Title Page
Abstract
Contents
Chapter 1. Introduction 16
1.1. Background 16
1.1.1. Underwater Acoustic Cellular Systems 17
1.2. Motivations, Achievements and Challenges 25
1.3. Aim of The Thesis 32
1.4. Organization of Thesis 35
Chapter 2. Link Adaptation in UAC Network 36
2.1. System Model and Problem Formulation 37
2.2. Underwater Acoustic Communications Network Channel Model 52
2.2.1. Path Loss Model 53
2.2.2. Ambient Noise 55
2.3. OFDM Architecture and Parameters 56
2.3.1. Characteristics of Link Level Simulation 57
Chapter 3. Measured Data Analysis with Principal Component Analysis 63
3.1. Dataset Overview 64
3.1.1. Taean Dataset 64
3.1.2. Incheon Dataset 65
3.2. Unsupervised ML: Principal Component Analysis 71
3.2.1. Power Method 72
3.2.2. Covariance Method 73
3.2.3. SVD Method 80
Chapter 4. ML Algorithms, RB Strategies, and Their Significances in LA of UACN 84
4.1. Machine Learning Algorithms 86
4.1.1. Supervised Learning Algorithms 87
4.2. Rule-based Strategy 106
4.2.1. Normal Distribution 106
4.2.2. Scatter Plot 114
4.2.3. 3D/4D Scatter Plot 114
4.3. Significances of ML and RB in Link Adaptation 115
Chapter 5. Multi-Dimensional Data Analysis for Modulation Classification 116
5.1. Machine Learning Approach for Modulation Classification 117
5.1.1. 1D Analysis 121
5.1.2. Multi-Dimensional Analysis 123
5.2. Rule-based Strategy for Modulation Classification 127
5.2.1. 1D Analysis 127
5.2.2. Multi-Dimensional Analysis 136
Chapter 6. Multi-Dimensional Data Analysis for Coding Rate Classification 141
6.1. Machine Learning Approach for Coding Rate Classification 141
6.2. Rule-based Strategy for Coding Rate Classification 146
6.2.1. 1D Analysis 153
6.2.2. Multi-Dimensional Analysis for Modulation and Coding Rate Classification 156
6.3. Performance Comparison Among Different Machine Learning Approaches 165
6.4. Link Adaptation Through AMC 167
Chapter 7. Conclusions 172
Appendix 174
References 182
Table 2.1. Distances Between all Sensor Nodes 40
Table 2.2. UAC Network Parameter Details 41
Table 2.3. AMC for UAC 44
Table 2.4. The System Parameters for UACN 45
Table 2.5. OFDM Symbol Parameters for Downlink 61
Table 2.6. OFDM Symbol Parameters for Uplink 61
Table 2.7. Details Frame Structure for Downlink and Uplink 62
Table 3.1. Taean Dataset Parameters 64
Table 3.2. Incheon Dataset Parameters 66
Table 3.3. Incheon Dataset, Parameter Available at Transmitter 66
Table 3.4. Incheon Dataset, Parameters Available at Receiver 68
Table 3.5. Values of 'Explained' and 'Latent' of all PCs 76
Table 3.6. Parameters Found in PCA Analysis 79
Table 4.1. Pseudocode of K-Nearest Neighbor Algorithm 92
Table 4.2. Pseudocode of Boosted Regression Tree Algorithm (AdaBoost) 96
Table 4.3. Pseudocode of Boosted Regression Tree Algorithm (LogitBoost) 98
Table 4.4. Pseudocode of Support Vector Machine Algorithm 101
Table 4.5. Pseudocode of Accuracy Algorithm for RB Strategy,... 109
Table 5.1. ML Performance of Different Combination of PCA Parameters 117
Table 5.2. Constant Values of Doppler Spread and Frequency Shift... 119
Table 5.3. More Observation with Doppler Spread and Frequency Shift 120
Table 5.4. Overall Accuracy and Execution Time of Single Parameter 122
Table 5.5. Overall Accuracy of Modulation Classification with... 124
