WLAN and Zigbee Signals Classification Using AI 📡

Table of Contents

• Project Overview
  • YouTube Link
• Classification
  • Zigbee
  • WLAN
  • Background (Noise)
• Hardware
• Installation and Usage
  • Prerequisites
  • Steps
• References

Project Overview

The 2.4 GHz ISM band is a busy place, with protocols like Wi-Fi and Zigbee constantly vying for space. In environments like smart homes, industrial IoT, or crowded offices, distinguishing between these signals can be a significant challenge.

This project addresses that problem by developing a lightweight, AI-based classifier that identifies Zigbee and Wi-Fi signals from spectrogram images. Built with MATLAB and deep learning, the system classifies time-frequency spectrograms into three categories: Zigbee, WLAN, and Background. The solution supports both synthetic signal generation and live signal classification using an ADALM-PLUTO SDR.

This project is designed as part of the Mathworks AI Challenge.

YouTube Link

Classification

Zigbee

To evaluate the model's performance, each signal is first generated using the MATLAB Wireless Waveform Generator and then transmitted over the air using the ADALM-PLUTO SDR.

The figure below shows a Zigbee signal prepared for transmission:

The following image presents the classification results for the Zigbee signal.

• Top left: The received spectrogram captured by the ADALM-PLUTO SDR.
• Bottom left: Pixel-wise classification results. Zigbee pixels are correctly labeled in pink, and background noise is labeled in blue.
• Right: The frequency mask highlights the detected Zigbee band in pink, consistent with the transmitted signal (centered at 2.4 GHz, 2 MHz wide).

WLAN

The next figure shows the WLAN signal generated for transmission:

The following classification result shows:

• Top left: The received WLAN signal from the ADALM-PLUTO SDR.
• Bottom left: Pixel classification, where WLAN regions are marked in purple and background in blue.
• Right: The model identifies the WLAN operating band in gray, correctly corresponding to the transmitted signal (centered at 2.4 GHz, 20 MHz wide).

Background (Noise)

To verify that the model is not producing false positives, it's also tested without transmitting any signal. As shown below, the model correctly labels the entire frame as background (blue), indicating no active Zigbee or WLAN signals:

Hardware

• ADALM-PLUTO Software Defined Radio (SDR)

Installation and Usage

Prerequisites

• Communications Toolbox
• Deep Learning Toolbox
• Signal Processing Toolbox
• Image Processing Toolbox
• Computer Vision Toolbox
• WLAN Toolbox
• Parallel Computing Toolbox
• ADALM-PLUTO Radio Support from Communications Toolbox

Steps

1. Clone the repository:

   git clone https://github.com/Gonzaleski/WLAN-Zigbee-Classifier.git
   

2. Change the directory:

   cd WLAN-Zigbee-Classifier
   

3. Open MATLAB

4. Generate the training data:

   • Go to scripts/dataGeneration
   • In MATLAB, run generateData.m

5. Train the model:

   • Go to scripts/training
   • In MATLAB, run trainModel.m

6. Connect the SDR to the laptop

7. Initialize the SDR:

   • Go to scripts/testing
   • In MATLAB, run initializeSDR.m

8. Run the test and see the results:

   • Go to scripts/testing
   • In MATLAB, run testWithSDR.m

References

• Create Waveforms Using Wireless Waveform Generator App
• Semantic Segmentation Using Deep Learning
• What is ZigBee Protocol?
• What is 5 GHz Network? List of Devices Compatible with 5 GHz Network
• Popular technologies operating in the 2.4 GHz ISM band
• Spectrum Sensing with Deep Learning to Identify 5G, LTE, and WLAN Signals
• ADALM-PLUTO_SDR
• List of WLAN channels
• Home Assistant