Muhammad Tahir Akhtar
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
Muhammad Tahir Akhtar has a total of 47 co-authors. Here are the top co-authors by number of shared publications:
Wataru Mitsuhashi
The University of Electro-Communications, 1-5-1 Chofugaoka, Chofu, Tokyo 182-8585, Japan
17 shared publicationsAkinori Nishihara
9 shared publicationsM. Abe
Graduate School of Engineering, Tohoku University 6-6-05 Aza-Aoba, Aramaki, Aoba-ku, Sendai, 980-8579, JAPAN
8 shared publicationsM. Kawamata
Graduate School of Engineering, Tohoku University 6-6-05 Aza-Aoba, Aramaki, Aoba-ku, Sendai, 980-8579, JAPAN
8 shared publicationsShakeel Ahmed
Department of Electrical Engineering, Pakistan Institute of Engineering and Applied Sciences, Islamabad, Pakistan
6 shared publicationsMasahide Abe
5 shared publications
Latest Publications
- Open Access
- 13 August 2026
- Published by Springer Nature in Discover Internet of Things
- Vol. 6 (1)
Resource-aware edge deployment and trade-off analysis of TinyML classifiers for IoT health monitoring on an ESP32 microcontroller
- 13 August 2026
- Published by Springer Nature in Discover Internet of Things
- Vol. 6 (1)
Open AccessResource-aware edge deployment and trade-off analysis of TinyML classifiers for IoT health monitoring on an ESP32 microcontroller
- Open Access
- 2 May 2026
- Published by Springer Nature in Discover Artificial Intelligence
Hybrid CNN–LSTM network for cross-subject EEG emotion recognition
- 2 May 2026
- Published by Springer Nature in Discover Artificial Intelligence
Open AccessHybrid CNN–LSTM network for cross-subject EEG emotion recognition
- Open Access
Modality-Specific Sparse Autoencoders for Efficient Multimodal ICU Alignment: A Symmetry–Asymmetry Learning Framework
Intensive care units (ICUs) generate heterogeneous data streams, including structured electronic health records, physiological time series, and medical imaging, that describe the same patient state through different observational forms. Effective multimodal learning in this setting requires a principled balance between representation-level symmetry and architectural asymmetry. Clinically corresponding patient states should exhibit cross-modal representational symmetry, whereas each modality retains intrinsic asymmetry in dimensionality, temporal resolution, noise characteristics, and missingness. This study proposes a modality-specific sparse autoencoder framework for efficient multimodal ICU representation learning under this symmetry–asymmetry principle. Separate sparse encoders are assigned to each modality to preserve the modality-dependent structure while suppressing redundant latent activity through adaptive gating. Representation-level symmetry is encouraged through a sparsity-aware contrastive objective that aligns paired latent embeddings across modalities only on active informative dimensions. To further model inter-patient dependencies, the framework incorporates a graph neural network (GNN) whose message-passing operations respect modality-specific sparsity patterns. Experimental results indicate that the proposed framework improves predictive performance and computational efficiency relative to conventional multimodal baselines, while also exhibiting stronger robustness under missing-modality conditions and more selective latent representations. Overall, the method provides an effective and clinically relevant multimodal learning strategy for ICU decision support while offering a measurable symmetry-aware and asymmetry-preserving formulation for heterogeneous medical data.
