Author ORCID Identifier
0000-0002-0359-0689
Biosketch
Dr. B. Keerthana is a researcher and academic specializing in Digital Signal Processing, multirate systems, and biomedical signal analysis. She completed her Ph.D. from SASTRA Deemed University, Thanjavur (2025) with research focused on algorithmic optimization of filter banks and their applications in neurological disorder detection. She holds an M.E. in VLSI Design from Anna University (CGPA 8.43) and a B.E. in Electronics and Communication Engineering (CGPA 8.77).
She has over seven years of teaching and research experience in engineering institutions, serving as a Research Assistant at SASTRA University (2021–2025) and previously as an Assistant Professor at several government engineering colleges in Tamil Nadu. Her teaching portfolio includes courses such as Digital Signal Processing, Signals and Systems, Communication Theory, VLSI Design, Embedded Systems, and FPGA laboratories, reflecting strong academic and laboratory expertise.
Her research contributions include SCI/SCIE indexed journal publications in areas such as cosine modulated filter banks, non-uniform filter banks, and biomedical signal processing for sleep disorder diagnosis using EMG and EOG signals. She has also published conference papers and contributed to international book series in signal processing and communication technologies.
Dr. Keerthana’s technical expertise spans MATLAB, C, Verilog HDL, VHDL, FPGA systems, and embedded platforms such as Raspberry Pi. Her research interests focus on multirate DSP, optimized filter bank design, biomedical signal analysis, and intelligent healthcare applications, aiming to bridge signal processing theory with real-world biomedical and communication system implementations.
Date of Award
21-7-2025
Document Type
Thesis
School
School of Electrical & Electroncis Engineering
Programme
Ph.D.-Doctoral of Philosophy
First Advisor
Dr.N.Raju
Keywords
Filter Bank Design, Biomedical Signal Processing, Sleep Disorder Detection, Optimization Algorithm, Cosine Modulation
Abstract
A novel optimization framework is proposed for the design of both uniform and non-uniform filter banks, aimed at improving the accuracy and computational efficiency of biomedical signal classification tasks, with a particular emphasis on the detection of sleep disorders and Alzheimer’s disease. The proposed algorithm is grounded in multirate signal processing theory and aims to achieve Near Perfect Reconstruction (NPR) with minimal computational overhead. The process begins with the optimization of a uniform cosine-modulated filter bank (CMFB), achieved through iterative frequency-domain analysis and parameter tuning.
This is followed by the derivation of a non-uniform filter bank via selective merging of bandpass filters, based on predefined decimation factors relevant to biomedical signal characteristics. An analytic design for a prototype filter in M-channel maximally decimated CMFBs is proposed using a constrained least-squares (CLS) method with weighted passband and stopband constraints. The design reduces the optimization problem to a single parameter and ensures rapid convergence by analytically determining the optimal step size. This guarantees a 3 dB cutoff frequency at π/2M and results in low amplitude and aliasing distortion values—2.4489 × 10⁻⁴ and 3.4907 × 10⁻⁹, respectively.
Additionally, a rapidly converging optimization algorithm for non-uniform CMFBs is introduced, where the prototype filter’s cutoff frequency is adaptively varied using analytically computed step sizes to maintain 0.707 magnitude at the quadrature frequency while satisfying objective constraints. A constrained equiripple FIR filter is used, with design parameters selected based on stopband attenuation, passband ripple, and desired filter order. The optimized filter banks are applied to electromyogram (EMG) and electrooculogram (EOG) signals, decomposed into physiologically relevant sub-bands.
Extracted features such as Hjorth parameters and time–frequency representations are classified using machine learning models including SVM, Random Forest, and Bi- LSTM. Experimental results show sleep disorder classification accuracy of 99.97%, with corresponding precision, sensitivity, and specificity values exceeding 99.9%. For Alzheimer’s disease detection using EEG signals, a GRU-based model achieved 86.11% accuracy, with 86.11% precision and sensitivity, and 93.06% specificity.
To demonstrate real-time feasibility, the proposed models were functionally implemented on a Raspberry Pi platform, validating their execution under embedded resource constraints. While detailed hardware profiling was beyond the scope of this algorithm-focused work, the deployment confirms suitability for low-power biomedical applications. Finally, the thesis discusses the potential integration of quantum optimization techniques to accelerate convergence in high-dimensional filter bank design problems. This future direction positions the work toward scalable, high-performance solutions in real-time signal processing for healthcare applications.
Recommended Citation
B, Keerthana Ms, "Algorithmic Optimization of Filter Bank: Harnessing Rapid Convergence and its applications" (2025). Theses and Dissertations. 184.
https://knowledgeconnect.sastra.edu/theses/184