Recursive validation and clustering for distributed spectrum sensing in CR-MANET

In cognitive radio networks, secondary users need to accurately identify primary user spectrum occupancy in order to use it. Accurate spectrum sensing is hindered by signal fading, hidden terminal problems, byzantine failures, etc. Centralized cooperative spectrum sensing works well if the secondary user network is infrastructure based and there is a centralized basestation making network wide decisions. When the secondary users network is a cognitive radio mobile ad-hoc network (CR-MANET), then decisions need to be made in a distributed manner and cooperative spectrum sensing introduces additional problems due to the presence of malicious users. These malicious secondary users encourage other secondary users to make a wrong spectrum occupancy decision by feeding inaccurate measurements. We study this problem and present a solution to improve primary user spectrum occupancy identification accuracy in the presence of malicious users. A virtual neighbor cluster is created in which the mobile device forms an evolving cluster of past neighbor devices that aids in validating the input gathered from the current neighboring devices. Next, a recursive partitioning around medoids based clustering is performed to identify a tightly bound set of valid inputs. The validated inputs from both the methods form a decision cluster and the data is fused to get the decision on primary user occupancy. Two data fusion strategies are presented and their use depends on the amount of dynamism in the CR-MANET. The analysis and results show the accuracy of primary user occupancy detection even in the presence of large number of malicious users and signal measurement errors.