Title Page
Contents
List of Abbreviation 8
Abstract 10
Chapter 1. Introduction 12
Chapter 2. GNN-Based Meta-Learning Approach for Adaptive Power Control in Dynamic D2D 14
2.1. Motivation 14
2.2. Problem Description 18
2.3. GNN-based meta-learning for environment adaptive power control 20
2.4. Simulation 24
2.5. Summary 31
Chapter 3. Reliable Modulation Classification: Laplace Approximation LSTM-Based Approach 32
3.1. Motivation 32
3.2. System Model 34
3.3. Bayesian Learning via Laplace Approximation 35
3.4. Simulation 39
3.5. Summary 45
Chapter 4. Conclusion 46
References 48
Table Ⅰ. Simulation Parameters for MPNN-based meta-training algorithm 24
Table Ⅱ. System Parameters for Meta-Tasks 25
Table Ⅲ. Simulation Parameters for modulation classification 39
Figure 1. Sum-throughput ratio in dynamic network topology 27
Figure 2. Outage Probability in dynamic network topology 27
Figure 3. Sum-throughput ratio in dynamic rate requirements 28
Figure 4. Outage Probability in dynamic rate requirements 28
Figure 5. Sum-throughput ratio in dynamic deployment area 29
Figure 6. Outage Probability in dynamic deployment area 29
Figure 7. Accuracy and confidence versus SNR on in-domain data 40
Figure 8. The accuracy and confidence versus input sequence length on in-domain data 42
Figure 9. The average confidence score on out-of-domain data PAM4 43
Figure 10. The average confidence score on out-of-domain data PAM4 44
Figure 11. The average confidence score on out-of-domain data PAM4 44