FMCW-Radar-ML-Classification

📘 Overview

This project implements a software-only FMCW radar simulator in MATLAB and combines it with a simple machine learning classifier to distinguish between three target types: Human, Car, and Drone using their Doppler / micro-Doppler signatures.

The project is designed as a compact RF + ML portfolio piece showcasing radar signal processing, feature engineering, and classification — all done in MATLAB.

---

⚙️ Features

🔹 FMCW Radar Signal Chain

• 77 GHz carrier frequency (typical automotive radar band)
• 150 MHz sweep bandwidth
• Linear FMCW chirp generation
• Range FFT (fast-time processing)
• Doppler FFT (slow-time processing)
• Range–Doppler Map visualisation
• Micro-Doppler-like spectrogram at a selected range bin

🔹 Machine Learning Classification

• Synthetic classes: Human, Car, Drone
• Velocity profiles generated to mimic each class:
  • Human: low speed, strong micro-motion
  • Car: higher average speed, smoother profile
  • Drone: moderate speed with oscillatory hovering motion
• RF-inspired features extracted from velocity:
  • Mean velocity
  • Standard deviation of velocity
  • Maximum absolute velocity
  • Mean absolute derivative of velocity
• Multi-class SVM classifier (fitcecoc) trained and evaluated
• Prints overall classification accuracy in the MATLAB command window

---

🧩 File Summary

File / Folder	Description
fmcw_radar_ml_project.m	Main MATLAB script: FMCW radar simulation + ML classification
figures/	Folder containing exported plots (q1, q2, q3)
figures/q1.png	Range–Doppler Map (Range vs Velocity)
figures/q2.png	Micro-Doppler-like spectrogram at a chosen range bin
figures/q3.png	Feature space scatter plot (Human / Car / Drone)

---

🧪 Tools & Environment

• MATLAB (tested with R2024b; works with most recent versions)
• Recommended Toolboxes:
  • Signal Processing Toolbox
  • Statistics and Machine Learning Toolbox

No external hardware is required — everything runs as a simulation.

---

📊 Results

1️⃣ Range–Doppler Map

The first figure shows the Range–Doppler Map, where each bright region corresponds to a simulated target at a particular range and radial velocity.

Range–Doppler Map

---

2️⃣ Micro-Doppler-Like Signature

At the strongest range bin, the script extracts the slow-time signal and computes a spectrogram, giving a micro-Doppler-like signature that reflects the velocity variations of the target over time.

Micro-Doppler Spectrogram

---

3️⃣ ML Feature Space

The final figure shows the feature space (e.g., mean vs max velocity) with different classes labelled, giving intuition about how well the RF-inspired features separate Human, Car, and Drone motion.

Feature Space (Human / Car / Drone)

The script prints the classification accuracy in the MATLAB command window, giving a quick sense of how well the simple SVM model performs on the synthetic dataset.

---

▶️ How to Run

1. Open MATLAB.
2. Add this project folder to the MATLAB path or set it as the Current Folder.
3. Open fmcw_radar_ml_project.m.
4. Click Run (or press F5).
5. Three figures will be generated:
   • Range–Doppler map
   • Micro-Doppler spectrogram
   • Feature space scatter
6. Optionally, save the figures into the figures/ folder as q1.png, q2.png, q3.png.

---

📡 Applications

• Radar signal processing education and demos
• RF + ML portfolio for automotive / sensing roles
• Feature-engineering concepts for Doppler / micro-Doppler analysis
• Baseline project for extending to real hardware or more advanced ML models

---

Future Work

• Add CFAR detection
• Add MIMO virtual array simulation
• Add deep-learning based micro-Doppler classifier
• Export dataset for Python-based ML

---

👤 Author

Brian Rono
Electrical & Computer Engineer • RF & Wireless • Embedded Systems • Machine Learning
🔗 GitHub Profile