CSI ESP32 Tool Kit

This project provides tools for capturing and classifying Channel State Information (CSI) data using an ESP32 device for user authentication.

Overview

The system captures CSI data from an ESP32 via UART and processes it for multi-class user authentication using machine learning models. It is adapted from the sbrc2024-csi repository for use with ESP32.

Experimental Setup

The experiments were conducted to evaluate the feasibility of using Channel State Information (CSI) for user authentication in Wi-Fi environments, considering the impact of environmental noise. For this purpose, an experimental setup was used based on two ESP32 devices: an ESP32 DEVKIT V1 model for capturing CSI data and an ESP32 C3 Super Mini model for sending Wi-Fi frames. The samples collected from this structure are made possible through Espressif's ESP-CSI framework.

Each CSI sample contains three hundred and eighty-four (384) floating-point values, corresponding to 192 complex I/Q samples (I = in-phase component, Q = quadrature component), which are converted into complex numbers (I + jQ). These 192 complex samples are organized into 64 subcarriers for each of the 3 simulated spatial streams by the ESP32 DevKit V1 firmware, even though the hardware has only one physical antenna. Thus, the data structure consists of 64 subcarriers multiplied by 3 antennas, totaling 192 complex samples, whose I and Q values are interleaved among the 384 floating points of each CSI sample. The data was collected in a controlled environment with three distinct users: Aline, Giovani, and Ines. As a criterion, the user had to be close to the receiving ESP32, in a static position with their hand over the device. The collection was performed using the capture_csi.py script, which captures raw CSI data via the ESP32's UART serial port and stores it in a CSV file. Each capture includes metadata such as user, position, and environment, ensuring data traceability.

After collection, the data was processed with the classifier_csi.py script, which converts raw CSI values into amplitudes, filters selected users, encodes labels, splits the data into training and test sets (with a stratified split of 50%), normalizes features, and trains multiple classifiers. The evaluated models include Random Forest, SVC, KNN, Logistic Regression, and Gradient Boosting. It is important to note that, to date, no filters have been applied to remove null or pilot subcarriers.

Requirements

• Python 3.8 or higher
• Packages listed in requirements.txt
• ESP-IDF environment configured
• ESP32 running the CSI firmware (via ESP-CSI)

File Structure

File	Description
capture_csi.py	Captures CSI data via UART from ESP32 and saves it to a CSV file
classifier_csi.py	Processes CSI data and performs user classification using machine learning models

How to Use

1. Install Dependencies

Create a virtual environment (recommended) and install the required packages:

python -m venv venv
source venv/bin/activate  # Linux/Mac
venv\Scripts\activate     # Windows
pip install -r requirements.txt

2. Capture CSI Data

Run the following command to capture CSI data from the ESP32 and save it to a CSV file:

python capture_csi.py --port /dev/ttyUSB0 -t 60 -u giovani -o data/data_csi.csv

• --port: Serial port where the ESP32 is connected (e.g., /dev/ttyUSB0 or COM3 on Windows).
• -t: Capture duration in seconds (e.g., 60 for 60 seconds).
• -u: User associated with the capture (required, e.g., giovani).
• -o: Output CSV file (default: data/data_csi.csv).

3. Classify Users

Run the classifier to process the CSI data and perform user authentication:

python classifier_csi.py

This script processes the CSV file (data/data_csi.csv by default), trains machine learning models (RandomForest, GradientBoosting, SVC, KNN, LogisticRegression), and generates visualizations such as PCA, t-SNE, confusion matrices, and ROC curves.

Based On

• sbrc2024-csi

License

MIT