Description:

Mechanism for Real-Time Channel-Resilient Optimization of
Deep Learning-based Radio Fingerprinting Algorithm

Background

A core security challenge for the Internet of Things is to devise reliable and energy-efficient authentication techniques. Most existing authentication mechanisms are not well-suited to the IoT since they rely heavily on cryptography-based algorithms and protocols which are difficult to run on tiny, energy-constrained IoT devices.

Radio fingerprinting addresses this problem by leveraging the unique hardware-level imperfections imposed on the received wireless signal by the transmitter's radio circuitry, avoiding energy intensive cryptography at the transmitter's side. Traditional techniques for radio fingerprinting rely on complex feature-extraction techniques that leverage protocol-specific characteristics to extract hardware impairments. This is of limited use in IoT applications where a number of different wireless protocols are used.

Technology Overview

To overcome the limitations of existing radio fingerprinting techniques, researchers at Northeastern have developed DeepRadioID, a system which utilizes deep learning to design general-purpose, high-performance, and optimizable radio fingerprinting algorithms.

There are several critical challenges in applying deep learning techniques to RF fingerprinting. First, deep learning models usually require a significant time to be re-trained, even with modern GPUs. Second, a fingerprinting system must evaluate the impact of adversarial actions to determine the ability of an adversary to imitate a legitimate device’s fingerprint. Lastly, it is critical to address the action of the wireless channel on the system’s fingerprinting accuracy. As a result of channel action, two identical waveforms transmitted by the same RF interface at two different moments in time are usually different from each other. This significantly decreases the model’s fingerprinting accuracy when the classifier is used on data collected with a wireless channel that is significantly different from the one used to train it.

A key aspect of this innovation is the application of a carefully-optimized digital finite input response filter (FIR) at the transmitter's side. The inventors apply tiny modifications to the waveform to strengthen its fingerprint according to the current channel conditions. Since the FIR is tailored to the specific device's hardware imperfections, they can demonstrate that an adversary will not be able to use the FIR to imitate a legitimate device's fingerprint.

Key Benefits

  • Optimizes radio fingerprinting accuracy of wireless devices
  • Decreases ability of adversary to imitate a devices fingerprint
  • Works with hundreds of devices across wireless protocols
  • No requirement for retraining the underlying deep learning model
  • Can be implemented on any off-the-shelf device with no additional cost

Commercial Applications

  • Radio Fingerprinting Optimization of Wireless Devices
  • Military/DoD applications

Development Status

The research team has extensively evaluated DeepRadioID on an experimental testbed of 20 nominally identical software-defined radios, as well as on two datasets made up by 500 ADS-B devices and by 500 WiFi devices. Experimental results show that DeepRadioID (i) increases fingerprinting accuracy by about 35%, 50% and 58% on the three scenarios consideredÍž and (ii) decreases an adversary's accuracy by about 54% when trying to imitate other device's fingerprints by using their filters.

 

 

Patent Information:
Category(s):
-Communications
For Information, Contact:
Mark Saulich
Associate Director of Commercialization
Northeastern University
m.saulich@northeastern.edu
Inventors:
Keywords: