DSP Learning — Signal Processing Journey

ECE student at PSG College of Technology working through Digital Signal Processing from fundamentals to research-grade projects, with the long-term goal of MS/PhD admission in signal processing.

Status: Two complete IEEE-format paper drafts (Month 1 and Month 2-4), both awaiting professor review before conference submission. Upcoming semesters (Signals and Systems, Control Systems) will deepen the theoretical foundation before final submission push.

---

Project Overview

Core project — ECG Signal Denoising. Started as a filter comparison study (Butterworth IIR vs. FIR Kaiser), grew into a full clinical-validation pipeline, then extended into adaptive filtering (LMS/RLS), wavelet analysis, a hybrid filter architecture, and a real FPGA hardware implementation — now spanning two complete paper-length bodies of work.

---

Paper 1 (Month 1) — ECG Denoising: IIR vs FIR

Comparative study of Butterworth IIR and FIR Kaiser Window filters for ECG denoising on the MIT-BIH Arrhythmia Database, extended with clinical validation (Pan-Tompkins R-peak detection, QRS duration, paired t-test) beyond standard signal-domain metrics.

Key finding: Butterworth IIR outperforms FIR Kaiser on QRS morphology preservation and cross-record consistency despite FIR's linear-phase advantage. Both filters share a fundamental limitation: neither can remove noise falling inside the ECG's own diagnostic band (0.5–40Hz).

Status: Complete IEEE paper draft (Overleaf, IEEEtran, 5 pages). Professor-reviewed and approved; awaiting conference venue selection.

Notebook	Topic
ecg_denoising_project.ipynb	Full Month 1 pipeline

---

Paper 2 (Month 2–4) — Adaptive and Hybrid Filtering for In-Band Noise

Directly extends Paper 1's limitation. Systematically tests whether wavelet thresholding, LMS, RLS, and a hybrid IIR+LMS cascade can resolve the in-band noise problem that defeats both static filters from Paper 1.

Key findings:

• Wavelet DWT thresholding fails identically to static filters (F1=83.33%) — noise-character-dependent, not noise-frequency-dependent
• LMS adaptive filtering resolves it (F1=96.30%), generalizing across 4 diverse cardiac conditions (96.04% ± 0.40%) and the entire 6–14Hz vulnerable frequency band, with no per-condition parameter re-tuning required
• Hybrid Butterworth+LMS cascade improves SNR by a statistically validated 4.9× over LMS alone (paired t-test, n=30, t(29)=69.23, p<0.001) at only 14% additional computational cost

Status: Complete IEEE paper draft (Overleaf, IEEEtran). Literature review grounded against Thakor & Zhu (1991), Makdessy et al. (2020), and Khiter et al. (2020). Not yet reviewed by professor.

Notebook	Topic
month2_wavelet_analysis.ipynb	CWT scalograms, DWT denoising, LMS/RLS, in-band noise resolution
Month 4 extensions (multi-record, frequency sweep, hybrid, robustness)	Generalization and statistical validation

---

Month 3 — FPGA Hardware Implementation

Implements the Paper 1 FIR Kaiser filter (873 taps) in VHDL on a real Cyclone V FPGA, verified against the Python software implementation.

Key result: 1000/1000 samples verified against Python reference (max error 4 Q15 units). Time-multiplexed architecture uses 1 DSP block (of 156 available), 9% logic utilization. Meets 50MHz real-time timing target with +2.036ns positive slack under worst-case silicon conditions.

Full writeup: paper_notes/month3_fpga_report.md

File	Purpose
fir_filter_873tap.vhd	Filter entity + 3-process architecture (FSM + datapath)
fir_filter_tb.vhd	Questa testbench with file-based verification
fir_coeffs.vhd	Q1.15 coefficient ROM (873 entries)
fir_filter_873tap.sdc	Timing constraint (50MHz clock)

---

Foundational Notebooks (Pre-Paper-1)

Notebook	Topic
week1-signals-basics.ipynb	Sine waves, sampling, aliasing
week1-filtering-basics.ipynb	Low-pass filter, SNR measurement
signal-composition-demo.ipynb	Complete signal analysis pipeline
week2_day1_fft.ipynb	FFT fundamentals

---

Paper Notes / Literature Review

File	Paper
paper_notes/week1_paper1.md	ECG Baseline Denoising — Shi et al. 2021
paper_notes/month3_fpga_report.md	Month 3 FPGA implementation full report
Das & Chakraborty (2017), Gond et al. (2024)	Paper 1 literature grounding
Thakor & Zhu (1991), Makdessy et al. (2020), Khiter et al. (2020)	Paper 2 literature grounding

---

Tools

Python · NumPy · SciPy · Matplotlib · PyWavelets · MIT-BIH PhysioNet Dataset · VHDL · Questa · Quartus Prime (Cyclone V) · Overleaf/LaTeX (IEEEtran)