Collaboration among cognitive nodes is a feature of many cognitive radio applications. Collaboration can take place in the context of sensing, in the context of any self-organising process, in the context of sharing resources, for bargaining purposes, during normal operation (especially in distributed scenarios in which nodes forward information for each other) and in many other instances. Collaboration is an enabling feature of a cognitive network. And again, while other more traditional networks involve elements of collaboration, the cognitive network seems to go further. The problem with relying on or using collaboration extensively is that a malicious or misbehaving node in a collaborative network can cause damage. It may even be possible for a few well-placed malicious nodes to cause a disproportionately large amount of damage. In fact this topic would have been equally relevant in the previous section on observations, especially as collaborative sensing is key in many cognitive radio applications: for example it can be crucial in dealing with hidden node problems.
A misbehaving or malicious radio can purposely not report the fact it has sensed an incumbent, leading to spectrum being detected as unoccupied when in fact it is occupied. It can purposely insert all sorts of other false information into the system – for example, it can report false neighbor numbers and give the impression that the network is more sparse that it actually is. It can give false accounts of traffic flows and congestion, and can encourage the use of unsuitable routes.
On the one hand we can solve this issue with security measures. Strong authentication of all users and careful admission control policies might prevent such an event happening. Content integrity measures, such as those discussed earlier in the chapter, can also be applied. And some kind of reputation system may prove useful. A reputation system is a type of collaborative filtering algorithm which attempts to determine ratings for a collection of entities, given a collection of opinions that those entities hold about each other.
The reputation system leads back to a more important point. Security attacks and malicious behavior aside, a cognitive network will typically often be heterogeneous in nature. Different nodes will have different abilities to sense. And, despite every good intention, a node may insert false or poor data into the system. To deal with this, there needs to be careful design of all collaborative protocols. For example, thresholds can be set that more or less say ‘unless X neighbors say they saw Y , then Y did not happen’ and these thresholds can be dynamic. This threshold example is very simple but there are more sophisticated filtering and information fusion options and ways of weighting observations. The idea of weighting observations also suggests the use of a reputation system. Hence the design of robust collaboration mechanisms for cognitive networks should be good enough to force the attacker to have to compromise a large number of nodes in order to make any impact.
