User Authentication Using WiFi-enabled IoT

Project Objectives

In recent years, user authentication has become increasingly vital due to the growing concern of user security and privacy leakage. In addition, the emerging applications in smart homes/offices are also exploring the ability to identify users and launch customized services (e.g., adjusting room temperature and lighting conditions) according. In this project, we aim to develop a device-free system that can perform user authentication by leveraging wireless signals provided by smart and IoT devices.

Technology Rationale

  1. Unique Physiological and Behavioral Characteristics. The advent of Internet of Things (IoT) over the past decade makes every electronic in indoor environments interconnected wirelessly. The wireless connection among IoT devices provides a rich web of reflected rays that spread every indoor corner, which provides rich opportunities for us to leverage the wireless signals to perform user authentication. We find that the wireless signals can capture human’s unique physiological and behavioral characteristics inherited from people’s daily activities when operating the smart and IoT devices (e.g., opening the refrigerator, entering the restricted office).

  2. Adapting to Environmental Changes. We face two key challenges to perform user authentication based on wireless signals: (i) small-scale location variations: a user usually conducts the same behavior in proximity with small location variations; and (ii) large-scale environmental changes: The status of a physical environment (e.g., the placements of furniture and home appliances) could vary from day to day in practical scenarios. Our system needs to mitigate these two issues for practical user authentication.

Technical Approach

  1. Unique physiological and behavioral characteristics. We propose to extract representative features based on both amplitude and phase information in of channel state information (CSI) measurements in wireless signals, which have the potential to reveal unique physiological and behavioral biometrics of different users. Furthermore, a three-layer deep neural network (DNN) model is developed to learn high-level abstractions of human physiological and behavioral characteristics for both activity recognition and human identification. In particular, the DNN scheme detects the activity type (i.e., stationary or walking) in the first layer and obtains the activity details (e.g., walking paths, opening a refrigerator) in the second layer. In the third layer, the model can learn the highest level non-linear abstractions from the representative features obtained from human activities and authenticate the user accordingly. 

  2. Adapting to Environmental Changes. To address these two issues encountered in practice, we develop a deep learning model together with an unsupervised domain discriminator, which are built upon both labeled data (i.e., WiFi signals containing user behaviors with user-identity labels) and unlabeled data (i.e., user behavior data without labels), to mitigate the impact of varying location and environmental changes and achieve reliable user authentication. This developed deep learning model could be utilized in a new environment when working with the collected user behaviors (i.e., unlabeled data) to perform user authentication and thus to achieve environment independency.

Project Status

This project has led to two conference papers in MobiHoc’17, MASS’20 and one journal paper in ACM TIOT. We report the performance of our system using unique physiological and behavioral characteristics for user authentication in Figure 1. We also show the system’s robustness by comparing our domain-discriminator-based approach with a baseline model based on a convolutional neural network (CNN), and a traditional domain adaptation approach grounded on transformable component (TCA) analysis. 

  1. Unique physiological and behavioral characteristics.
  1. Adapting to environmental changes.

References

Cong Shi, Jian Liu, Hongbo Liu, Yingying Chen. Smart User Authentication through Actuation of Daily Activities Leveraging WiFi-enabled IoT. In Proceedings of the 18th ACM International Symposium on Mobile Ad Hoc Networking and Computing (ACM MobiHoc), pp. 1-10, 2017.

Cong Shi, Jian Liu, Nick Borodinov, Bruno Leao, Yingying Chen. Towards Environment-independent Behavior-based User Authentication Using WiFi. In Proceedings of the 17th IEEE International Conference on Mobile Ad-Hoc and Smart Systems (IEEE MASS), pp. 666-674, 2020.

Cong Shi, Jian Liu, Hongbo Liu, Yingying Chen. WiFi-enabled User Authentication through Deep Learning in Daily Activities. ACM Transactions on Internet of Things (ACM TIOT), No. 165, pp 1–25, 2021.