| Popis: |
While the existing anti-jamming solutions tend to “escape” the attacks by finding another communication channel or adapting, waiting until the attacks cease, this work proposes an unprecedented method to combat jammers by leveraging the jamming signals to transmit data based on the recent advances in ambient backscatter communication technology. When the jammer attacks the channels, the transmitter modulates the jamming signals to backscatter information to the receiver. To deal with the uncertainty of jamming attacks and environment conditions, we first develop a Markov decision process framework with the Q-learning algorithm to obtain the optimal policy for the system. However, the Q-learning algorithm is widely known for its slow convergence, especially for systems with a large number of states and/or actions. For that, we develop a novel deep reinforcement learning algorithm based on the dueling neural network architecture that converges to the optimal policy much faster than the conventional Q-learning. Extensive simulations show that our proposed solution can improve the average throughput up to 426% and reduce the packet loss by 24% compared to other anti-jamming solutions. |
