Chengzhu Yang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Chengzhu Yang has a total of 17 co-authors. Here are the top co-authors by number of shared publications:
- Desheng Chen
Beijing Institute of Technology
4 shared publications - Hing Cheung So
City University of Hong Kong
4 shared publications - Tao Tang
School of Information and Electronics, Beijing Institute of Technology, Beijing, China
4 shared publications - Yuantao Gu
Department of Electronic Engineering, Beijing National Research Center for Information Science and Technology, Tsinghua University, Beijing, China
4 shared publications - Badong Chen
Xi’an Jiaotong University
3 shared publications - Hongbing Ma
Tsinghua University
3 shared publications
Latest Publications
- 23 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Aerospace and Electronic Systems
- Vol. 62 (00189251) , 5066-5083
A Sparse Bayesian Learning-Based Approach With Indian Buffet Process Prior for Joint Wideband DOA and Frequency Band Estimation
Sparse Bayesian learning (SBL)-based methods for wideband direction of arrival (DOA) estimation have shown impressive performance in terms of high-resolution. It generally assumes that all signals share the same frequency band, resulting in performance degradation when signals occupy different frequency bands. To deal with this problem, we propose a wideband DOA estimation method utilizing the Indian Buffet Process (IBP) prior, namely IBP-SBL, to explore the frequency band correlation structure of target signals and hence their DOAs. Specifically, IBP-SBL regards the spatial spectrum as the combination of multiple latent features to be estimated, and infers the occupied frequency band of each signal by exploring the activation of the latent features at each frequency point. Subsequently, the DOA refinement procedure based on maximum likelihood (ML) is applied to reduce the quantization error caused by grid mismatch problem. Compared to the previous methods, the proposed algorithm can associate each sub-band with the relevant signals even in scenarios with severe overlap of signal frequency bands, thereby realizing more accurate DOA estimation by using the estimated frequency band related to each signal. Numerical simulation results of 500 Monte Carlo experiments show that our method achieves lower DOA estimation error and convergence time. In real-world data experiments, our method also exhibits superior DOA estimation accuracy and clearer target trajectory.
- 23 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Aerospace and Electronic Systems
- Vol. 62 (00189251) , 5066-5083
A Sparse Bayesian Learning-Based Approach With Indian Buffet Process Prior for Joint Wideband DOA and Frequency Band Estimation
Sparse Bayesian learning (SBL)-based methods for wideband direction of arrival (DOA) estimation have shown impressive performance in terms of high-resolution. It generally assumes that all signals share the same frequency band, resulting in performance degradation when signals occupy different frequency bands. To deal with this problem, we propose a wideband DOA estimation method utilizing the Indian Buffet Process (IBP) prior, namely IBP-SBL, to explore the frequency band correlation structure of target signals and hence their DOAs. Specifically, IBP-SBL regards the spatial spectrum as the combination of multiple latent features to be estimated, and infers the occupied frequency band of each signal by exploring the activation of the latent features at each frequency point. Subsequently, the DOA refinement procedure based on maximum likelihood (ML) is applied to reduce the quantization error caused by grid mismatch problem. Compared to the previous methods, the proposed algorithm can associate each sub-band with the relevant signals even in scenarios with severe overlap of signal frequency bands, thereby realizing more accurate DOA estimation by using the estimated frequency band related to each signal. Numerical simulation results of 500 Monte Carlo experiments show that our method achieves lower DOA estimation error and convergence time. In real-world data experiments, our method also exhibits superior DOA estimation accuracy and clearer target trajectory.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75, 9704624
Adaptive Space-Frequency Basis Selection Methods for Signal Frequency Band Sensing and Fast Wideband DOA Estimation With Passive Sonar Data Verification
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75, 9704624
Adaptive Space-Frequency Basis Selection Methods for Signal Frequency Band Sensing and Fast Wideband DOA Estimation With Passive Sonar Data Verification
- 19 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Aerospace and Electronic Systems
- Vol. 61 (1) , 1104-1119
Joint Localization and Source Association Sparse Bayesian Learning Under Multipath Propagation
This paper focuses on the topic of joint direction of arrival (DOA), source association and attenuation coefficient estimation under multipath environment. Most existing methods adopt the sequential three-phase estimation, resulting in the nuisance dependency between the estimation accuracy of the current phase and the previous phase. Besides, they also require some accurate prior information, including the accurate DOA initialization, the number of incoherent sources and spatial paths, which is unrealistic in practice. To solve this problem, the joint localization and source association sparse Bayesian learning (JLSA-SBL) algorithm is proposed to integrate the source association process, DOA and attenuation coefficient estimation into a unified parameter estimation framework. The proposed method exploits the underlying sparsity and coherent structure of the incident signals to achieve more accurate joint parameter estimation. Compared to the previous methods, JLSA-SBL can directly estimate the latent multipath propagation parameters even in the absence of prior information. Besides, the JLSA-SBL also has superior performance in distinguishing the closely spaced multipath signals belonging to different sources. Numerical simulation experiments have been performed to demonstrate the superior performance of the proposed method.
