📊 BioSignal AI - Multi-Modal Signal Viewer

🔬 Signal Viewer AI Platform

Multi-Modal Signal Analysis with Deep Learning

DSP Course - Task 01 | Team 08 | Spring 2026

---

📋 Table of Contents

1. Project Overview
2. System Architecture
3. Medical Signals (ECG/EEG)
4. Acoustic Signals
5. Trading Signals
6. Microbiome Signals
7. Installation & Setup
8. Usage Guide
9. Technical Implementation
10. Demo Video

---

🎯 Project Overview

BioSignal AI is a comprehensive multi-modal signal analysis platform that implements all requirements of Task 1: Signal Viewer with basic processing. The system handles four distinct signal types with specialized visualization and AI-powered analysis.

Key Features at a Glance

Signal Type	Key Capabilities	AI Models Used
Medical	Multi-channel ECG/EEG, 4+ viewer types, abnormality detection	HuBERT + MLP Classifier
Acoustic	Doppler simulation, vehicle velocity estimation, drone detection	Classic algorithms (non-AI)
Trading	Stock/currency/mineral analysis, LSTM prediction	Global LSTM (3-layer)
Microbiome	Disease profiling, diversity metrics, clinical reports	Random Forest

---

🏗️ System Architecture

┌─────────────────────────────────────────────────────────────┐
│                     Flask Backend (app.py)                   │
│         Routes, File Handling, Model Loading, API            │
└─────────────────────────────────────────────────────────────┘
                              │
        ┌─────────────────────┼─────────────────────┐
        ▼                     ▼                     ▼
┌───────────────┐    ┌───────────────┐    ┌───────────────┐
│  Medical API  │    │  Acoustic API │    │  Trading API  │
│  /upload      │    │/upload_sound  │    │ /api/upload   │
│/analyze_signal│    │ /detect_drone │    │ /api/predict  │
└───────────────┘    └───────────────┘    └───────────────┘
        │                     │                     │
        ▼                     ▼                     ▼
┌───────────────┐    ┌───────────────┐    ┌───────────────┐
│   Models/     │    │   Audio       │    │   Models/     │
│   Medical/    │    │ Processing    │    │   Trading/    │
└───────────────┘    └───────────────┘    └───────────────┘

Project Structure

SIGNAL_VIEWER/
│
├── Data/                          # Raw signal data
│   ├── Acoustic Signals/          # Vehicle and drone audio
│   ├── Medical Signals/           # ECG/EEG recordings
│   ├── Microbiome Signals/        # Microbial abundance data
│   └── Trading Signals/           # Stock/currency/mineral CSVs
│
├── Models/                         # Pre-trained AI models
│   ├── Medical/                    # ECG classification models
│   ├── Microbiome/                  # Random Forest models
│   └── Trading/                     # LSTM models
│
├── static/                          # Frontend assets
│   ├── CSS/                         # Styling files
│   └── JS/                          # JavaScript logic
│
├── Templates/                        # HTML pages
│   ├── index.html                    # Main entry
│   ├── viewer.html                    # Medical viewer
│   ├── sound.html                      # Acoustic analyzer
│   ├── stock.html                      # Trading dashboard
│   └── micro.html                      # Microbiome analysis
│
├── docs/                              # Documentation
│   ├── images/                        # Screenshots
│   └── Video/                         # Demo video
│
└── app.py                             # Main Flask application

---

❤️ Medical Signals (ECG/EEG)

Overview

The medical signal viewer supports multi-channel ECG (heart) and EEG (brain) signals with multiple visualization modes and AI-based abnormality detection.

