Intelligent bandit learning for jamming strategy generation.

Current jamming strategy generation in communication relies on prior information or has a slow learning rate. To overcome these drawbacks, an intelligent strategy generation algorithm is proposed. This paper first creates a jamming strategy space which has three dimensions: jamming power, modulation, and duty cycle. Then, a communication jamming reward function is constructed and its property of continuity is proved. The proof of continuity provides the theoretical basis for optimization. Then, the local boundedness of the reward function in the strategy space is demonstrated. Finally, an intelligent algorithm for jamming strategy generation in communication is introduced based on the multiarmed bandit framework. Compared with the current algorithm, the new one can significantly improve the optimization speed by enlarging its exploring area from one point to a certain region. Simulations show that, by using the regional exploration, this algorithm can explore much more jamming strategies compared with the existing methods in the equal training cycles, and learn the optimal jamming strategy faster. [ABSTRACT FROM AUTHOR]