| Underwater sensor networks(UWSNs)are vulnerable to jamming attacks due to serious multipath effects and long transmission delay.Jammers send faked or replayed signals to degree the underwater communication quality and even interrupt the ongoing sensing data transmission.Underwater communication can apply relay nodes to improve the anti-jamming transmission quality.However,existing anti-jamming underwater relay schemes rely on prior knowledge such as underwater channel and jamming models,which appreciably affects the bit error rate(BER)and the energy consumption of the relay node.Therefore,this paper studies the anti-jamming underwater cooperative communication,which applies the reinforcement learning(RL)and game theory to improve the underwater transmission quality and extend UWSN life cycle against jamming attacks.First,a prospective theory(PT)based anti-jamming underwater rely game model is proposed to reveal the impact of the subjectivity of the jammers on the underwater transmission quality.This paper analyzes the equilibrium strategies of the PT based underwater communication game under different underwater channel environments and jamming strategies,and reveals the impact of the factors such as the subjectivity of the jammer,jamming energy consumption and channel gains on the anti-jamming underwater transmission.Compared with the existing expected utility theory based studies,this paper reveals the impact of the factors such as the subjectivity of the jammer,jamming model and energy consumption on the anti-jamming underwater transmission performance.Secondly,a RL based anti-jamming underwater cooperative communication mechanism named FQR is investigated to optimize the relay power allocation and the trajectory strategies,and improve the underwater cooperative communication performance consisting of the BER of the underwater sensing data and the relay energy consumption.Relay observes the current underwater communication status and dynamically adjusts the relay transmission power and trajectory strategy without knowing the underwater channel,network and the jamming models.In addition,a deep reinforcement learning based anti-jamming underwater cooperative communication mechanism named FDQR is proposed for the underwater relay nodes that can support the computational complexity of deep learning,in which a convolutional neural network is used to compress the underwater communication state space and accelerate the learning speed.The non-anechoic pool experimental results show that the FQR reduces 14.5%of the BER of the underwater sensing data,saves 26.4%of the relay energy consumption and increases the relay utility by 27.1%compared with the benchmark relay scheme QLR.In addition,our proposed FDQR further reduces 31.5%of the BER of the underwater sensing data,saves 48.4%of the relay energy consumption and increases the relay utility by 44.1%compared with the QLR. |