📸 Medical Page

Main medical dashboard with sidebar controls and chart grid

Supported File Formats

• .hea + .dat (WFDB format)
• .edf (European Data Format)
• .csv (custom format)

🎛️ Control Panel Features

Signal Type Selection

Choose between ECG and EEG signals

Signal Type Selection 2

Alternative signal type selector

📊 Viewer Types Implemented

1. Continuous-Time Signal Viewer

Standard time-domain visualization with amplitude vs time

Features:

• Viewport of fixed time-length
• Speed control (1x to 20x)
• Window size adjustment (100-5000 samples)
• Play/Stop controls
• Zoom in/out capability

2. Polar Graph Viewer

Polar representation where r = magnitude, θ = time

Advanced Polar Modes:

• Standard Overlay - Multiple channels overlaid
• Ratio Mode (V1/V2) - Channel ratio visualization
• Ratio Mode (V2/V1) - Inverse ratio visualization

Two-channel polar overlay with ratio analysis

Multiple channels overlaid in polar coordinates

Enhanced polar overlay with multiple channels

3. XOR Graph Viewer

Signal divided into time chunks and XOR-ed

How it works:

• Signal divided into chunks (window size)
• Each chunk plotted on top of previous
• Identical chunks cancel out (XOR effect)
• Highlights differences between periods

XOR visualization with adjustable chunk size

XOR view with multiple channels overlaid

4. Recurrence/Poincaré Map

2D Scatter plot betwee X[n] and X[n+1]

Customization:

• X-axis channel selection
• Y-axis channel selection
• Color map selection (Hot, Viridis, Plasma, Jet)
• Shows signal dynamics and patterns

Enhanced recurrence plot with density contours

Standard overlay view with multiple channels

🎨 Channel Customization

Channel Selection

Multi-select channels with Ctrl+Click

Per-Channel Properties

• Color picker - Custom color per channel
• Line thickness - Adjustable width (0.5-5.0)
• Animation toggle - Enable/disable per channel
• Show/Hide - Toggle visibility in overlay mode

🤖 AI Abnormality Detection

AI Comparison Panel

Side-by-side comparison of Deep Learning and Classic ML

Deep Learning Model (HuBERT + MLP)

• Feature Extractor: HuBERT Transformer model
• Classifier: 3-layer MLP (256→128→64 neurons)
• Classes: 5 ECG abnormality types
  • Normal (N)
  • Supraventricular ectopic beat (S)
  • Ventricular ectopic beat (V)
  • Fusion beat (F)
  • Unknown beat (Q)
• Multi-channel analysis: Analyzes up to 3 channels

Classic ML Algorithm

• R-peak detection using scipy.signal.find_peaks
• Heart rate calculation (BPM)
• Statistical features: SDNN, RMSSD
• Rule-based classification:
  • SDNN > 0.12 → Arrhythmia suspected
  • BPM > 100 → Tachycardia
  • BPM < 50 → Bradycardia
  • Otherwise → Normal rhythm

⚙️ Playback Controls

Speed and Window Controls

Adjust playback speed and viewing window

• Speed Slider: 1x to 20x
• Window Size: 100 to 5000 samples
• Real-time animation across all viewers
• Synchronized playback in multi-view mode

📈 All Views Showcase

Complete set of viewer types in single mode

🧠 EEG-Specific Features

EEG-specific analysis options

Choose from multiple analysis modes

---

🔊 Acoustic Signals

Overview

The acoustic module provides Doppler effect simulation and real-world vehicle/drone analysis using classic signal processing algorithms.

📸 Acoustic Page

Main acoustic analysis dashboard

🚗 Doppler Effect Simulation

Sound Generation Controls

Velocity and frequency sliders with play/pause controls

Physics Model:

f_approach = f * (v_sound + v) / (v_sound - v)
f_recede = f * (v_sound - v) / (v_sound + v)

Parameters:

• Velocity: 1-300 m/s
• Frequency: 100-2000 Hz
• Duration: 5 seconds
• Envelope: Triangular amplitude envelope

Features:

• Real-time slider updates
• Play/Pause generated sound
• Download as WAV file

🚙 Real Vehicle Analysis

Velocity Estimation

Upload real vehicle sounds for analysis

Algorithm Steps:

1. Band-pass filtering (100-800 Hz) - removes noise
2. Autocorrelation - finds dominant frequency
3. Sliding windows - analyzes approach and recede phases
4. Doppler formula - calculates velocity

Results Display:

• Velocity in m/s and km/h
• Approach frequency
• Receding frequency
• Noise filter toggle

🚁 Drone Detection

Detection Results

Frequency-based drone detection algorithm

Detection Method:

• Frequency range: 150-800 Hz (typical drone range)
• Window size: 200ms sliding windows
• Autocorrelation frequency estimation
• Average frequency across all windows

Output:

• ⚠️ DRONE DETECTED! (if frequency in range)
• ✅ No Drone Detected (if frequency outside range)
• Average frequency display

---

📈 Trading Signals

Overview

The trading module supports stocks, currencies, and minerals with multiple chart types and LSTM-based prediction.