- 19 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Aerospace and Electronic Systems
- Vol. 61 (1) , 1104-1119
Joint Localization and Source Association Sparse Bayesian Learning Under Multipath Propagation
This paper focuses on the topic of joint direction of arrival (DOA), source association and attenuation coefficient estimation under multipath environment. Most existing methods adopt the sequential three-phase estimation, resulting in the nuisance dependency between the estimation accuracy of the current phase and the previous phase. Besides, they also require some accurate prior information, including the accurate DOA initialization, the number of incoherent sources and spatial paths, which is unrealistic in practice. To solve this problem, the joint localization and source association sparse Bayesian learning (JLSA-SBL) algorithm is proposed to integrate the source association process, DOA and attenuation coefficient estimation into a unified parameter estimation framework. The proposed method exploits the underlying sparsity and coherent structure of the incident signals to achieve more accurate joint parameter estimation. Compared to the previous methods, JLSA-SBL can directly estimate the latent multipath propagation parameters even in the absence of prior information. Besides, the JLSA-SBL also has superior performance in distinguishing the closely spaced multipath signals belonging to different sources. Numerical simulation experiments have been performed to demonstrate the superior performance of the proposed method.
- 8 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 21 (1545598X) , 1-5
FBC-SBL: Frequency Band Clustering Sparse Bayesian Learning for Off-Grid Wideband DOA Estimation With Different Frequency Bands
Direction of arrival (DOA) estimation methods for wideband signals based on sparse Bayesian learning (SBL) have shown impressive performance. It generally assumes that wideband signals share the same frequency band, whereas deteriorating the performance of DOA estimation when target signals locate at different bands. To deal with this problem, we propose an off-grid DOA estimation framework, namely FBC-SBL, based on SBL incorporating frequency band clustering (FBC). FBC can exploit the cluster structure of signal distribution in frequency domain by applying a specific clustering algorithm. Compared to other existing algorithms, FBC-SBL can infer the frequency band occupation thereby estimating DOAs more accurately. Specifically, the whole frequency band is firstly divided into finer subbands and the variance vector of each subband is estimated. Then, FBC gathers the subbands with similar signal distribution. Finally, the estimated DOAs can be obtained by analyzing each cluster through SBL framework. Numerical simulations and real data experiments show that our method has superior performance compared to other algorithms, and achieves optimal average absolute error of 1.6394 degree in real data experiments.
- 8 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 21 (1545598X) , 1-5
FBC-SBL: Frequency Band Clustering Sparse Bayesian Learning for Off-Grid Wideband DOA Estimation With Different Frequency Bands
Direction of arrival (DOA) estimation methods for wideband signals based on sparse Bayesian learning (SBL) have shown impressive performance. It generally assumes that wideband signals share the same frequency band, whereas deteriorating the performance of DOA estimation when target signals locate at different bands. To deal with this problem, we propose an off-grid DOA estimation framework, namely FBC-SBL, based on SBL incorporating frequency band clustering (FBC). FBC can exploit the cluster structure of signal distribution in frequency domain by applying a specific clustering algorithm. Compared to other existing algorithms, FBC-SBL can infer the frequency band occupation thereby estimating DOAs more accurately. Specifically, the whole frequency band is firstly divided into finer subbands and the variance vector of each subband is estimated. Then, FBC gathers the subbands with similar signal distribution. Finally, the estimated DOAs can be obtained by analyzing each cluster through SBL framework. Numerical simulations and real data experiments show that our method has superior performance compared to other algorithms, and achieves optimal average absolute error of 1.6394 degree in real data experiments.
- 18 September 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 68 (4) , 1492-1496
2-D Learned Proximal Gradient Algorithm for Fast Sparse Matrix Recovery
Many real-world problems can be modeled as sparse matrix recovery from two-dimensional (2D) measurements, which is recognized as one of the most important topics in signal processing community. Benefited from the roaring success of compressed sensing, many classical iterative algorithms can be directly applied or reinvented for matrix recovery, though they are computationally expensive. To alleviate this, we propose a neural network named 2D learned proximal gradient algorithm (2D-LPGA), which aims to quickly reconstruct the target matrix. Theoretical analysis reveals that if the parameters of the network satisfy certain conditions, it can reconstruct the sparse signal with linear convergence rate. Moreover, numerical experiments demonstrate the superiority of the proposed method over other classical schemes.
