Sybil-Resilient Clustering for Cognitive Radio Networks

Cooperative sensing can generally be classified into two classes: hard-decision-based, and soft-decision-based.  In hard-decision cooperative sensing protocols, each radio makes a binary decision on whether a channel is occupied by the primary users; whereas in soft-decision cooperative sensing protocols, each radio reports how confident it is that the channel is occupied by the primary users.

Since soft-decision cooperative sensing gathers more information, in a network with only honest radios, soft-decision cooperative sensing must outperform hard-decision cooperative sensing.  However, soft-decision cooperative sensing is particularly susceptible to false-report attacks, in which an attacker reports incorrect confidence.  For example, an attacker that constantly claims that he is 100% sure the channel is occupied can render the cognitive radio network useless by triggering primary-avoidance.  The community has thus proposed several outlier-detection-based protocols in hope to mitigate the false-report attack.  The outlier-detection protocols are, in one way or another, analogous to voting schemes where the radios that do not conform to the crowd are ignored.  Thus, like all voting schemes, outlier-detecting cooperative sensing is susceptible to the Sybil attack.

We at ReCognize propose countering the false-reporting-Sybil attack by using Sybil-resilient clustering.  We first cluster the cognitive radios based loosely on their geographical placement such that the cluster heads are mostly benign.  Then since one Sybil attacker (and his many other identities) can only occupy one location at one time, all Sybil identities are likely clustered into only a few clusters.  We then propose that the cluster heads reach intra-cluster sensing decision based on soft-decision so as to take advantage of the improved sensing capability; and that the fusion center reach inter-cluster decision based on hard-decision so as to avoid being overwhelmed by the Sybil identities.