Research Direction

Research Areas: Core Methods & Representative Applications

The core methods explain how research problems are addressed. The application areas that follow describe the engineering contexts in which these methods are applied.

1. Robust and Computationally Efficient Adaptive Signal Processing

Real-world measurements can be affected by impulsive noise, outliers, and changing environmental conditions. This research uses robust error criteria, parameter adaptation, and data selection to reduce the influence of anomalous data on learning and avoid unnecessary parameter updates. Theoretical analyses and simulations examine convergence, steady-state error, and tracking behavior to assess the trade-offs between estimation accuracy and computational requirements.

Research questions: How can algorithms remain stable when sudden noise disturbances occur? Which data should be used to update the model?

Keywords: Adaptive filtering; Robust estimation; Correntropy; Data selection; Variable kernel width

Supporting research: [1] Correntropy-based data selection; [2] robust risk-sensitive criteria and kernel-width adaptation.

2. Kernel Adaptive Filtering and Nonlinear System Identification

Kernel methods enable the learning of nonlinear input–output relationships from streaming data. Recent work focuses on biases in nonlinear mappings and parameter estimates caused by measurement errors in the inputs. To address these challenges, the research develops bias-compensation techniques and update rules that are robust to impulsive output noise. Quantization and dictionary sparsification are also used to control model size, exploring the balance between nonlinear modeling capability and computational cost.

Research question: How can reliable nonlinear relationships be learned from imperfect measurements?

Keywords: Kernel adaptive filtering; KLMS; KMCC; Noisy inputs; Bias compensation; Dictionary sparsification

[1] Correntropy-Based Data Selective Adaptive Filtering

Y.-R. Chien, S.-T. Wu, H.-W. Tsao, and P. S. R. Diniz.

IEEE Transactions on Circuits and Systems I: Regular Papers, 71(2), 754–766, 2024.

https://doi.org/10.1109/TCSI.2023.3339632

[2] Optimizing kernel width for new risk-sensitive loss: A generalized algorithmic approach

Y. Tang, Y.-R. Chien, and G. Qian.

Digital Signal Processing, 154, 104655, 2024.

https://doi.org/10.1016/j.dsp.2024.104655

2. Kernel Adaptive Filtering and Nonlinear System Identification

Kernel methods enable the learning of nonlinear input–output relationships from streaming data. Recent work focuses on biases in nonlinear mappings and parameter estimates caused by measurement errors in the inputs. To address these challenges, the research develops bias-compensation techniques and update rules that are robust to impulsive output noise. Quantization and dictionary sparsification are also used to control model size, exploring the balance between nonlinear modeling capability and computational cost.

Research question: How can reliable nonlinear relationships be learned from imperfect measurements?

Keywords: Kernel adaptive filtering; KLMS; KMCC; Noisy inputs; Bias compensation; Dictionary sparsification

[3] Bias Compensation for Kernel Least-Mean-Square Algorithms

Y.-R. Chien, J.-L. Liu, E.-T. Lin, and G. Qian.

IEEE Sensors Letters, 9(4), 7001704, 2025.

https://doi.org/10.1109/LSENS.2025.3553594

[4] Robust Nonlinear Sensor Data Prediction in the Presence of Input Noise and Output Impulsive Outliers: A Bias-Compensated KMCC Approach

Y.-R. Chien, E.-T. Lin, and G. Qian.

IEEE Sensors Journal, 26(15), 23025–23037, 2026.

https://doi.org/10.1109/JSEN.2026.3704308

[5] On Quantized Kernel Affine Projection Algorithm: Sparsification via Delayed Vector Quantization

E.-T. Lin and Y.-R. Chien.

International Symposium on Intelligent Signal Processing and Communication Systems, ISPACS 2024.

https://doi.org/10.1109/ISPACS62486.2024.10868163

3. Distributed Adaptive Learning and Cooperative Estimation

This research investigates how multiple nodes use their own measurements and information from neighboring nodes to cooperatively identify unknown systems or estimate signals. Key topics include distributed nonlinear modeling, robustness to non-Gaussian noise, and the effects of network cooperation on convergence and estimation error. Recent work further explores cooperative weighting when input noise conditions differ across nodes, linking sensor networks with online learning.

Research question: How can sensors cooperate in estimation when they observe different data and experience different levels of noise?

