Lei Cheng
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
Lei Cheng has a total of 121 co-authors. Here are the top co-authors by number of shared publications:
- Yik-Chung Wu
Department of Electrical and Electronic Engineering, The University of Hong Kong, Pokfulam, Hong Kong
16 shared publications - Jianlong Li
College of Information Science and Electronic Engineering, Zhejiang University 1 , Hangzhou 310027
13 shared publications Yik-Chung Wu
12 shared publications- Ting Zhang
College of Information Science and Electronic Engineering, Zhejiang University 1 , Hangzhou 310027
11 shared publications Zhongtao Chen
The University of Hong Kong
11 shared publicationsSiyuan Li
College of Information Science and Electronic Engineering, Zhejiang University 1 , Hangzhou, 310027
7 shared publications
Latest Publications
- 1 September 2026
- Published by Acoustical Society of America (ASA) in The Journal of the Acoustical Society of America
- Vol. 160  (3) , 2356-2369
Dynamic mode decomposition by tensor evolution for source depth
Depth estimation of shallow acoustic sources using a deep-ocean near-bottom vertical line array relies on depth-sensitive interference, which appears as an oscillatory structure in the frequency-angle domain in broadband matched beam-intensity processing (MBIP). For low signal-to-noise ratio, the nonlinear feature extraction in MBIP transforms array-domain noise into feature-domain distortions, making depth estimation fragile. Addressing this issue, we propose a tensor evolution-based depth estimation method inspired by dynamic mode decomposition, which interprets a sequence of broadband MBIP surfaces as the output of a dynamic system from a time-varying autoregressive operator. The operator is represented by a low-rank Tucker decomposition, which separates coherent feature modes and their temporal modes from incoherent noise. The target depth is inferred via Fourier summation applied to a dominant feature mode. Deep-ocean numerical experiments demonstrate that tensor evolution-based depth estimation improves robustness at low signal-to-noise ratio, consistently outperforming matched field processing and broadband MBIP.
- 1 September 2026
- Published by Acoustical Society of America (ASA) in The Journal of the Acoustical Society of America
- Vol. 160  (3) , 2356-2369
Dynamic mode decomposition by tensor evolution for source depth
Depth estimation of shallow acoustic sources using a deep-ocean near-bottom vertical line array relies on depth-sensitive interference, which appears as an oscillatory structure in the frequency-angle domain in broadband matched beam-intensity processing (MBIP). For low signal-to-noise ratio, the nonlinear feature extraction in MBIP transforms array-domain noise into feature-domain distortions, making depth estimation fragile. Addressing this issue, we propose a tensor evolution-based depth estimation method inspired by dynamic mode decomposition, which interprets a sequence of broadband MBIP surfaces as the output of a dynamic system from a time-varying autoregressive operator. The operator is represented by a low-rank Tucker decomposition, which separates coherent feature modes and their temporal modes from incoherent noise. The target depth is inferred via Fourier summation applied to a dominant feature mode. Deep-ocean numerical experiments demonstrate that tensor evolution-based depth estimation improves robustness at low signal-to-noise ratio, consistently outperforming matched field processing and broadband MBIP.
- Open Access
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 12054-12065
GRASP: Manifold-Aware Structural Guidance for Generative Atmospheric Super-Resolution
- 8 August 2026
- Published by Association for Computing Machinery (ACM)
- p. 12054-12065
Open AccessGRASP: Manifold-Aware Structural Guidance for Generative Atmospheric Super-Resolution
- 1 May 2026
- Published by Acoustical Society of America (ASA) in The Journal of the Acoustical Society of America
- Vol. 159  (5) , 4204-4222
Continuous forecasting of range-dependent ocean sound speed field: Diffusion model meets multi-output Gaussian process
Environment-aware underwater acoustic detection and communication require precise forecasting of the range-dependent sound speed field (SSF) at any given time. Recently, methods such as Gaussian process regression (GPR) and conditional diffusion models have shown advanced performance in SSF forecasting. However, limitations remain: standard GPR fails to capture the range-depth spatial correlations, and conditional diffusion models struggle with continuous forecasting. To address these issues, we integrate multi-output GPR and conditional diffusion models to enable continuous forecasting of range-dependent SSFs, employing careful designs for diffusion noise, neural architecture, and training strategies. Our experiments, conducted on HYCOM hindcast datasets from the South China Sea, demonstrate that our proposed model outperforms state-of-the-art baselines in forecasting range-dependent SSFs at any given time and the associated underwater transmission losses.