Open AccessModality-Specific Sparse Autoencoders for Efficient Multimodal ICU Alignment: A Symmetry–Asymmetry Learning Framework
Intensive care units (ICUs) generate heterogeneous data streams, including structured electronic health records, physiological time series, and medical imaging, that describe the same patient state through different observational forms. Effective multimodal learning in this setting requires a principled balance between representation-level symmetry and architectural asymmetry. Clinically corresponding patient states should exhibit cross-modal representational symmetry, whereas each modality retains intrinsic asymmetry in dimensionality, temporal resolution, noise characteristics, and missingness. This study proposes a modality-specific sparse autoencoder framework for efficient multimodal ICU representation learning under this symmetry–asymmetry principle. Separate sparse encoders are assigned to each modality to preserve the modality-dependent structure while suppressing redundant latent activity through adaptive gating. Representation-level symmetry is encouraged through a sparsity-aware contrastive objective that aligns paired latent embeddings across modalities only on active informative dimensions. To further model inter-patient dependencies, the framework incorporates a graph neural network (GNN) whose message-passing operations respect modality-specific sparsity patterns. Experimental results indicate that the proposed framework improves predictive performance and computational efficiency relative to conventional multimodal baselines, while also exhibiting stronger robustness under missing-modality conditions and more selective latent representations. Overall, the method provides an effective and clinically relevant multimodal learning strategy for ICU decision support while offering a measurable symmetry-aware and asymmetry-preserving formulation for heterogeneous medical data.
- Open Access
- 19 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 9696-9716
Spectral Filtering Using Periodic Autoregressive Moving Average Graph Neural Networks for Heterophilic Graphs
The importance of Graph Neural Networks (GNNs) has increased over the years due to their ability to handle non-Euclidean data. Most of the existing research mainly focuses on spatial relationships using neighboring nodes to aggregate information, which usually corresponds to lowpass filtering from the perspective of graph signal processing theory. However, many real-world applications involve heterophilic graphs, where nodes with dissimilar labels or features frequently connect, necessitating highpass filtering capabilities. Existing highpass GNNs are typically parameter-intensive, limiting their efficiency and practical development. This problem can be addressed by utilizing Autoregressive Moving Average graph (ARMA) filters. Thus, a novel periodic ARMA GNN (pARMA-GNN) model that efficiently approximates rational spectral filters using periodic recursive filtering is proposed. The proposed pARMA-GNN model consists of periodic ARMA convolutional (pARMAConv) layers that significantly reduce computational complexity and parameter count compared to the current state-of-the-art models specializing in capturing both high-frequency and low-frequency information. The theoretical analysis of the frequency response of the proposed model has shown its ability to capture high-frequency and low-frequency components. Additionally, several variants of the proposed model have been developed to solve specific issues related to homophilic and heterophilic datasets. The results of experiments conducted on synthetic data have shown the practical capabilities of the model to filter different frequencies compared to other baseline models. Moreover, results on real-world datasets indicate that pARMA-GNN provides competitive performance on homophilic graphs, while outperforming prior models by about 1-10% on heterophilic graph datasets. The analysis of computational complexity of ARMA-based GNNs indicates that the proposed model has a great advantage in certain scenarios. Finally, ablation study on the effect of number of period has been conducted to show the impact of periodic weights on both homophilic and heterophilic datasets.
- 19 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 9696-9716
Open AccessSpectral Filtering Using Periodic Autoregressive Moving Average Graph Neural Networks for Heterophilic Graphs
The importance of Graph Neural Networks (GNNs) has increased over the years due to their ability to handle non-Euclidean data. Most of the existing research mainly focuses on spatial relationships using neighboring nodes to aggregate information, which usually corresponds to lowpass filtering from the perspective of graph signal processing theory. However, many real-world applications involve heterophilic graphs, where nodes with dissimilar labels or features frequently connect, necessitating highpass filtering capabilities. Existing highpass GNNs are typically parameter-intensive, limiting their efficiency and practical development. This problem can be addressed by utilizing Autoregressive Moving Average graph (ARMA) filters. Thus, a novel periodic ARMA GNN (pARMA-GNN) model that efficiently approximates rational spectral filters using periodic recursive filtering is proposed. The proposed pARMA-GNN model consists of periodic ARMA convolutional (pARMAConv) layers that significantly reduce computational complexity and parameter count compared to the current state-of-the-art models specializing in capturing both high-frequency and low-frequency information. The theoretical analysis of the frequency response of the proposed model has shown its ability to capture high-frequency and low-frequency components. Additionally, several variants of the proposed model have been developed to solve specific issues related to homophilic and heterophilic datasets. The results of experiments conducted on synthetic data have shown the practical capabilities of the model to filter different frequencies compared to other baseline models. Moreover, results on real-world datasets indicate that pARMA-GNN provides competitive performance on homophilic graphs, while outperforming prior models by about 1-10% on heterophilic graph datasets. The analysis of computational complexity of ARMA-based GNNs indicates that the proposed model has a great advantage in certain scenarios. Finally, ablation study on the effect of number of period has been conducted to show the impact of periodic weights on both homophilic and heterophilic datasets.
