Vibration Dataset

Description:

In the on wrist scenario, we collect device credentials with different settings on the smartwatches to evaluate the efficacy and robustness of the system. In particular, we conduct experiments with 5 different vibration patterns, and 3 jamming attacks under different sound noises. In total, we have two participants collecting around 600 device credentials in the on desk scenario and 1680 device credentials in the on wrist scenario across over 4 weeks. Unless stated otherwise, we use the vibration strength of 50 with 1 second vibration duration and 1 second sleep duration to generate vibrations. We use the maximum sampling rate of the smartwatches' accelerometers (i.e., 50Hz) to collect data. We randomly select 30 device credentials from a legitimate device and 30 device credentials from the other four watches (as attackers) to construct a training dataset. The rest of the data (i.e., 90 device credentials from the legitimate user and 90 device credentials from the attacker) is used for testing. We repeat the training and testing five times and use the average results to evaluate our system's performance.

Format: 

.csv format

Section 1: Device Configuration:

Our vibration data is collected on  five commodity smartwatches (Three Fossil Gen 5 watches and two Moto 360 Gen 3 watches). Specifically we use Google Wear OS (version 2.27) Google: Wear OS on these smartwatches to change the vibration strength of the built-in vibration motors within a range of 0 to 255 and vibration durations. We place each watch on a wooden table with its face up and program the app to keep the watch still for 1 second and vibrating for 2 second with the vibration strength set to 50. Meanwhile, the app uses the watch's accelerometers to capture the vibration signals using their maximum sampling rate of 50Hz. We repeat the same vibration pattern for comparison.

Section 2: Data Format:

The collected device credentials are downloaded to a desktop as CSV type format file. 

Data file format rule:

mmddyy_dayversion_participantName_scenario_devicemodel.csv

Example:

020623_01_pA_desk_fossil.csv

data recorded on: Feb 6th,2023

daily try version: 01

participant id: A

scenario: on desk

device model: Fossil smartwatch

Section 3: Experimental Setups:

We use three Fossil Gen 5 and two Moto 360 Gen 3 smartwatches to evaluate the performance of WatchID. We evaluate the system under two scenarios: on desk and on wrist for practical usage situations. In the wrist scenario, we collect the vibration-based device credentials of each smartwatch when it is  fixed on the desk, while in the second, we carry the operation when the watch is worn on a human wrist.

In the on desk scenario, we focus on studying the efficacy of the vibration-based device credentials. For each smartwatch, we collect 120 device credentials.

In the on wrist scenario, we collect device credentials with different settings on the smartwatches to evaluate the efficacy and robustness of the system. In particular, we conduct experiments with 5 different vibration patterns and 3 jamming attacks under different sound noises. In total, we have two participants collecting around 600 device credentials in the on desk scenario and 1680 device credentials in the on wrist scenario across over 4 weeks.

Section 4: Data Description:

• fossil1.csv: Collected vibration signals from Smartwatches(WearOS), including accelerometer readings of x,y,z axe

• fossil1_filtered.csv: Noise-removed data

• Column 1: x axes

  • accelerometer readings of x

• Column 2: y axes

  • accelerometer readings of y

• Column 3: z axes

  • accelerometer readings of z

• Column 4: label

• Column 5: model

  • device model

Section 5: Codes

• Signal Denoising.ipynb: Applying a Butterworth bandpass filter to mitigate noise.
• 1D CNN.ipynb: A simple 1-dimensional Convolutional Neural Network, including Max-Min normalization, signal segmentation and classification result using 1D CNN.
• fossil1.h5: Saved CNN model.

Section 6: Citations

Cite this paper

Cheng, J.Q., Wang, Z., Wang, Y., Zhao, T., Wan, H., Xie, E. (2022). WatchID: Wearable Device Authentication via Reprogrammable Vibration. In: Hara, T., Yamaguchi, H. (eds) Mobile and Ubiquitous Systems: Computing, Networking and Services. MobiQuitous 2021. Lecture Notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering, vol 419. Springer, Cham. https://doi.org/10.1007/978-3-030-94822-1_53