- 1 May 2026
- Published by Acoustical Society of America (ASA) in The Journal of the Acoustical Society of America
- Vol. 159  (5) , 4204-4222
Continuous forecasting of range-dependent ocean sound speed field: Diffusion model meets multi-output Gaussian process
Environment-aware underwater acoustic detection and communication require precise forecasting of the range-dependent sound speed field (SSF) at any given time. Recently, methods such as Gaussian process regression (GPR) and conditional diffusion models have shown advanced performance in SSF forecasting. However, limitations remain: standard GPR fails to capture the range-depth spatial correlations, and conditional diffusion models struggle with continuous forecasting. To address these issues, we integrate multi-output GPR and conditional diffusion models to enable continuous forecasting of range-dependent SSFs, employing careful designs for diffusion noise, neural architecture, and training strategies. Our experiments, conducted on HYCOM hindcast datasets from the South China Sea, demonstrate that our proposed model outperforms state-of-the-art baselines in forecasting range-dependent SSFs at any given time and the associated underwater transmission losses.
- Open Access
- 1 March 2026
- Published by Acoustical Society of America (ASA) in The Journal of the Acoustical Society of America
- Vol. 159  (3) , 2670-2684
Direction-of-arrival estimation for weak underwater targets via flexible sparsity-aware modeling
Direction-of-arrival (DOA) estimation plays a critical role in underwater acoustic applications such as target detection, localization, tracking, and identification. Unlike image or speech processing, underwater array signal processing faces unique challenges because of low signal-to-noise ratio (SNR) and limited snapshots. In such conditions, whether traditional methods (e.g., conventional beamforming, multiple signal classification) or learning-based approaches (e.g., sparse Bayesian learning) suffer from a degradation in accuracy and resolution of DOA estimation. To improve performance at low SNR and limited observations, we propose a Bayesian probabilistic model leveraging the flexible sparsity-promoting properties of the generalized hyperbolic distribution. Within this framework, the closed-form inference updates are derived using the variational inference technique. Furthermore, to mitigate grid mismatch, we introduce a modified marginal likelihood computation, based on which an efficient adaptive approach is developed for refined DOA estimation. Numerical results, including simulation and SWellEx-96 Experiment data, validate the superior performance of the proposed method, especially in the low SNR and limited snapshots regime.
- 1 March 2026
- Published by Acoustical Society of America (ASA) in The Journal of the Acoustical Society of America
- Vol. 159  (3) , 2670-2684
Open AccessDirection-of-arrival estimation for weak underwater targets via flexible sparsity-aware modeling
Direction-of-arrival (DOA) estimation plays a critical role in underwater acoustic applications such as target detection, localization, tracking, and identification. Unlike image or speech processing, underwater array signal processing faces unique challenges because of low signal-to-noise ratio (SNR) and limited snapshots. In such conditions, whether traditional methods (e.g., conventional beamforming, multiple signal classification) or learning-based approaches (e.g., sparse Bayesian learning) suffer from a degradation in accuracy and resolution of DOA estimation. To improve performance at low SNR and limited observations, we propose a Bayesian probabilistic model leveraging the flexible sparsity-promoting properties of the generalized hyperbolic distribution. Within this framework, the closed-form inference updates are derived using the variational inference technique. Furthermore, to mitigate grid mismatch, we introduce a modified marginal likelihood computation, based on which an efficient adaptive approach is developed for refined DOA estimation. Numerical results, including simulation and SWellEx-96 Experiment data, validate the superior performance of the proposed method, especially in the low SNR and limited snapshots regime.