- 18 September 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems II: Express Briefs
- Vol. 68 (4) , 1492-1496
2-D Learned Proximal Gradient Algorithm for Fast Sparse Matrix Recovery
Many real-world problems can be modeled as sparse matrix recovery from two-dimensional (2D) measurements, which is recognized as one of the most important topics in signal processing community. Benefited from the roaring success of compressed sensing, many classical iterative algorithms can be directly applied or reinvented for matrix recovery, though they are computationally expensive. To alleviate this, we propose a neural network named 2D learned proximal gradient algorithm (2D-LPGA), which aims to quickly reconstruct the target matrix. Theoretical analysis reveals that if the parameters of the network satisfy certain conditions, it can reconstruct the sparse signal with linear convergence rate. Moreover, numerical experiments demonstrate the superiority of the proposed method over other classical schemes.
- 5 March 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Signal Processing
- Vol. 68 (1053587X) , 5244-5259
Learning Proximal Operator Methods for Nonconvex Sparse Recovery with Theoretical Guarantee
Sparse recovery has attracted considerable attention in signal processing community these years, because of its widespread usage in many applications. Though lots of convex and nonconvex methods have been proposed for this topic, most of them are computationally expensive. Recently, learned iterative shrinkage-thresholding algorithm (LISTA) and its variants are incorporated to sparse recovery, which are based on specific deep neural networks designed by unfolding ISTA. However, the problem they are able to solve is regularized by ${\ell_1}$ norm, which is less effective at promoting sparseness than some nonconvex sparsity-inducing penalties (SIPs). Motivated by the fact that the problems regularized by these SIPs can be solved by proximal operator methods, we devise a network named learning proximal operator method (LePOM) for sparse recovery. For LePOM on a general class of SIPs, a necessary condition of its convergence is established. Based on this condition, a simplified network is proposed with less parameters. Theoretical analysis of this network inspires us to further reduce the number of parameters and arrive at proposing Analytical LePOM (ALePOM). ALePOM determines most parameters by solving an optimization problem and significantly reduces the number of parameters. Theoretical analysis shows if the signal is sufficiently sparse, ALePOM converges linearly. Simulations confirm our analyses and demonstrate that the proposed solutions outperform state-of-the-art sparse recovery algorithms and neural network-based methods.- 5 March 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Signal Processing
- Vol. 68 (1053587X) , 5244-5259
Learning Proximal Operator Methods for Nonconvex Sparse Recovery with Theoretical Guarantee
Sparse recovery has attracted considerable attention in signal processing community these years, because of its widespread usage in many applications. Though lots of convex and nonconvex methods have been proposed for this topic, most of them are computationally expensive. Recently, learned iterative shrinkage-thresholding algorithm (LISTA) and its variants are incorporated to sparse recovery, which are based on specific deep neural networks designed by unfolding ISTA. However, the problem they are able to solve is regularized by ${\ell_1}$ norm, which is less effective at promoting sparseness than some nonconvex sparsity-inducing penalties (SIPs). Motivated by the fact that the problems regularized by these SIPs can be solved by proximal operator methods, we devise a network named learning proximal operator method (LePOM) for sparse recovery. For LePOM on a general class of SIPs, a necessary condition of its convergence is established. Based on this condition, a simplified network is proposed with less parameters. Theoretical analysis of this network inspires us to further reduce the number of parameters and arrive at proposing Analytical LePOM (ALePOM). ALePOM determines most parameters by solving an optimization problem and significantly reduces the number of parameters. Theoretical analysis shows if the signal is sufficiently sparse, ALePOM converges linearly. Simulations confirm our analyses and demonstrate that the proposed solutions outperform state-of-the-art sparse recovery algorithms and neural network-based methods.- 19 August 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Signal Processing
- Vol. 67 (19) , 5046-5061
Weakly Convex Regularized Robust Sparse Recovery Methods With Theoretical Guarantees
Robust sparse signal recovery against impulsive noise is a core issue in many applications. Numerous methods have been proposed to recover the sparse signal from measurements corrupted by various impulsive noises, but most of them either lack theoretical guarantee for robust sparse recovery or are not efficient enough for large-scale problems. To this end, a general optimization problem for robust sparse signal recovery, which includes many existing works as concrete instances, is analyzed by a freshly defined Double Null Space Property (DNSP), and its solution is proved to be able to robustly reconstruct the sparse signal under mild conditions. Moreover, for computational tractability, weakly convex sparsity-inducing penalties are applied to the general problem, and properties of the solution to the resultant non-convex problem are further studied. Based on these properties, an algorithm named Robust Projected Generalized Gradient (RPGG) is devised to solve the weakly convex problem. Theoretical results prove that the sparse signal can be precisely reconstructed by RPGG from compressive measurements with sparse noise or robustly recovered from those with impulsive noise. Meanwhile, simulations demonstrate that RPGG with tuned parameters outperforms other robust sparse recovery algorithms.