- Open Access
- 23 November 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14 (23) , 10866
A Hybrid Approach to Enhanced Signal Denoising Using Data-Driven Multiresolution Analysis with Detrended-Fluctuation-Analysis-Based Thresholding and Stationary Wavelet Transform
In this work, a new method for denoising signals is developed that is based on variational mode decomposition (VMD) and a novel metric using detrended fluctuation analysis (DFA). The proposed method first decomposes the signal into band-limited intrinsic mode functions (BLIMFs) using VMD. Then, a DFA-based developed metric is employed to identify the ‘noisy’ BLIMFs (based on their DFA-based scaling exponent and frequency content). The existing DFA-based methods use a single-slope threshold to detect noise, assuming all signals have the same noise pattern and ignoring their unique characteristics. In contrast, the proposed DFA-based metric sets adaptive thresholds for each mode based on their specific frequency and correlation properties, making it more effective for diverse signals and noise types. These predominantly noisy BLIMFs are then denoised using shrinkage techniques in the framework of stationary wavelet transform (SWT). This step allows efficient denoising of components, mainly the noisy BLIMFs identified by the adaptive threshold, without losing important signal details. Extensive computer simulations have been carried out for both synthetic and real electrocardiogram (ECG) signals. It is demonstrated that the proposed method outperforms the state-of-the-art denoising methods and with a comparable computational complexity.
- 23 November 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14 (23) , 10866
Open AccessA Hybrid Approach to Enhanced Signal Denoising Using Data-Driven Multiresolution Analysis with Detrended-Fluctuation-Analysis-Based Thresholding and Stationary Wavelet Transform
In this work, a new method for denoising signals is developed that is based on variational mode decomposition (VMD) and a novel metric using detrended fluctuation analysis (DFA). The proposed method first decomposes the signal into band-limited intrinsic mode functions (BLIMFs) using VMD. Then, a DFA-based developed metric is employed to identify the ‘noisy’ BLIMFs (based on their DFA-based scaling exponent and frequency content). The existing DFA-based methods use a single-slope threshold to detect noise, assuming all signals have the same noise pattern and ignoring their unique characteristics. In contrast, the proposed DFA-based metric sets adaptive thresholds for each mode based on their specific frequency and correlation properties, making it more effective for diverse signals and noise types. These predominantly noisy BLIMFs are then denoised using shrinkage techniques in the framework of stationary wavelet transform (SWT). This step allows efficient denoising of components, mainly the noisy BLIMFs identified by the adaptive threshold, without losing important signal details. Extensive computer simulations have been carried out for both synthetic and real electrocardiogram (ECG) signals. It is demonstrated that the proposed method outperforms the state-of-the-art denoising methods and with a comparable computational complexity.
- 21 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
On Developing a Robust Filtered-Reference RLS Algorithm and its Application for ANC Systems Targeting Impulsive Noise Sources
This paper develops a robust filtered-reference recursive least squares (RLS) algorithm for active noise control (ANC) systems targeting impulsive sources. The developed algorithm is based on a previous work exploiting an objective function showing robustness to tackle impulsive sources. The update equation in the previous algorithm incorporates a fixed step-size as a learning parameter. Furthermore, the forgetting factor (signifying the memory of least squares adaptation) is kept constant as in the traditional RLS algorithm. This paper first develops a generalized sigmoid activation function computed on the basis of internally generated error signal in the ANC system. The activation function is tuned such that it stays close to unity during the transient state and decays towards zero at the steady-state. This allows to tune the step-size as well as the forgetting factor such that the step-size parameter (the forgetting factor) is adjusted to a large value (small value) during the transient state to improve upon the convergence speed. On the other hand, the step-size parameter (the forgetting factor) is automatically tuned to a small value (large value) as the ANC system approaches the steady-state. Therefore, the proposed algorithm completely solves the trade-off situation. Extensive numerical simulations have been carried out which show the effective performance of the proposed algorithm.
