-armed bandit algorithms with non-stationary environment. The attacker's goal is to force the victim algorithm to choose a suboptimal arm most of the time while incurring a small attack cost. We consider three increasingly general attack scenarios, each of which has different assumptions about the environment, victim algorithm and information available to the attacker. We propose three attack strategies, one for each considered scenario, and prove that they are successful in terms of expected target arm selection and attack cost. We also propose a defense non-stationary algorithm that is able to defend any attacker whose attack cost is bounded by a budget, and prove that it is robust to attacks. The simulation results validate our theoretical analysis.
{"title":"Stochastic Bandits With Non-Stationary Rewards: Reward Attack and Defense","authors":"Chenye Yang;Guanlin Liu;Lifeng Lai","doi":"10.1109/TSP.2024.3486240","DOIUrl":"10.1109/TSP.2024.3486240","url":null,"abstract":"In this paper, we investigate rewards attacks on stochastic multi-armed bandit algorithms with non-stationary environment. The attacker's goal is to force the victim algorithm to choose a suboptimal arm most of the time while incurring a small attack cost. We consider three increasingly general attack scenarios, each of which has different assumptions about the environment, victim algorithm and information available to the attacker. We propose three attack strategies, one for each considered scenario, and prove that they are successful in terms of expected target arm selection and attack cost. We also propose a defense non-stationary algorithm that is able to defend any attacker whose attack cost is bounded by a budget, and prove that it is robust to attacks. The simulation results validate our theoretical analysis.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"72 ","pages":"5007-5020"},"PeriodicalIF":4.6,"publicationDate":"2024-10-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142490485","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 : 2024-10-25DOI: 10.1109/TSP.2024.3486533
Ruifu Li;Danijela Cabric
Atomic norm minimization is of great interest in various applications of sparse signal processing including super-resolution line-spectral estimation and signal denoising. In practice, atomic norm minimization (ANM) is formulated as semi-definite programming (SDP) that is generally hard to solve. This work introduces a low-complexity solver for a type of ANM known as atomic norm soft thresholding (AST). The proposed method uses the framework of coordinate descent and exploits the sparsity-inducing nature of atomic norm regularization. Specifically, this work first provides an equivalent, non-convex formulation of AST. It is then proved that applying a coordinate descent algorithm on the non-convex formulation leads to convergence to the global solution. For the case of a single measurement vector of length $N$ and complex exponential basis, the complexity of each step in the coordinate descent procedure is $mathcal{O}(Nlog N)$ , rendering the method efficient for large-scale problems. Through simulations, for sparse problems the proposed solver is shown to be faster than alternating direction method of multiplier (ADMM) or customized interior point SDP solver. Numerical simulations demonstrate that the coordinate descent solver can be modified for AST with multiple dimensions and multiple measurement vectors as well as a variety of other continuous basis.
{"title":"A Coordinate Descent Approach to Atomic Norm Denoising","authors":"Ruifu Li;Danijela Cabric","doi":"10.1109/TSP.2024.3486533","DOIUrl":"10.1109/TSP.2024.3486533","url":null,"abstract":"Atomic norm minimization is of great interest in various applications of sparse signal processing including super-resolution line-spectral estimation and signal denoising. In practice, atomic norm minimization (ANM) is formulated as semi-definite programming (SDP) that is generally hard to solve. This work introduces a low-complexity solver for a type of ANM known as atomic norm soft thresholding (AST). The proposed method uses the framework of coordinate descent and exploits the sparsity-inducing nature of atomic norm regularization. Specifically, this work first provides an equivalent, non-convex formulation of AST. It is then proved that applying a coordinate descent algorithm on the non-convex formulation leads to convergence to the global solution. For the case of a single measurement vector of length \u0000<inline-formula><tex-math>$N$</tex-math></inline-formula>\u0000 and complex exponential basis, the complexity of each step in the coordinate descent procedure is \u0000<inline-formula><tex-math>$mathcal{O}(Nlog N)$</tex-math></inline-formula>\u0000, rendering the method efficient for large-scale problems. Through simulations, for sparse problems the proposed solver is shown to be faster than alternating direction method of multiplier (ADMM) or customized interior point SDP solver. Numerical simulations demonstrate that the coordinate descent solver can be modified for AST with multiple dimensions and multiple measurement vectors as well as a variety of other continuous basis.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"72 ","pages":"5077-5090"},"PeriodicalIF":4.6,"publicationDate":"2024-10-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142490473","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 : 2024-10-24DOI: 10.1109/TSP.2024.3486169