- 19 August 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Signal Processing
- Vol. 67 (19) , 5046-5061
Weakly Convex Regularized Robust Sparse Recovery Methods With Theoretical Guarantees
Robust sparse signal recovery against impulsive noise is a core issue in many applications. Numerous methods have been proposed to recover the sparse signal from measurements corrupted by various impulsive noises, but most of them either lack theoretical guarantee for robust sparse recovery or are not efficient enough for large-scale problems. To this end, a general optimization problem for robust sparse signal recovery, which includes many existing works as concrete instances, is analyzed by a freshly defined Double Null Space Property (DNSP), and its solution is proved to be able to robustly reconstruct the sparse signal under mild conditions. Moreover, for computational tractability, weakly convex sparsity-inducing penalties are applied to the general problem, and properties of the solution to the resultant non-convex problem are further studied. Based on these properties, an algorithm named Robust Projected Generalized Gradient (RPGG) is devised to solve the weakly convex problem. Theoretical results prove that the sparse signal can be precisely reconstructed by RPGG from compressive measurements with sparse noise or robustly recovered from those with impulsive noise. Meanwhile, simulations demonstrate that RPGG with tuned parameters outperforms other robust sparse recovery algorithms.
- 1 November 2018
- Published by Elsevier in Signal Processing
- Vol. 152, 84-89
Robust sparse recovery via weakly convex optimization in impulsive noise
- 1 November 2018
- Published by Elsevier in Signal Processing
- Vol. 152, 84-89
Robust sparse recovery via weakly convex optimization in impulsive noise
- 2 August 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Signal Processing
- Vol. 66 (19) , 5014-5028
Sparse Recovery Conditions and Performance Bounds for $\ell _p$-Minimization
In sparse recovery, a sparse signal ${\mathbf x}\in \mathbb {R}^N$ with $K$ nonzero entries is to be reconstructed from a compressed measurement $\mathbf y=Ax$ with ${\mathbf A}\in \mathbb {R}^{M\times N}$ ( $M< N$ ). The $\ell _p$ $(0\leq p < 1)$ pseudonorm has been found to be a sparsity inducing function superior to the $\ell _1$ norm, and the null space constant (NSC) and restricted isometry constant (RIC) have been used as key notions in the performance analyses of the corresponding $\ell _p$ -minimization. In this paper, we study sparse recovery conditions and performance bounds for the $\ell _p$ -minimization. We devise a new NSC upper bound that outperforms the state-of-the-art result. Based on the improved NSC upper bound, we provide a new RIC upper bound dependent on the sparsity level $K$ as a sufficient condition for precise recovery, and it is tighter than the existing bound for small $K$ . Then, we study the largest choice of $p$ for the $\ell _p$ -minimization problem to recover any $K$ -sparse signal, and the largest recoverable $K$ for a fixed $p$ . Numerical experiments demonstrate the improvement of the proposed bounds in the recovery conditions over the up-to-date counterparts.
- 2 August 2018
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Signal Processing
- Vol. 66 (19) , 5014-5028
Sparse Recovery Conditions and Performance Bounds for $\ell _p$-Minimization
In sparse recovery, a sparse signal ${\mathbf x}\in \mathbb {R}^N$ with $K$ nonzero entries is to be reconstructed from a compressed measurement $\mathbf y=Ax$ with ${\mathbf A}\in \mathbb {R}^{M\times N}$ ( $M< N$ ). The $\ell _p$ $(0\leq p < 1)$ pseudonorm has been found to be a sparsity inducing function superior to the $\ell _1$ norm, and the null space constant (NSC) and restricted isometry constant (RIC) have been used as key notions in the performance analyses of the corresponding $\ell _p$ -minimization. In this paper, we study sparse recovery conditions and performance bounds for the $\ell _p$ -minimization. We devise a new NSC upper bound that outperforms the state-of-the-art result. Based on the improved NSC upper bound, we provide a new RIC upper bound dependent on the sparsity level $K$ as a sufficient condition for precise recovery, and it is tighter than the existing bound for small $K$ . Then, we study the largest choice of $p$ for the $\ell _p$ -minimization problem to recover any $K$ -sparse signal, and the largest recoverable $K$ for a fixed $p$ . Numerical experiments demonstrate the improvement of the proposed bounds in the recovery conditions over the up-to-date counterparts.