Keywords: Distributed adaptive filtering; Diffusion learning; Cooperative estimation; Non-Gaussian noise

[6] Robust Diffusion Recursive Algorithm for Distributed Widely-Linear Exponential Functional Link Network

G. Qian, J. Wang, L. Shen, Y.-R. Chien, J. Qian, and S. Wang.

IEEE Transactions on Signal and Information Processing over Networks, 11, 845–858, 2025.

https://doi.org/10.1109/TSIPN.2025.3589685

4. Audio Signal Processing and Hearing Assistance

Adaptive and robust learning methods are applied to active noise control, acoustic echo cancellation, acoustic feedback cancellation, and spatial noise suppression in digital hearing aids. The research addresses changing acoustic paths, impulsive noise, and nonlinearities, while investigating adaptive directivity and speech enhancement methods. The aim is to reduce unwanted sounds while preserving useful speech information.

Research questions: How can headphones adapt to changing noise conditions? How can hearing aids reduce acoustic feedback and suppress sounds arriving from other directions?

Keywords: Active noise control; Acoustic echo cancellation; Acoustic feedback cancellation; Hearing aids

[7] Affine-Projection-Like Maximum Correntropy Criteria Algorithm for Robust Active Noise Control

Y.-R. Chien, C.-H. Yu, and H.-W. Tsao.

IEEE/ACM Transactions on Audio, Speech, and Language Processing, 30, 2255–2266, 2022.

https://doi.org/10.1109/TASLP.2022.3190720

[8] Robust Volterra Filter for Nonlinear Censored Regression and Its Applications to Nonlinear Acoustic Echo Cancellation

G. Qian, L. Shen, Y. Wang, Y.-R. Chien, J. Qian, and S. Wang.

IEEE Transactions on Consumer Electronics, 71(2), 2567–2578, 2025.

https://doi.org/10.1109/TCE.2025.3570854

[9] Variable Step-Size Modified One-Tap Affine-Projection Algorithm for Adaptive Null-Forming in Digital Hearing Aids

Y.-C. Huang, Y.-H. Lin, F.-J. Tey, and Y.-R. Chien.

International Symposium on Intelligent Signal Processing and Communication Systems, ISPACS 2025.

https://doi.org/10.1109/ISPACS68724.2025.11382579

5. Radar Sensing and Array Signal Processing

This research combines millimeter-wave radar, signal processing, and machine learning to study non-contact human activity recognition, range estimation, and direction-of-arrival (DOA) estimation. Representative methods include the fractional Fourier transform, clutter suppression, radar data representations, and robust adaptive array processing. The focus is on how noise, interference, and limited measurement data affect sensing performance.

Research question: How can radar signals be used to recognize human activities, measure range, or determine the direction from which a signal arrives?

Keywords: Millimeter-wave radar; Range estimation; Human activity recognition; Fractional Fourier transform; DOA estimation

[10] Robust Fractional Fourier Transform-Based Range Estimation Using Millimeter-Wave Radar for IoT Applications

S.-H. Fang, Y.-C. Jie, Y.-R. Chien, and H.-C. Wu.

IEEE Internet of Things Journal, 12(22), 47424–47436, 2025.

https://doi.org/10.1109/JIOT.2025.3602469

[11] Noninvasive Human Activity Recognition Using Millimeter-Wave Radar

C. Yu, Z. Xu, K. Yan, Y.-R. Chien, S.-H. Fang, and H.-C. Wu.

IEEE Systems Journal, 16(2), 3036–3047, 2022.

https://doi.org/10.1109/JSYST.2022.3140546

[12] Robust DOA estimation with combined adaptive filtering and alternating maximum versoria based denoising

Y. Tang, J. Peng, Y.-R. Chien, and H. Zeng.

Signal Processing, 242, 110439, 2026.

https://doi.org/10.1016/j.sigpro.2025.110439

6. Interference Mitigation and Noise Modeling for Communications

This research investigates how communication receivers can handle impulsive noise and external interference, and develops noise models that reflect real-world conditions. Recent work on powerline communications uses real-world measurements and generative adversarial networks to learn the statistical characteristics of complex noise, supporting algorithm testing and synthetic data generation. Earlier representative work includes adaptive notch filtering and anti-jamming designs for GPS receivers.

Research question: How can realistic communication noise be simulated, and how can the effects of strong interference on receivers be reduced?

Keywords: Powerline communications; Impulsive noise; Measurement-based modeling; Generative adversarial networks; Anti-jamming

[13] NGGAN: Noise Generation GAN Based on the Practical Measurement Dataset for Narrowband Powerline Communications

Y.-R. Chien, P.-H. Chou, Y.-J. Peng, C.-Y. Huang, H.-W. Tsao, and Y. Tsao.

IEEE Transactions on Instrumentation and Measurement, 74, 2505415, 2025.

https://doi.org/10.1109/TIM.2024.3523361

[14] Design of GPS Anti-Jamming Systems Using Adaptive Notch Filters

Y.-R. Chien.

IEEE Systems Journal, 9(2), 451–460, 2015.

https://doi.org/10.1109/JSYST.2013.2283753