📸 Trading Page

Main trading dashboard with category selection

📂 Asset Categories

Category	Chart Types	Example Assets
Stock Market	Candlestick + Volume, MA Overlay, % Comparison, Line Chart	AAPL, MSFT, GOOGL
Currency	Line Chart, Bollinger Bands, Rolling Volatility, MA Overlay	EUR/USD, GBP/USD, USD/JPY
Mineral	Candlestick, MA 50/200 Cross, Seasonality, Line Chart	Gold, Silver, Copper

📤 File Upload

Upload Interface

CSV upload with automatic format detection

Intelligent Parser Features:

• Detects Yahoo Finance format (skips first 2 rows)
• Handles standard CSV with headers
• Works with files without headers
• Automatic date column detection
• Extracts price data from various formats

📊 Chart Types in Action

Candlestick + Volume

OHLC candlesticks with volume bars

MA Overlay

Moving averages (20/50) overlaid on price

Bollinger Bands

Volatility bands with MA20 and ±2σ

Seasonality

Average price by month for minerals

🔮 LSTM Prediction

Prediction Interface

Forecast days selection and prediction button

Model Architecture:

• Input: 60-day sequences
• Features: open, high, low, close returns, volume change
• LSTM layers: 128 → 64 → 32 units
• Dropout: 0.2 after each LSTM layer
• Output: Next day's price

Smart Prediction Taming:

1. Calculate historical volatility
2. Apply realistic drift
3. Clip extreme predictions
4. Generate 95% confidence bands

Prediction Results

Historical data + forecast with confidence intervals

🔄 View Modes

Static Mode

Full historical data displayed

Over Time Mode

Animated window view with playback

Multi-View Mode

Four chart types displayed simultaneously

Currencies Multi-View

Over time view of all currency charts

Static view of all currency charts

Minerals Multi-View

Over time view of all mineral charts

Static view of all mineral charts

---

🧬 Microbiome Signals

Overview

The microbiome module analyzes microbial abundance data to predict disease states and calculate diversity metrics.

📸 Microbiome Page

Main microbiome analysis dashboard

📤 Data Upload

Upload Interface

TSV/CSV upload with sample selection

Supported Format:

• Rows: Samples/patients
• Columns: Bacterial taxa
• Last column: Diagnosis (CD, UC, Healthy, nonIBD)

🎯 AI Diagnosis

Diagnosis Results

AI-predicted disease with confidence score

Model Details:

• Algorithm: Random Forest (100 trees)
• Training data: iHMP dataset
• Classes: Crohn's Disease, Ulcerative Colitis, Healthy, nonIBD
• Features: 200+ bacterial taxa

📊 Scientific Metrics

Diversity Indices

Bar chart and radar visualization

Shannon Index (Alpha diversity):

H = -∑(p_i * ln(p_i))
where p_i = relative abundance of taxon i

F/B Ratio (Phyla balance):

• Firmicutes / Bacteroidetes
• Balanced range: 0.5 - 1.5
• Low (<0.5): Bacteroidetes dominance
• High (>1.5): Firmicutes dominance

Clinical Levels

Beneficial bacteria:

• Faecalibacterium
• Bifidobacterium
• Lactobacillus

Pathogen load:

• Escherichia
• Shigella
• Enterobacteriaceae

📋 Clinical Interpretation

Report Generation

AI-generated clinical report with color-coded advice

Report Components:

1. Diagnosis - Disease classification
2. Phyla Balance - F/B ratio interpretation
3. Summary - Personalized advice
4. Diversity - Shannon index value

