Game Theoretic Study On Security Techniques In Next-Generation Mobile Networks

The non-orthogonal multiple access(NOMA)technology improves the spectrum utilization efficiency for the wireless communication of the next-generation mobile networks,and the mobile crowdsensing(MCS)system combines mobile cloud computing and the Internet of things to provide social sensing services.However,NOMA wireless networks are vulnerable to jamming attacks that aim to interrupt the transmission of mobile users,and the MCS system is faced with the failure risk due to the faked sensing attacks from selfish users.Therefore,the thesis provides the insight into the security mechanism of the next-generation mobile networks from the above techniques.Firstly,the downlink NOMA transmission system contending with a smart jammer is formulated as a Stackelberg game,in which the reinforcement learning based NOMA power allocation strategy is proposed to improve the anti-jamming transmission efficiency.A Stackelberg equilibrium of the anti-jamming NOMA transmission game is derived and the conditions assuring its existence are provided to disclose the impact of multiple antennas and radio channel states.In the dynamic games without the knowledge of jamming models,the Q-learning based NOMA power allocation scheme is proposed to derive the optimal strategy.The model based learning method and the transfer learning technique are used to accelerate the learning speed of the power allocation learning process.For example,compared with the benchmark Q-learning strategy,the proposed strategy in the NOMA based wireless network with 3 mobile users improves the user signal-to-interference-plus-noise-ratio and the sum data rates by 16.1%and 25.4%,respectively.Secondly,the interactions between the cloud server and a number of mobile users with embedded sensors are formulated as a secure MCS game,and the deep Q-network based MCS payment strategy is proposed to suppress the faked sensing attacks in the dynamic MCS games.The Stackelberg equilibria of the static game are presented,which analyzes the conditions to motivate accurate sensing.Without being aware of sensing parameters of mobile users,a cloud server can apply the Q-learning based payment strategy to maximize its utility in the dynamic games.The deep Q-network based payment system is proposed to accelerate the learning process and improve the performance against faked sensing attacks by using a deep convolutional neural network to extract the state information.Simulation results show that the proposed system decreases the probability of the faked sensing attack by 30.2%compared with the random payment system at the 200-th time slot.