- 21 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
On Developing a Robust Filtered-Reference RLS Algorithm and its Application for ANC Systems Targeting Impulsive Noise Sources
This paper develops a robust filtered-reference recursive least squares (RLS) algorithm for active noise control (ANC) systems targeting impulsive sources. The developed algorithm is based on a previous work exploiting an objective function showing robustness to tackle impulsive sources. The update equation in the previous algorithm incorporates a fixed step-size as a learning parameter. Furthermore, the forgetting factor (signifying the memory of least squares adaptation) is kept constant as in the traditional RLS algorithm. This paper first develops a generalized sigmoid activation function computed on the basis of internally generated error signal in the ANC system. The activation function is tuned such that it stays close to unity during the transient state and decays towards zero at the steady-state. This allows to tune the step-size as well as the forgetting factor such that the step-size parameter (the forgetting factor) is adjusted to a large value (small value) during the transient state to improve upon the convergence speed. On the other hand, the step-size parameter (the forgetting factor) is automatically tuned to a small value (large value) as the ANC system approaches the steady-state. Therefore, the proposed algorithm completely solves the trade-off situation. Extensive numerical simulations have been carried out which show the effective performance of the proposed algorithm.
- 11 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 698-702
On Adaptive Feedback Active Noise Control Systems With Online Adaptation of Secondary Path Estimation Filter
This paper focuses on identifying the secondary path (SP) in feedback active noise control (FBANC) systems during online operation. In FBANC, it's not feasible to install a reference microphone for capturing the reference signal needed for the (adaptive) active noise control (ANC) filter. Consequently, an appropriate reference signal is required to be electronically generated, necessitating an accurate estimation of the SP. The accuracy of SP estimation (SPE) directly impacts the stability and noise reduction capabilities of ANC systems. Building upon prior work that utilized two adaptive filters; one for ANC and another for SPE; this paper presents a method with a similar structure. The key idea in this paper is to tailor a generalized sigmoid function for developing a variable step-size (VSS)-based adaptive algorithm for (adaptive) ANC filter. The proposed method exhibits improved noise reduction performance as well as good SPE performance, as evidenced by the detailed numerical results provided in the paper.
- 11 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 698-702
On Adaptive Feedback Active Noise Control Systems With Online Adaptation of Secondary Path Estimation Filter
This paper focuses on identifying the secondary path (SP) in feedback active noise control (FBANC) systems during online operation. In FBANC, it's not feasible to install a reference microphone for capturing the reference signal needed for the (adaptive) active noise control (ANC) filter. Consequently, an appropriate reference signal is required to be electronically generated, necessitating an accurate estimation of the SP. The accuracy of SP estimation (SPE) directly impacts the stability and noise reduction capabilities of ANC systems. Building upon prior work that utilized two adaptive filters; one for ANC and another for SPE; this paper presents a method with a similar structure. The key idea in this paper is to tailor a generalized sigmoid function for developing a variable step-size (VSS)-based adaptive algorithm for (adaptive) ANC filter. The proposed method exhibits improved noise reduction performance as well as good SPE performance, as evidenced by the detailed numerical results provided in the paper.
- 8 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-5
Engaging Undergraduate Students in Signal Processing Research: Vision, Strategy, and A Success Story
There is a great deal of effort (and literature) on preparing undergraduate students to work on research-based problems in the area of signal processing during their undergraduate studies. However, most of the undergraduate students are in fact not ready to carry out some meaningful research until their graduation. One possible reason could be that most of the research problems in the area of signal processing require significant background knowledge in advanced linear algebra and stochastic processes. As a result, students are typically assigned mini-projects (derived from a larger problem being solved), or student-driven design projects in some independent study course. It is expected that after graduating, they will be ready and motivated to do research during their graduate-level higher studies. The focus of this paper, on the other hand, is to present the efforts and results on engaging students in doing research during their undergraduate studies. The Electrical and Computer Engineering (ECE) Department at Nazarbayev University (NU) attracts some of the best students in the country. Considering their potential and motivation level, a strategy has envisaged to engage them for doing research during the early stage of their undergraduate studies. The results of this strategy have been very promising. The selected undergraduate students learned to do independent research in the area of signal processing (and machine learning). Furthermore, their research work has resulted in numerous publications in top-ranked journals and conferences.
