Security in Cognitive Radio Networks: Threats and Mitigation

This paper describes a new class of attacks specific to cognitive radio networks. Wireless devices that can learn from their environment can also be taught things by malicious elements of their environment. By putting artificial intelligence in charge of wireless network devices, we are allowing unanticipated, emergent behavior, fitting a perhaps distorted or manipulated level of optimality. The state space for a cognitive radio is made up of a variety of learned beliefs and current sensor inputs. By manipulating radio sensor inputs, an adversary can affect the beliefs of a radio, and consequently its behavior. In this paper we focus primarily on PHY-layer issues, describing several classes of attacks and giving specific examples for dynamic spectrum access and adaptive radio scenarios. These attacks demonstrate the capabilities of an attacker who can manipulate the spectral environment when a radio is learning. The most powerful of which is a self-propagating AI virus that could interactively teach radios to become malicious. We then describe some approaches for mitigating the effectiveness of these attacks by instilling some level of "common sense" into radio systems, and requiring learned beliefs to expire and be relearned. Lastly we provide a road-map for extending these ideas to higher layers in the network stack.