---

🔧 Installation & Setup

Prerequisites

Python 3.8+
pip (Python package manager)
FFmpeg (for audio processing)
Git (optional)

Step 1: Clone Repository

git clone <repository-url>
cd SIGNAL_VIEWER

Step 2: Install Dependencies

pip install -r requirements.txt

requirements.txt:

flask==2.3.3
tensorflow==2.13.0
torch==2.0.1
transformers==4.35.0
wfdb==4.1.2
mne==1.5.1
librosa==0.10.1
scikit-learn==1.3.0
pandas==2.0.3
numpy==1.24.3
plotly==5.17.0
joblib==1.3.2
scipy==1.11.2

Step 3: Prepare Directory Structure

# Create data directories
mkdir -p Data/Acoustic\ Signals/{car,Drones}
mkdir -p Data/Medical\ Signals/{ECG\ Data,EEG}
mkdir -p Data/Microbiome\ Signals
mkdir -p Data/Trading\ Signals/{currencies,minerals,Stock}

# Create temp upload directories
mkdir -p temp_uploads_{ecg,micro,sound,eeg,acoustic}

Step 4: Add Model Files

Medical Models (Models/Medical/):

• hubert_ecg.py - Custom HuBERT implementation
• ecg_classifier.pkl - Trained MLP classifier
• model.safetensors - HuBERT weights
• config.json - Model configuration

Microbiome Models (Models/Microbiome/):

• microbiome_model.pkl - Random Forest classifier
• model_features.pkl - Feature names

Trading Models (Models/Trading/saved/):

• global_lstm_model.h5 - Trained LSTM weights
• global_lstm_model_scaler_X.pkl - Feature scaler
• global_lstm_model_scaler_y.pkl - Target scaler
• asset_mapping.json - Asset ID mapping

Step 5: Add Sample Data

Medical Data:

• Download MIT-BIH Arrhythmia Database samples
• Place in Data/Medical Signals/ECG Data/

Acoustic Data:

• Add vehicle pass-by recordings to Data/Acoustic Signals/car/
• Add drone recordings to Data/Acoustic Signals/Drones/

Trading Data:

• Download CSV files from Yahoo Finance
• Place in respective category folders

Microbiome Data:

• Download iHMP dataset
• Save as Data/Microbiome Signals/iHMP_data.csv

Step 6: Run Application

python app.py

Step 7: Access Web Interface

Open browser and navigate to:

http://127.0.0.1:5000

---

🎮 Usage Guide

Medical Viewer Workflow

1. Select Signal Type

   • Choose ECG or EEG from dropdown

2. Upload Data

   • Click "Choose Files"
   • Select .hea/.edf/.csv files
   • Wait for processing

3. Select Channels

   • Ctrl+Click to select multiple
   • Customize colors/thickness

4. Choose Viewer Type

   • Signal (time-domain)
   • Polar (magnitude vs time)
   • XOR (difference detection)
   • Recurrence (2D scatter)

5. Run AI Analysis

   • Click "Run AI Prediction"
   • Compare DL vs Classic results

6. Control Playback

   • Adjust speed slider
   • Adjust window size
   • Click Play/Stop

Acoustic Analysis Workflow

For Simulation:

1. Adjust velocity and frequency
2. Click "Generate Sound"
3. Use Play/Pause to listen
4. Click "Download" to save

For Vehicle Analysis:

1. Upload vehicle audio file
2. Toggle noise filter if needed
3. Click "Analyze Velocity"
4. View estimated speed and frequencies

For Drone Detection:

1. Upload test audio file
2. Click "Run Detection"
3. View detection result

Trading Dashboard Workflow

1. Select Category

   • Stock Market / Currency / Mineral

2. Load Data

   • Upload CSV or use pre-loaded
   • View file info panel

3. Choose View Mode

   • Single (one large chart)
   • Multi (four charts)

4. Select Chart Type

   • Based on category selection

5. Set Display Mode

   • Static (full history)
   • Over Time (animated window)

6. Run Prediction

   • Set forecast days
   • Click "Predict Future Behavior"
   • View forecast with confidence bands

Microbiome Analysis Workflow

1. Upload File

   • Select TSV/CSV file
   • Wait for sample extraction

2. Select Sample

   • Choose from dropdown
   • Click "Run AI Analysis"

3. Review Results

   • AI Diagnosis with confidence
   • Shannon diversity index
   • F/B ratio
   • Beneficial/pathogen percentages

4. Examine Visualizations

   • Top taxa bar chart
   • Microbial radar plot
   • Clinical interpretation report

---

💻 Technical Implementation

Backend Architecture (app.py)

# Key routes and their functions

@app.route('/upload')              # Medical signal upload
@app.route('/analyze_signal_ai')    # AI analysis for medical
@app.route('/upload_sound')         # Acoustic file upload  
@app.route('/micro_upload')         # Microbiome data upload
@app.route('/analyze_micro_sample') # Microbiome analysis
@app.route('/api/upload')           # Trading data upload
@app.route('/api/predict')          # LSTM prediction

Medical Signal Processing

# Feature extraction with HuBERT
with torch.no_grad():
    outputs = ECG_AI_MODEL(sig_tensor)
    features = outputs.last_hidden_state.mean(dim=1)

# Classic ML detection
peaks, _ = scipy.signal.find_peaks(work_sig_norm, 
                                   distance=int(0.4 * fs))
rr_intervals = np.diff(peaks) / fs
sdnn = np.std(rr_intervals)
rmssd = np.sqrt(np.mean(np.diff(rr_intervals) ** 2))

Acoustic Signal Processing

# Band-pass filter implementation
def bandPassFilter(segment, sampleRate, lowHz, highHz):
    RC_low = 1 / (2 * np.pi * lowHz)
    RC_high = 1 / (2 * np.pi * highHz)
    dt = 1 / sampleRate
    # Filter logic...
    
# Autocorrelation frequency estimation
def estimateFrequency(segment, sampleRate):
    for lag in range(minLag, maxLag):
        corr = np.sum(segment[:-lag] * segment[lag:])
        if corr > maxCorr:
            maxCorr = corr
            bestLag = lag

Trading LSTM Model

class GlobalLSTM:
    def build_model(self):
        model = Sequential([
            Input(shape=(sequence_length, n_features)),
            LSTM(128, return_sequences=True),
            Dropout(0.2),
            LSTM(64, return_sequences=True),
            Dropout(0.2),
            LSTM(32),
            Dropout(0.2),
            Dense(16, activation='relu'),
            Dense(1)
        ])

Frontend Visualization (Plotly.js)

// Dynamic chart creation
function renderChart(divId, type, data, predictionData) {
    let traces = [];
    
    if (type === 'candlestick') {
        traces.push({
            x: dates,
            open: data.prices.open,
            high: data.prices.high,
            low: data.prices.low,
            close: data.prices.close,
            type: 'candlestick'
        });
    }
    
    Plotly.newPlot(divId, traces, layout);
}

---

🎥 Demo Video

A comprehensive demonstration video is available:

📹 Watch the Demo Video

---

📝 Conclusion

BioSignal AI successfully implements all requirements of Task 1 with:

Category	Requirements Met	Key Achievement
✅ Medical	14/14	Multi-channel ECG/EEG with 4 viewer types + AI comparison
✅ Acoustic	7/7	Doppler simulation + real vehicle/drone analysis
✅ Trading	7/7	Multi-category charts + LSTM prediction
✅ Microbiome	6/6	Disease profiling + diversity metrics + clinical reports

Key Highlights:

• Unified interface for four distinct signal types
• Real-time visualization with playback controls
• AI-powered analysis (HuBERT, LSTM, Random Forest)
• Classic algorithms for comparison
• Comprehensive customization options
• Professional documentation with 40+ screenshots
• Demo video showcasing all features

---

📝 License

TThis project is created for educational purposes as part of the Digital Signal Processing Course - Task 01: Signal Viewer with Basic Processing.

Course: Digital Signal Processing Task1: Signal Viewer with Basic Processing Semester: Spring 2026 Team: 08

---

🌟 Thank you for reviewing our project! 🌟

For questions or support, please contact Team 08

---