Abstract
In the evolving landscape of smart cities, mobile cooperative sensing nodes generate vast amounts of data requiring offloading to edge servers. However, the open nature of wireless transmission exposes these networks to intelligent attackers capable of eavesdropping, jamming, and spoofing. This paper proposes a secure communication strategy that integrates game theory and Q-learning to optimize power allocation, maximizing secrecy data rates while minimizing system latency.
Introduction: The Security Challenge in Smart Cities
Smart cities rely on a dense network of mobile nodes for environmental monitoring, traffic management, and public safety. Due to limited local processing power and battery life, these nodes frequently offload computational tasks to edge servers. This offloading process is inherently vulnerable to "intelligent" radio devices that can disrupt transmission or steal sensitive information.
The primary challenge is balancing two often-conflicting goals: maintaining high security (secrecy data rate) and ensuring low latency for time-sensitive computational tasks. Traditional security methods often overlook the performance impact on latency, necessitating a unified approach that considers both factors in an adversarial setting.
System Model and Adversarial Environment
The researchers model a scenario involving three key entities: a transmitter (Alice), a legitimate receiver (Bob), and an intelligent attacker (Eve). Alice must decide how much power to allocate for transmitting task data to Bob while Eve attempts to degrade the communication.
Eve is considered "intelligent" because she can switch between four distinct attack modes to maximize Alice's system latency:
- Silent: No active interference.
- Eavesdropping: Passive listening to capture data.
- Jamming: Active interference to reduce the signal-to-noise ratio at Bob.
- Spoofing: Sending deceptive signals to confuse the receiver.
Methodology: Combining Game Theory and Q-Learning
To address this adversarial environment, the paper employs a two-pronged mathematical approach combining game theory and reinforcement learning.
Non-Cooperative Game Theory
The interaction between Alice and Eve is modeled as a non-cooperative game. Alice aims to minimize a "benefit function" that incorporates both task latency and power consumption. Conversely, Eve aims to maximize this function to disrupt Alice's operations. The authors derive the Nash Equilibrium (NE) for this game, identifying the stable state where neither party can improve their outcome by changing strategies independently.
Q-Learning for Dynamic Power Allocation
Since Alice may not have prior knowledge of Eve's specific attack strategy or the environmental channel conditions, a reinforcement learning approach is used. Alice implements a Q-learning algorithm to iteratively learn the optimal transmit power. By interacting with the environment and receiving feedback through repeated trials, the system converges toward the Nash Equilibrium, allowing Alice to adapt to the attacker's behavior dynamically.
Results and Performance Analysis
The proposed strategy was evaluated through MATLAB simulations using standard synthetic parameters for channel gain and noise. Key findings include:
- Convergence: The Q-learning algorithm successfully stabilized Alice's power allocation within 1,000 to 2,500 trials.
- Secrecy Rate: The system maintained a stable secrecy data rate between 1.72 and 1.75 bps/Hz, depending on the offloading ratio.
- Latency Optimization: For a task offloading ratio of 0.1, the total system latency (including local computation, transmission, and edge processing) stabilized at approximately 1.1 seconds.
- Resilience: The effectiveness of Eve's interference significantly decreased as Alice's learning progressed, demonstrating the algorithm's ability to "out-learn" and mitigate active threats.
Conclusion and Future Directions
This research provides a robust framework for securing edge computational networks in smart cities. By treating security and latency as a joint optimization problem solved through game theory and reinforcement learning, the system can effectively counter intelligent adversaries.
Future work will explore more complex network topologies involving multiple users and attackers. Additionally, the authors suggest that incorporating Deep Reinforcement Learning (DRL) and Unmanned Aerial Vehicles (UAVs) as mobile relays could further enhance the scalability and performance of secure offloading in dynamic urban environments.
Summary updated