Samira Davarghalati - Department of Computer Engineering and Information Technology (QIAU) Qazvin, Iran
Mohammad Reza Meybodi - Department of Computer Engineering and Information Technology, Amirkabir University Tehran, Iran

one of the most challenging issues in cognitive networks is the selection of appropriate mechanism and presenting a learning model in the course of complete cognitive engine. In fact, in cognitive process, learning is to complete decision making and it helps future decision making and planning through maintaining the previous decisions under a set of specified conditions, or by revealing behavioral patterns in network environment. In this paper, we use Q learning algorithm, which is one of reinforcement learning techniques, to obtain appropriate value of each parameters of protocol stack in cognitive engine of cognitive wireless sensor networks. Using Q algorithm provides learning opportunity to solve the problem in a large operational environment where there is limited information. The proposed protocol is simulated by NS2.35 simulator. In this paper, we compare the proposed method with the existed method based on simulated annealing. Some network's end-to-end criteria such as end-to-end delay average, oscillation of delay, throughput and packet loss, and generally the useful function which is measured by these criteria are evaluated. The results of simulation show that the proposed method has better performance than the previous one and increase significantly the utility function.

wireless sensor networks; cognitive network; Q-learning; cognitive engine