- 23 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Signal Processing
- Vol. 74  (1053587X) , 654-669
Rank-Revealing Bayesian Block-Term Tensor Completion With Graph Information
Block-term decomposition (BTD), particularly its rank-(Lr,Lr, 1) special case, is widely used in signal processing. Traditional methods for computing BTD either unrealistically assume the number of blocks and block ranks are known or require exhaustive tuning of these parameters. While sparsity-promoting regularization has been introduced to estimate these parameters more efficiently, it still requires regularization parameter tuning. Bayesian learning addresses these issues by employing sparsity-promoting priors on the number of blocks and block ranks, but so far is limited to fully observed BTD tensors. To process incomplete BTD tensors, only a few optimization-based methods have been proposed, and they continue to suffer from heavy parameter tuning. To enable tuning-free BTD completion, a prior that simultaneously enforces block-wise sparsity and withinblock column-wise sparsity while incorporating graph structure is introduced within the Bayesian framework. Besides theoretically establishing the legitimacy of the prior distribution, a mean-field design is developed to obtain a closed-form updating variational inference (VI) algorithm without loss of graph information. Extensive experiments on both synthetic datasets and real-world datasets demonstrate the superiority of the proposed method over existing optimization-based algorithms and the Bayesian model without graph information, in terms of rank learning, tensor recovery, and factor recovery.
- 23 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Signal Processing
- Vol. 74  (1053587X) , 654-669
Rank-Revealing Bayesian Block-Term Tensor Completion With Graph Information
Block-term decomposition (BTD), particularly its rank-(Lr,Lr, 1) special case, is widely used in signal processing. Traditional methods for computing BTD either unrealistically assume the number of blocks and block ranks are known or require exhaustive tuning of these parameters. While sparsity-promoting regularization has been introduced to estimate these parameters more efficiently, it still requires regularization parameter tuning. Bayesian learning addresses these issues by employing sparsity-promoting priors on the number of blocks and block ranks, but so far is limited to fully observed BTD tensors. To process incomplete BTD tensors, only a few optimization-based methods have been proposed, and they continue to suffer from heavy parameter tuning. To enable tuning-free BTD completion, a prior that simultaneously enforces block-wise sparsity and withinblock column-wise sparsity while incorporating graph structure is introduced within the Bayesian framework. Besides theoretically establishing the legitimacy of the prior distribution, a mean-field design is developed to obtain a closed-form updating variational inference (VI) algorithm without loss of graph information. Extensive experiments on both synthetic datasets and real-world datasets demonstrate the superiority of the proposed method over existing optimization-based algorithms and the Bayesian model without graph information, in terms of rank learning, tensor recovery, and factor recovery.
- 5 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Signal Processing
- Vol. PP  (1053587X) , 1-14
Radio Map Estimation via Latent Domain Plug-and-Play Denoising
Radio map estimation (RME), also known as spectrum cartography, aims to reconstruct the strength of radio interference across different domains (e.g., space and frequency) from sparsely sampled measurements. To tackle this typical inverse problem, state-of-the-art RME methods rely on handcrafted or data-driven structural information of radio maps. However, the former often struggles to model complex radio frequency (RF) environments and the latter requires excessive trainingßžmaking it hard to quickly adapt to in situ sensing tasks. This work presents an RME approach based on plug-and-play (PnP) denoising, a technique from computational imaging. The idea is to leverage the observation that the denoising operations of signals like natural images and radio maps are similarßždespite the nontrivial differences of the signals themselves. Hence, sophisticated denoisers designed for or learned from natural images can be directly employed to assist RME, avoiding using radio map data for training. Unlike conventional PnP methods that operate directly in the data domain, the proposed method exploits the classical spatio-spectral decomposition model of radio maps and proposes an ADMM algorithm that denoises in the latent factor that only represents the spatial domain. This design significantly improves computational efficiency and enhances noise robustness. Theoretical aspects, e.g., recoverability of the complete radio map and convergence of the ADMM algorithm are analyzed. Synthetic and real data experiments are conducted to demonstrate the effectiveness of our approach.
