The latest video from Intellisense Systems shows two demonstrations of RADEC – the Radio Frequency Ensemble Classifier – which provides trustworthy A.I. to combat adversarial attacks in electromagnetic spectrum operations.
Today, much of modern warfare is waged on the electromagnetic spectrum. Every air, maritime, and ground platform has radars, communications, and a radio frequency (RF) signature. The abilities to identify and manipulate these signatures can provide an enormous advantage both on the physical battlefield and in an electromagnetic spectrum fight. To maintain the U.S.’s all-domain advantage, the Department of Defense (DoD) is changing its approach to a more unified treatment for electromagnetic spectrum operations.
In this arena, U.S. Armed Forces are constantly challenged by both peers and near-peer adversaries in the electromagnetic spectrum. These challenges have exposed the cross-cutting reliance of U.S. forces on the electromagnetic spectrum and are driving changes in how the DoD approaches these activities to maintain its combat advantages. The U.S. and its adversaries are increasingly leveraging deep neural networks for RF communications because of their improved accuracy over traditional approaches; however, they are vulnerable to adversarial attacks that use highly optimized noise masks to deceive RF classifiers. These attacks are imperceptible to humans and RF decoders.
The Radio Frequency Ensemble Classifier (RADEC) provides trustworthy A.I. that is resilient to these attacks. RADEC’s hardened ensemble RF classifier is trained to see through and detect adversarial perturbations in the data and correctly recover the true class, thus avoiding deception.
This video features two demonstrations of RADEC’s offensive and defensive capabilities. In the first scenario, an enemy surface reactive jammer is set to initiate jamming when its DNN classifier detects a specific protocol or mod scheme, such as one the one used by an unmanned aerial system. In this demonstration, the RADEC’s hardened classifier is not affected by Gaussian noise transmitted by software-defined radios. This demo also illustrates the program’s smart jamming capabilities, wherein the enemy perturbation is transmitted by a third independent device over-the-air to fool all RF classifiers in the vicinity.
In the second demo, UAS and ground sensors are surveying and collecting signals of interest to identify, locate, and track enemy targets. The enemy attempts to confuse the DNN classifiers using the same adversarial signals. Again, RADEC detects enemy perturbations and hardens the classifier against such attacks.