Zhongyuan Zhao;Kailei Xu;Wei Hong;Mugen Peng;Zhiguo Ding;Tony Q. S. Quek;Howard H. Yang
Analog over-the-air (A-OTA) computing is an effective approach to achieving distributed learning among multiple end-user devices within a bandwidth-constrained spectrum. In this paradigm, users’ intermediate parameters, such as gradients, are modulated onto a set of common waveforms and concurrently transmitted to the parameter server. Benefiting from the superposition property of multi-access channels, the server can obtain an automatically aggregated global gradient from the received signal without decoding individual user's information. Nonetheless, the scarcity of orthogonal waveforms constrains such a paradigm from adopting complex deep learning models. In this paper, we develop model pruning strategies for A-OTA distributed learning, balancing the tradeoff between communication efficiency and learning performance. Specifically, we design an importance measure to evaluate the contribution of each entry in the model parameter based on the noisy aggregated gradient introduced by A-OTA computing. We also derive an analytical expression for the training error bound, which shows that the proposed scheme can converge even when the aggregated gradient is corrupted by heavy-tailed interference with unbounded variance. We further improve the developed algorithm by incorporating the momentum method to (a) enhance the design of the importance measure and (b) accelerate the model convergence rate. Extensive experiments are conducted to validate the performance gains achieved by our proposed scheme and verify the correctness of analytical results.
{"title":"Model Pruning for Distributed Learning Over the Air","authors":"Zhongyuan Zhao;Kailei Xu;Wei Hong;Mugen Peng;Zhiguo Ding;Tony Q. S. Quek;Howard H. Yang","doi":"10.1109/TSP.2024.3486169","DOIUrl":"10.1109/TSP.2024.3486169","url":null,"abstract":"Analog over-the-air (A-OTA) computing is an effective approach to achieving distributed learning among multiple end-user devices within a bandwidth-constrained spectrum. In this paradigm, users’ intermediate parameters, such as gradients, are modulated onto a set of common waveforms and concurrently transmitted to the parameter server. Benefiting from the superposition property of multi-access channels, the server can obtain an automatically aggregated global gradient from the received signal without decoding individual user's information. Nonetheless, the scarcity of orthogonal waveforms constrains such a paradigm from adopting complex deep learning models. In this paper, we develop model pruning strategies for A-OTA distributed learning, balancing the tradeoff between communication efficiency and learning performance. Specifically, we design an importance measure to evaluate the contribution of each entry in the model parameter based on the noisy aggregated gradient introduced by A-OTA computing. We also derive an analytical expression for the training error bound, which shows that the proposed scheme can converge even when the aggregated gradient is corrupted by heavy-tailed interference with unbounded variance. We further improve the developed algorithm by incorporating the momentum method to (a) enhance the design of the importance measure and (b) accelerate the model convergence rate. Extensive experiments are conducted to validate the performance gains achieved by our proposed scheme and verify the correctness of analytical results.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"72 ","pages":"5533-5549"},"PeriodicalIF":4.6,"publicationDate":"2024-10-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142489954","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 : 2024-10-23DOI: 10.1109/TSP.2024.3484908
Yuhui Song;Zijun Gong;Yuanzhu Chen;Cheng Li
Sparse Bayesian Learning (SBL) has emerged as a powerful tool for sparse signal recovery, due to its superior performance. However, the practical implementation of SBL faces a significant computational complexity associated with matrix inversion. Despite numerous efforts to alleviate this issue, existing methods are often limited to specifically structured sparse signals. This paper aims to provide a universal inversion-free approach to accelerate existing SBL algorithms. We unify the optimization of SBL variants with different priors within the expectation-maximization (EM) framework, where a lower bound of the likelihood function is maximized. Due to the linear Gaussian model foundation of SBL, updating this lower bound requires maximizing a quadratic function, which involves matrix inversion. Thus, we employ the minorization-maximization (MM) framework to derive two novel lower bounds that diagonalize the quadratic coefficient matrix, thereby eliminating the need for any matrix inversions. We further investigate their properties, including convergence guarantees under the MM framework and the slow convergence rate due to reduced curvature. The proposed approach is applicable to various types of structured sparse signals, such as row-sparse, block-sparse, and burst-sparse signals. Our simulations on synthetic and real data demonstrate remarkably shorter running time compared to state-of-the-art methods while achieving comparable recovery performance.
稀疏贝叶斯学习（SBL）因其卓越的性能，已成为稀疏信号恢复的有力工具。然而，SBL 的实际应用面临着与矩阵反演相关的巨大计算复杂性。尽管为缓解这一问题做出了许多努力，但现有方法往往局限于特定结构的稀疏信号。本文旨在提供一种通用的免反演方法，以加速现有的 SBL 算法。我们在期望最大化（EM）框架内统一了具有不同先验的 SBL 变体的优化，其中最大化了似然函数的下限。由于 SBL 以线性高斯模型为基础，更新该下限需要最大化二次函数，其中涉及矩阵反转。因此，我们采用最小化-最大化（MM）框架，推导出两个新的下界，将二次系数矩阵对角化，从而消除了任何矩阵反转的需要。我们进一步研究了它们的特性，包括 MM 框架下的收敛保证以及曲率减小导致的收敛速度缓慢。所提出的方法适用于各种类型的结构稀疏信号，如行稀疏、块稀疏和突发稀疏信号。我们在合成数据和真实数据上进行的仿真表明，与最先进的方法相比，该方法的运行时间显著缩短，同时恢复性能相当。
{"title":"Towards Inversion-Free Sparse Bayesian Learning: A Universal Approach","authors":"Yuhui Song;Zijun Gong;Yuanzhu Chen;Cheng Li","doi":"10.1109/TSP.2024.3484908","DOIUrl":"10.1109/TSP.2024.3484908","url":null,"abstract":"Sparse Bayesian Learning (SBL) has emerged as a powerful tool for sparse signal recovery, due to its superior performance. However, the practical implementation of SBL faces a significant computational complexity associated with matrix inversion. Despite numerous efforts to alleviate this issue, existing methods are often limited to specifically structured sparse signals. This paper aims to provide a universal inversion-free approach to accelerate existing SBL algorithms. We unify the optimization of SBL variants with different priors within the expectation-maximization (EM) framework, where a lower bound of the likelihood function is maximized. Due to the linear Gaussian model foundation of SBL, updating this lower bound requires maximizing a quadratic function, which involves matrix inversion. Thus, we employ the minorization-maximization (MM) framework to derive two novel lower bounds that diagonalize the quadratic coefficient matrix, thereby eliminating the need for any matrix inversions. We further investigate their properties, including convergence guarantees under the MM framework and the slow convergence rate due to reduced curvature. The proposed approach is applicable to various types of structured sparse signals, such as row-sparse, block-sparse, and burst-sparse signals. Our simulations on synthetic and real data demonstrate remarkably shorter running time compared to state-of-the-art methods while achieving comparable recovery performance.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"72 ","pages":"4992-5006"},"PeriodicalIF":4.6,"publicationDate":"2024-10-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142488489","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}
Widely linear (WL) processing has been of great interest these last two decades for multi-user (MUI) interference mitigation in radiocommunications networks using rectilinear (R) or quasi-rectilinear (QR) signals in particular. Despite numerous papers on the subject, this topic remains of interest for several current and future applications which use R or QR signals, described hereafter. In this context, using a continuous time approach, it is shown in this paper the sub-optimality of most of the WL MMSE receivers of the literature, which are implemented at the symbol rate after a matched filtering operation to the pulse shaping filter, and the necessity to know the MUI channels, always cumbersome in practice, to implement the optimal WL MMSE receiver. Then, the main challenge addressed in the paper is to propose new WL MMSE receivers able to implement the optimal one without requiring the MUI channels knowledge. For this purpose, two new WL MMSE receivers, a two-input one and a three-input one, are proposed and analyzed in this paper for R and QR signals corrupted by data-like MUI. The two-input and three-input receivers are shown to be quasi-optimal respectively for R signals using Square Root Raised Cosine (SRRC) filters with a low roll-off and for R and QR signals whatever the pulse shaping filter, showing in particular the non-equivalence of R and QR signals for WL MMSE receivers. These two new receivers open new perspectives for the implementation of the optimal WL MMSE receiver in the presence of data-like MUI from the only knowledge of the SOI channel.
