Anti in-band full-duplex interception for wireless tactical networks

Abstract

Tactical Vehicular Ad Hoc Networks (TVANs) are a key enabler of modern battlefield communications, supporting coordination among highly mobile entities such as ground combat vehicles, command posts, and relay platforms. However, the broadcast nature of wireless transmissions makes TVANs inherently vulnerable to interception and eavesdropping, especially under advanced threats such as in-band full-duplex (IBFD) interceptors that can simultaneously jam and intercept legitimate links. In practice, anti-interception protection must preserve low probability of interception (LPI) while still guaranteeing the quality-of-service (QoS) requirements of mission-critical traffic under stringent mobility, latency, and energy constraints. This thesis develops a cross-layer anti-interception framework for DS-CDMA-based TVANs by jointly integrating three complementary protection mechanisms: (i) passive protection via communication resource adaptation (e.g., power or spreading control), (ii) active cooperative jamming to disrupt interception attempts, and (iii) AES-based physical-layer scrambling with adaptive key length to balance confidentiality, delay, and energy consumption. First, a double-layer strategy combining passive resource adaptation and cooperative jamming is formulated as a joint resource allocation problem involving transmit power, jamming power, and spreading-related parameters under QoS and LPI constraints. Since the resulting problem is non-convex and interference-coupled, an efficient optimization-based benchmark is derived using tractable approximations, including first-order Taylor linearization, difference-of-convex (DC) decomposition, and sequential/iterative decomposition to obtain near-optimal solutions. Second, the framework is extended to a triple-layer design by incorporating adaptive encryption into the resource allocation. To quantify the joint impact of physical-layer defenses and encryption, an encryption-aware secrecy performance metric is employed, revealing the trade-offs among interception resistance, QoS, encryption delay, and energy cost. While the optimization benchmark provides high-quality solutions, its computational burden increases with network size and mobility, which limits real-time deployment. To enable near real-time operation, this thesis further proposes a learning-assisted solution based on multi-agent deep reinforcement learning (MADRL) under a centralized training and decentralized execution (CTDE) paradigm. The MADRL formulation targets the most computationally intensive control dimensions and learns resource allocation policies that approximate optimization-quality performance with low online runtime. Simulations under dynamic tactical conditions demonstrate that the proposed optimization and MADRL solutions improve LPI protection and maintain QoS more effectively than representative single-layer baselines, while the MADRL approach achieves fast decision making suitable for high-mobility TVAN operation.

Date	4 May 2026
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Kim Khoa Nguyen (Supervisor)