Biosignal Processing (BM4152) Repository

This repository contains a collection of assignments completed for the BM4152 Biosignal Processing module, along with a research paper implementation focused on heart sound classification using deep learning and texture features.

Repository Structure

The repository is organized into four main sections: three assignments covering various signal processing techniques and one comprehensive research paper implementation.

---

📂 Assignment 01: Digital Filter Design Basics

Focus: Fundamentals of Digital Signal Processing and Filter Design.

This assignment explores the design and application of digital filters to process biosignals such as ECG (Electrocardiogram) and ABR (Auditory Brainstem Response).

• Key Topics:
  • FIR & IIR Filter Design: Designing lowpass, highpass, and bandpass filters.
  • Windowing Techniques: Analysis of different window functions (Hann, Hamming, Rectangular, Blackman).
  • Filter Types: Implementation of Butterworth and Kaiser filters.
  • Noise Removal: Applying designed filters (e.g., Comb filters) to remove noise and artifacts from raw signal recordings (ABR_rec.mat, ECG_with_noise.mat).
• Key Files: Assignment_01_script.mlx, Lowpass & highpass kaiser filter design.fda, comb_filter.fig.

---

📂 Assignment 02: Optimal & Adaptive Filtering

Focus: Advanced Statistical Signal Processing.

This assignment delves into optimal estimation and adaptive filtering algorithms, specifically targeting noise reduction and signal enhancement in non-stationary biosignals.

• Key Topics:
  • Adaptive Filtering: Implementation of LMS (Least Mean Squares) and RLS (Recursive Least Squares) algorithms for noise cancellation.
  • Wiener Filtering: Design of optimal filters for enhancing noisy ECG signals.
  • Linear Prediction: Modeling biosignals using linear prediction methods to estimate Signal-to-Noise Ratio (SNR) and underlying signal structure.
  • Spectral Analysis: Using Power Spectral Density (PSD) to analyze filter performance.
• Key Files: Q1.mlx, Q2.mlx, Bio_Signal_Processing_Assignment_02.pdf, and performance plots in Figures/ (e.g., LMS_algorithm.png, order15_weiner_filter_result.png).

---

📂 Assignment 03: Wavelet Analysis

Focus: Time-Frequency Analysis using Wavelet Transforms.

This assignment investigates the use of Wavelet Transforms for analyzing non-stationary signals, offering advantages over traditional Fourier methods for detecting transient features in ECG signals.

• Key Topics:
  • Discrete Wavelet Transform (DWT): Decomposition of signals using Haar and Daubechies (DB9) wavelets.
  • Signal Denoising: Removing high-frequency noise while preserving signal peaks.
  • Signal Compression: Reducing signal size by thresholding wavelet coefficients.
  • Time-Frequency Representation: Analyzing signals using spectrograms and scalograms to visualize frequency changes over time.
• Key Files: Assignment_3_Q2.mlx, Assignment_03_Q3.mlx, and resulting visualizations like spectrogram.png, Reconstructed Components...png.

---

🧬 Research Paper Implementation

Title: Heart sounds classification using convolutional neural network with 1D-local binary pattern and 1D-local ternary pattern features DOI: 10.1016/j.apacoust.2021.108152

This project implements a hybrid approach for classifying heart sounds (e.g., Normal vs. Abnormal/Murmurs) using the PhysioNet dataset.

• Methodology:
  • Feature Extraction: Utilizes 1D-Local Binary Patterns (LBP) and 1D-Local Ternary Patterns (LTP) to capture local texture and structural information from the 1D audio signals.
  • Feature Selection: Implements ReliefF algorithm to select the most discriminative features.
  • Deep Learning Model: A 1D Convolutional Neural Network (CNN) is trained on the processed features/signals for robust classification.
• Data: Uses heart sound recordings from the PhysioNet database.
• File: BSP_paper_Implementation_physionet_data.ipynb

---