在过去的二十年里，宽线性（WL）处理技术一直是无线电通信网络多用户（MUI）干扰缓解的重要研究课题，尤其是在使用直线性（R）或准直线性（QR）信号的无线电通信网络中。尽管有关这一主题的论文不胜枚举，但这一主题对于当前和未来使用 R 或 QR 信号的几种应用仍具有重要意义，下文将对此进行介绍。在此背景下，本文采用连续时间方法，展示了文献中大多数 WL MMSE 接收机的次优性，这些接收机是在对脉冲整形滤波器进行匹配滤波操作后，在符号率上实现的，而且必须知道 MUI 信道（在实践中总是很麻烦），才能实现最佳 WL MMSE 接收机。因此，本文要解决的主要难题是提出新的 WL MMSE 接收机，无需 MUI 信道知识即可实现最佳接收机。为此，本文针对被数据类 MUI 破坏的 R 和 QR 信号，提出并分析了两种新的 WL MMSE 接收机，一种是双输入接收机，另一种是三输入接收机。结果表明，对于使用低滚降平方根余弦（SRRC）滤波器的 R 信号，以及使用任何脉冲整形滤波器的 R 和 QR 信号，两输入接收器和三输入接收器分别是准最优的，这尤其表明了 R 和 QR 信号对于 WL MMSE 接收器的非等效性。这两种新型接收器为在仅了解 SOI 信道的情况下，在存在类数据 MUI 的情况下实现最佳 WL MMSE 接收器开辟了新的前景。
{"title":"New Insights Into Widely Linear MMSE Receivers for Communication Networks Using Data-Like Rectilinear or Quasi-Rectilinear Signals","authors":"Pascal Chevalier;Jean-Pierre Delmas;Roger Lamberti","doi":"10.1109/TSP.2024.3479875","DOIUrl":"10.1109/TSP.2024.3479875","url":null,"abstract":"Widely linear (WL) processing has been of great interest these last two decades for multi-user (MUI) interference mitigation in radiocommunications networks using rectilinear (R) or quasi-rectilinear (QR) signals in particular. Despite numerous papers on the subject, this topic remains of interest for several current and future applications which use R or QR signals, described hereafter. In this context, using a continuous time approach, it is shown in this paper the sub-optimality of most of the WL MMSE receivers of the literature, which are implemented at the symbol rate after a matched filtering operation to the pulse shaping filter, and the necessity to know the MUI channels, always cumbersome in practice, to implement the optimal WL MMSE receiver. Then, the main challenge addressed in the paper is to propose new WL MMSE receivers able to implement the optimal one without requiring the MUI channels knowledge. For this purpose, two new WL MMSE receivers, a two-input one and a three-input one, are proposed and analyzed in this paper for R and QR signals corrupted by data-like MUI. The two-input and three-input receivers are shown to be quasi-optimal respectively for R signals using Square Root Raised Cosine (SRRC) filters with a low roll-off and for R and QR signals whatever the pulse shaping filter, showing in particular the non-equivalence of R and QR signals for WL MMSE receivers. These two new receivers open new perspectives for the implementation of the optimal WL MMSE receiver in the presence of data-like MUI from the only knowledge of the SOI channel.","PeriodicalId":13330,"journal":{"name":"IEEE Transactions on Signal Processing","volume":"72 ","pages":"5156-5173"},"PeriodicalIF":4.6,"publicationDate":"2024-10-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142488491","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：