- 5 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Signal Processing
- Vol. PP  (1053587X) , 1-14
Radio Map Estimation via Latent Domain Plug-and-Play Denoising
Radio map estimation (RME), also known as spectrum cartography, aims to reconstruct the strength of radio interference across different domains (e.g., space and frequency) from sparsely sampled measurements. To tackle this typical inverse problem, state-of-the-art RME methods rely on handcrafted or data-driven structural information of radio maps. However, the former often struggles to model complex radio frequency (RF) environments and the latter requires excessive trainingßžmaking it hard to quickly adapt to in situ sensing tasks. This work presents an RME approach based on plug-and-play (PnP) denoising, a technique from computational imaging. The idea is to leverage the observation that the denoising operations of signals like natural images and radio maps are similarßždespite the nontrivial differences of the signals themselves. Hence, sophisticated denoisers designed for or learned from natural images can be directly employed to assist RME, avoiding using radio map data for training. Unlike conventional PnP methods that operate directly in the data domain, the proposed method exploits the classical spatio-spectral decomposition model of radio maps and proposes an ADMM algorithm that denoises in the latent factor that only represents the spatial domain. This design significantly improves computational efficiency and enhances noise robustness. Theoretical aspects, e.g., recoverability of the complete radio map and convergence of the ADMM algorithm are analyzed. Synthetic and real data experiments are conducted to demonstrate the effectiveness of our approach.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 33, 4311-4315
Spatio-Temporal Field-Modulated Fourier Neural Operator for Transmission Loss Estimation
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 33, 4311-4315
Spatio-Temporal Field-Modulated Fourier Neural Operator for Transmission Loss Estimation
- 2 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Signal Processing
- Vol. 73  (1053587X) , 5319-5335
When Bayesian Tensor Completion Meets Multioutput Gaussian Processes: Functional Universality and Rank Learning
Functional tensor decomposition can analyze multidimensional data with real-valued indices, paving the path for applications in machine learning and signal processing. A limitation of existing approaches is the assumption that the tensor rank—a critical parameter governing model complexity—is known. However, determining the optimal rank is a nondeterministic polynomial-time hard (NP-hard) task and there is a limited understanding regarding the expressive power of functional low-rank tensor models for continuous signals.We propose a rank-revealing functional Bayesian tensor completion (RRFBTC) method. Modeling the latent functions through carefully designed multioutput Gaussian processes, RR-FBTC handles tensors with real-valued indices while enabling automatic tensor rank determination during the inference process. We establish the universal approximation property of the model for continuous multi-dimensional signals, demonstrating its expressive power in a concise format. To learn this model, we employ the variational inference framework and derive an efficient algorithm with closed-form updates. Experiments on both synthetic and realworld datasets demonstrate the effectiveness and superiority of the RR-FBTC over state-of-the-art approaches.
- 2 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Signal Processing
- Vol. 73  (1053587X) , 5319-5335
When Bayesian Tensor Completion Meets Multioutput Gaussian Processes: Functional Universality and Rank Learning
Functional tensor decomposition can analyze multidimensional data with real-valued indices, paving the path for applications in machine learning and signal processing. A limitation of existing approaches is the assumption that the tensor rank—a critical parameter governing model complexity—is known. However, determining the optimal rank is a nondeterministic polynomial-time hard (NP-hard) task and there is a limited understanding regarding the expressive power of functional low-rank tensor models for continuous signals.We propose a rank-revealing functional Bayesian tensor completion (RRFBTC) method. Modeling the latent functions through carefully designed multioutput Gaussian processes, RR-FBTC handles tensors with real-valued indices while enabling automatic tensor rank determination during the inference process. We establish the universal approximation property of the model for continuous multi-dimensional signals, demonstrating its expressive power in a concise format. To learn this model, we employ the variational inference framework and derive an efficient algorithm with closed-form updates. Experiments on both synthetic and realworld datasets demonstrate the effectiveness and superiority of the RR-FBTC over state-of-the-art approaches.
