Protecting tactical wireless networks against dual interception

Abstract

Today, wireless tactical networks are constantly being upgraded to become more networkcentric and autonomous. The unprecedented development of artificial intelligence (AI) paves the way for the non-human involved military applications in which soldiers can be replaced by autonomous combat vehicles that are fully equipped with combat weapons as well as a powerful communication ability for tactical cooperation on the battlefield. However, increasing communication connectivity combined with decreasing human intervention makes the security and reliability of the system become riskier. A larger number of connections requires very scalable resource management, and the system is more vulnerable to electronic interception by the enemy. Thus, a self-defensive capability is required in these tactical systems to protect them against the enemy’s interception. This thesis investigates the problem of protecting high-mobility tactical networks against dual enemy interceptions. We design an anti-interception resource optimization strategy in which multiple system control variables are jointly optimized to not only protect the system from enemy interception but also maintain the quality of service (QoS). We apply the proposed strategy in two different tactical scenarios: i) a Warfighter Information Network-Tactical (WIN-T) system with the high mobility of ground combat vehicles (GCVs), and ii) a mixed radio frequency/freespace optical (RF/FSO) relay network where both the relay node and enemy interceptors are high-mobility objects. In both scenarios, we mathematically formulate the anti-interception resource allocation problem as a non-convex optimization model. We decompose this intractable optimization problem into two sub-problems, then solve the first sub-problem using an iterative method. To handle the non-convex form of the second sub-problem, we combine first-order Taylor approximation with the difference of convex functions (D.C) method. To obtain the optimized solution in near real-time, we propose Deep Reinforcement Learning (DRL) approaches, using both single agent and multi-agent model. Numerical results show the DRL method has the potential to be applicable in high-complexity military scenarios. To make a decision of selecting a DRL solution for a given dual anti-interception scenario, we compare the defensive capacity of two frameworks SADRL (Single-Agent Deep Reinforcement Learning) and MADRL (Multi-Agent Deep Reinforcement Learning) in different levels of mobility and user scalability. Numerical results show that both frameworks can approximate the optimal solution. However, in highly mobile and massive deployment scenarios, the defensive performance of MADRL performs better than that of SADRL but has a higher overhead cost.

Date	30 Aug 2023
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Kim Khoa Nguyen (Supervisor)