The number of wirelessly communicating devices increases every day, along with the number of communication standards and technologies that they use to exchange data. A relatively new form of research is trying to find a way to make all these co-located devices not only capable of detecting each other's presence, but to go one step further - to make them cooperate. One recently proposed way to tackle this problem is to engage into cooperation by activating 'network services' (such as internet sharing, interference avoidance, etc.) that offer benefits for other co-located networks. This approach reduces the problem to the following research topic: how to determine which network services would be beneficial for all the cooperating networks. In this paper we analyze and propose a conceptual solution for this problem using the reinforcement learning technique known as the Least Square Policy Iteration (LSPI). The proposes solution uses a self-learning entity that negotiates between different independent and co-located networks. First, the reasoning entity uses self-learning techniques to determine which service configuration should be used to optimize the network performance of each single network. Afterwards, this performance is used as a reference point and LSPI is used to deduce if cooperating with other co-located networks can lead to even further performance improvements.
展开▼