- 8 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-5
Engaging Undergraduate Students in Signal Processing Research: Vision, Strategy, and A Success Story
There is a great deal of effort (and literature) on preparing undergraduate students to work on research-based problems in the area of signal processing during their undergraduate studies. However, most of the undergraduate students are in fact not ready to carry out some meaningful research until their graduation. One possible reason could be that most of the research problems in the area of signal processing require significant background knowledge in advanced linear algebra and stochastic processes. As a result, students are typically assigned mini-projects (derived from a larger problem being solved), or student-driven design projects in some independent study course. It is expected that after graduating, they will be ready and motivated to do research during their graduate-level higher studies. The focus of this paper, on the other hand, is to present the efforts and results on engaging students in doing research during their undergraduate studies. The Electrical and Computer Engineering (ECE) Department at Nazarbayev University (NU) attracts some of the best students in the country. Considering their potential and motivation level, a strategy has envisaged to engage them for doing research during the early stage of their undergraduate studies. The results of this strategy have been very promising. The selected undergraduate students learned to do independent research in the area of signal processing (and machine learning). Furthermore, their research work has resulted in numerous publications in top-ranked journals and conferences.
- Open Access
- 10 August 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 85165-85183
A Systematic Review on Distributed Arithmetic-Based Hardware Implementation of Adaptive Digital Filters
Adaptive Digital Filters (ADFs) are computationally demanding Digital Signal Processing (DSP) systems with applications in diverse areas signal processing, such as active noise control, channel equalization, system identification. The most popular implementation structure for ADFs is FIR transversal filter which can be efficiently implemented without multipliers by Distributed Arithmetic (DA) method. DA is used to efficiently implement digital FIR filter with a precomputed weight DA combinations in a memory element. However, due to the operation of ADFs the memory content has to be recomputed at each iteration; the advantage of DA method disappears. The objective of this review is to cover the research gap on DA-ADF implementations. This article reviews the main features and contributions of DA-ADF designs present in literature. The DA-ADF designs have been grouped into three categories: non-pipelined, pipelined and block processing. Pipelining and block processing increase the throughput of the ADF. The reviewed designs implement only Least Mean Square (LMS) type adaptive algorithms due to the simplicity of implementation. Latest and most efficient designs tend to employ conjugate Offset Binary Coding (OBC) DA structure.
- 10 August 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 85165-85183
Open AccessA Systematic Review on Distributed Arithmetic-Based Hardware Implementation of Adaptive Digital Filters
Adaptive Digital Filters (ADFs) are computationally demanding Digital Signal Processing (DSP) systems with applications in diverse areas signal processing, such as active noise control, channel equalization, system identification. The most popular implementation structure for ADFs is FIR transversal filter which can be efficiently implemented without multipliers by Distributed Arithmetic (DA) method. DA is used to efficiently implement digital FIR filter with a precomputed weight DA combinations in a memory element. However, due to the operation of ADFs the memory content has to be recomputed at each iteration; the advantage of DA method disappears. The objective of this review is to cover the research gap on DA-ADF implementations. This article reviews the main features and contributions of DA-ADF designs present in literature. The DA-ADF designs have been grouped into three categories: non-pipelined, pipelined and block processing. Pipelining and block processing increase the throughput of the ADF. The reviewed designs implement only Least Mean Square (LMS) type adaptive algorithms due to the simplicity of implementation. Latest and most efficient designs tend to employ conjugate Offset Binary Coding (OBC) DA structure.