Adaptive Framework for Jamming Suppression in Wireless Networks : 무선 네트워크의 적응형 간섭 억제 프레임워크 = Adaptive Framework for Jamming Suppression in Wireless Networks

다국어 초록 (Multilingual Abstract)

This thesis presents an integrated examination of adaptive beamforming methodologies for jamming suppression in wireless networks operating under contested conditions with limited prior knowledge of interference environments. Two complementary frameworks are: (1) a model-driven approach leveraging adaptive subspace-based covariance matrix reconstruction (AS-CMR) with autonomous source detection and KullbackLeibler divergence-based beamformer selection, and (2) a data-driven approach employing hybrid CNN-Transformer architecture for end-to-end learning of optimal beamforming weights. Through extensive numerical validation across diverse jamming scenarios, both approaches achieve near-optimal performance within 1-2 dB of the ideal Minimum Variance Distortionless Response (MVDR) bound while outperforming conventional techniques by 10-19 dB.

목차 (Table of Contents)

• 1. Introduction
• 1.1 6G and Future Wireless Paradigms
• 1.2 Jamming Threat in Future Networks
• 1.3 Classical Physical Layer Anti-Jamming Countermeasures
• 1.4 Deep Learning Paradigm
• 1.5 Contributions and Organization
• 2. Related Works
• 2.1 Related Works
• 2.2 Optimal Beamforming: The MVDR Beamformer
• 2.3 Robust Adaptive Beamforming Variants
• 2.4 Covariance Matrix Reconstruction Methods
• 2.5 Deep Learning based Beamfoming
• 3. Subspace Based Adaptive Beamformer
• 3.1 Motivation and contribution
• 3.2 System Model
• 3.3 Proposed AS-CMR Beamformer
• 3.3.1 Sample Covariance Matrix and Dominant Sources Estimation
• 3.3.2 Angle of Arrival Estimation
• 3.3.3 Candidate Beamformer Generation and Optimal Selection
• 3.3.4 Lower-Complexity Beamformer Computation
• 3.4 Noise Robust Alternative
• 3.4.1 Fourth-Order Cumulant Matrix Construction and Stabilization
• 3.4.2 Angle of Arrival Estimation
• 3.5 Computational Complexity Analysis
• 3.6 Numerical Results
• 3.7 Summary
• 4 Hybrid CNN-Transformer
• 4.1 Motivation and Contribution
• 4.2 System Model
• 4.3 Proposed Deep Learning Framework
• 4.3.1 Architecture Overview
• 4.3.2 Multi-Task Loss Function with Directional Alignment
• 4.4 Simulation Results
• 4.4.1 Dataset Generation and Preprocessing
• 4.4.2 Baseline Methods
• 4.4.3 Inference Pipeline
• 4.4.4 Numerical Results
• 4.5 Summary
• 5 Conclusion and Future Works
• 5.1 Conclusion
• 5.2 Future Works
• References