EM Eye

This repository is the tutorial of eavesdropping on camera video from camera circuit's electromagnetic (EM) leakage, which is presented in the NDSS'24 paper "EM Eye: Characterizing Electromagnetic Side-channel Eavesdropping on Embedded Cameras" (contact: [email protected] and/or [email protected]). Please also refer to the research project introduction website.

To enable researchers to quickly reproduce the observations in the paper on different camera devices, we modified the TempestSDR tool into TempestSDR_EMEye. Specifically, range of eavesdropping parameters was adjusted to fit the scope of embedded cameras.

To help readers understand how this eavesdropping process works, we also provide a MATLAB simulation program.

TempestSDR_EMEye: Bill of Materials

Hardware:

• A decent laptop
• Software-defined radio: Ettus USRP B210 USRP
• Low noise amplifier: Foresight Intelligence FSTRFAMP06
• Antenna: A common outdoor Log-periodic directional antenna (LPDA) for long distances or a cheap near-field magnetic field probe for short distances.

Note that most hardware models can be changed. For example, the USRP can be replaced by cheaper SDR devices such as RTL-SDR. If you have expertise in software-defined radio, feel free to try different sets of hardware equipment.

Software:

• Operating System: Ubuntu/macOS/Debian (Bookworm)
• UHD 4.4.0.0 driver provided by Ettus Research
• libuhd-dev 4.4.0
• openjdk-8-jdk or OpenJDK 17.0.10
• TempestSDR_EMEye in this repo. Compile it by running "make all JAVA_HOME=/usr/lib/jvm/java-8-openjdk-amd64/" (change the path to your own jdk path) in the the JavaGUI/ folder

We suggest keeping the software models the same as ours. If you are compiling TempestSDR_EMEye with other software environments, you should read the instructions on the TempestSDR page.

Make sure your USRP is connected and you successfully ping it. Now the test environment is all set up.

TempestSDR_EMEye: Usage

You can go to the JavaGUI/ folder and start TempestSDR_EMEye by running "java -jar JTempestSDR.jar". You will see the GUI pop up. Go to File -> Load USRP (via UHD), and input "--rate=10000000" to set the sampling rate to 10 MHz. Then configure the eavesdropping reconstruction algorithm by setting Disp Width, Real Height, FPS, and Freq based on this table below.

Tested Devices:

No.	Vendor and Model	Disp Width	Real Height	FPS	Freq. (MHz)
1	Raspberry Pi Camera Module 1	2576	1126	30.01985577	204 or 255
2	Raspberry Pi Camera Module 2	2576	1790	29.55323582	457 or 570
3	Raspberry Pi Camera Module 3	2166	1244	30.08283973	335 or 1015
4	Google Pixel 1	2800	1557	29.89235398	600 or 1649
5	Google Pixel 3	6000	3055	30.0610185	515 or 680
6	Samsung S6	5000	3173	30.01527799	527 or 1054
7	ZTE Z557	4000	2453	23.98017526	522 or 1740
8	Wyze Cam Pan 2	2300	1687	20.0071563	890 or 1185
9	Xiaomi Dafang IP Camera	2567	2250	15.0008813	322 or 890
10	Baidu Xiaodu X9	2576	1250	15.0000375	204 or 1470
11	TeGongMao	2576	1500	15.15164801	763 or 1144
12	Goov V9	2000	750	30.30325987	546 or 656
13	QiaoDu	2576	2245	15.04149722	293 or 1191
14	360 M320 Dashcam	2620	1818	25.00241735	450 or 1261
15	Blackview Dashcam	2567	1196	30.09970806	155 or 1015

The configuration parameters are for specific camera modes. You may need to adjust the parameters when testing other modes.

• Raspberry Pi cameras were tested while they are in the 1080p mode (-md 1).
• Smartphone cameras (rear/main cameras) were tested in the video mode.
• Dash cams and home cameras were tested in their default modes.

Note that individual devices of the same model may require slightly different parameters than the ones reported in the table due to hardware manufacturing and software version variations. You may need to fine-tune the frequency, FPS, and height, and even the SDR sampling rate to get the best results. The width parameter only controls the scaled aspect ratio of the displayed reconstruction and does not have any impact on the EM reconstruction process. We refer you to the dissertation of Martin Marinov, the author of TempestSDR, for more information on how these parameters work.

Eavesdropping GUI:

Press the star button. You will see something similar to the figure below. The upper-left window displays the real-time reconstruction of the camera video.

MatlabSimulation_EMEye

The simulation program is in the MatlabSimulation_EMEye/ folder of this repo.

The program converts a RAW image frame (room.dng) captured by a Raspberry Pi Camera V1 into the bit stream transmitted over the MIPI CSI-2 2-lane interface used by the camera sensor, and then reconstructs the image from the simulated corresponding EM signals of the bit stream.

Citation Information

If you find our results useful to your research, please cite it as follows:

@inproceedings{long2024emeye,
    title={EM Eye: Characterizing Electromagnetic Side-channel Eavesdropping on Embedded Cameras},
    author={Long, Yan and Jiang, Qinhong and Yan, Chen and Alam, Tobias and Ji, Xiaoyu and Xu, Wenyuan and Fu, Kevin},
    booktitle={Network and Distributed Systems Security (NDSS) Symposium},
    year={2024}
    }