- 1 December 2025
- Published by Acoustical Society of America (ASA) in The Journal of the Acoustical Society of America
- Vol. 158  (6) , 5075-5089
Hankel-FNO: Fast underwater acoustic charting via physics-encoded Fourier neural operator
Fast and accurate underwater acoustic charting is crucial for downstream tasks such as environment-aware sensor placement optimization and autonomous vehicle path planning. Conventional methods rely on computationally expensive although accurate numerical solvers, which are not scalable for large-scale or real-time applications. Although deep learning–based surrogate models can accelerate these computations, they often suffer from limitations such as fixed-resolution constraints or dependence on explicit partial differential equation formulations. These issues hinder their applicability and generalization across diverse environments. We propose Hankel-FNO, a Fourier Neural Operator (FNO)-based model for efficient and accurate acoustic charting. By incorporating sound propagation knowledge and bathymetry, our method has high accuracy while maintaining high computational speed. Results demonstrate that the Hankel-FNO outperforms traditional solvers in speed and surpasses data-driven alternatives in accuracy, especially in long-range predictions. Experiments show the model's adaptability to diverse environments and sound source settings with minimal fine-tuning.
- 1 December 2025
- Published by Acoustical Society of America (ASA) in The Journal of the Acoustical Society of America
- Vol. 158  (6) , 5075-5089
Hankel-FNO: Fast underwater acoustic charting via physics-encoded Fourier neural operator
Fast and accurate underwater acoustic charting is crucial for downstream tasks such as environment-aware sensor placement optimization and autonomous vehicle path planning. Conventional methods rely on computationally expensive although accurate numerical solvers, which are not scalable for large-scale or real-time applications. Although deep learning–based surrogate models can accelerate these computations, they often suffer from limitations such as fixed-resolution constraints or dependence on explicit partial differential equation formulations. These issues hinder their applicability and generalization across diverse environments. We propose Hankel-FNO, a Fourier Neural Operator (FNO)-based model for efficient and accurate acoustic charting. By incorporating sound propagation knowledge and bathymetry, our method has high accuracy while maintaining high computational speed. Results demonstrate that the Hankel-FNO outperforms traditional solvers in speed and surpasses data-driven alternatives in accuracy, especially in long-range predictions. Experiments show the model's adaptability to diverse environments and sound source settings with minimal fine-tuning.
- 30 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 48  (2) , 1437-1455
To Fold or Not to Fold: Graph Regularized Tensor Train for Visual Data Completion
Tensor train (TT) representation has achieved tremendous success in visual data completion tasks, especially when it is combined with tensor folding. However, folding an image or video tensor breaks the original data structure, leading to local information loss as nearby pixels may be assigned into different dimensions and become far away from each other. In this paper, to fully preserve the local information of the original visual data, we explore not folding the data tensor, and at the same time adopt graph information to regularize local similarity between nearby entries. To overcome the high computational complexity introduced by the graph-based regularization in the TT completion problem, we propose to break the original problem into multiple sub-problems with respect to each TT core fiber, instead of each TT core as in traditional methods. Furthermore, to avoid heavy parameter tuning, a sparsity-promoting probabilistic model is built based on the generalized inverse Gaussian (GIG) prior, and an inference algorithm is derived under the mean-field approximation. Experiments on both synthetic data and real-world visual data show the superiority of the proposed methods.
- 30 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 48  (2) , 1437-1455
To Fold or Not to Fold: Graph Regularized Tensor Train for Visual Data Completion
Tensor train (TT) representation has achieved tremendous success in visual data completion tasks, especially when it is combined with tensor folding. However, folding an image or video tensor breaks the original data structure, leading to local information loss as nearby pixels may be assigned into different dimensions and become far away from each other. In this paper, to fully preserve the local information of the original visual data, we explore not folding the data tensor, and at the same time adopt graph information to regularize local similarity between nearby entries. To overcome the high computational complexity introduced by the graph-based regularization in the TT completion problem, we propose to break the original problem into multiple sub-problems with respect to each TT core fiber, instead of each TT core as in traditional methods. Furthermore, to avoid heavy parameter tuning, a sparsity-promoting probabilistic model is built based on the generalized inverse Gaussian (GIG) prior, and an inference algorithm is derived under the mean-field approximation. Experiments on both synthetic data and real-world visual data show the superiority of the proposed methods.