Table 5.6. Characteristics of [-sigma, sigma] Overlapped Truncated Data... 130
Table 5.7. Threshold Finding Trials for Every Parameters [Non-Overlapped Data] 133
Table 5.8. Threshold Values of all Parameters for Modulation Classification 135
Table 5.9. Accuracy Measure with 2D Rule-based Analysis... 139
Table 5.10. ML and 2D Rule-based Analysis for Modulation Classification 140
Table 6.1. Overall Accuracy of Coding Rate Classification... 143
Table 6.2. Accuracy using Threshold values of Individual Parameters... 154
Table 6.3. Accuracy Measure of Modulation and Coding Rate Classifications... 158
Table 6.4. Thresholds for Three Parameters of Incheon Data 163
Table 6.5. Summary of Three-Parameter cases of Incheon Data 165
Table 6.6. Data Rate for Two-parameters Case, Incheon dataset 168
Table 6.7. Data Rate for Three-parameters Case, Incheon dataset 169
Appendix Table I. Parameters Found in PCA Analysis (Taean dataset) 175
Appendix Table II. Accuracy Measure of Modulation and Coding Rate Classifications... 177
Appendix Table III. Summary of Performances with Three-parameter cases with Taean Dataset. 181
Figure 1.1. SIR as a Function of the Cell Radius R for Different Values... 21
Figure 1.2. The Admissible Region of (R,N). 23
Figure 1.3. Link Adaptation with ML and RB Classifiers in UACN. 34
Figure 2.1. Real Time Experimental Set up of Incheon Sea Dataset. 39
Figure 2.2. MTRL Single Cell Layout Structure. 40
Figure 2.3. Two-dimensional UW Network Structure. 40
Figure 2.4. 1-tier Cellular Layout for UAC Network. 42
Figure 2.5. BER vs. SINR (Coded vs. Uncoded). 48
Figure 2.6. Channel Estimation [soft radio]. 48
Figure 2.7. Adaptive vs. Fixed Schemes. 49
Figure 2.8. Statistical Varying Nature of Taean Dataset. 50
Figure 2.9. ACM Procedure. 50
Figure 2.10. Real-Time AMC Test. 51
Figure 2.11. Estimated Channel Scattering from AMC Tests. 51
Figure 2.12. Frequency vs. Absorption Coefficient Curve. 54
Figure 2.13. Frequency vs. Pathloss Curve. 54
Figure 2.14. Individual Turbulence Noise, Shipping Noise,... 55
Figure 2.15. Total Ambient Noise Curves with Varying Shipping and Wind Factors. 56
Figure 2.16. UCN Frame Structure. 57
Figure 2.17. OFDM Frame Structure. 59
Figure 2.18. Pilot Structure. 59
Figure 2.19. Downlink (UBSC-UBS) OFDM Symbol Representation. 60
Figure 2.20. Uplink (UBS-UBSC) OFDM Symbol Representation. 60
Figure 3.1. The Working Flowchart of Finding Classification Accuracy. 70
Figure 3.2. The Covariance Matrix Defines the Shape of the Data. 74
Figure 3.3. Scree Plot of Principal Components. 76
Figure 3.4. Orthonormal Plot of Covariances of Principal Component (PC) 1 and 2. 77
Figure 4.1. KNN Flowchart. 90
Figure 4.2. Flowchart of Adaboost.M2 Algorithm. 97
Figure 4.3. Two-class Feature Space with Linear Support Vector Machine. 100
Figure 4.4. The confidence intervals correspond to 3-sigma rule... 107
Figure 5.1. Confusion Matrix for ML analysis for SNR and CB(-15 dB)(corr 5). 125
Figure 5.2. pdf of Gaussian Distribution of PCA Selected Single Parameter from Incheon... 129
Figure 5.3. Scatter Plot of 2D Rule-based Analysis: (a) SNR vs. Data Rate (bps),... 138
Figure 6.1. Confusion Matrix using ML algorithms for Coding Rate Classification with... 146
Figure 6.2. 3D and 4D Scatter Plot for Selected Parameters for Coding Rate 152
Figure 6.3. [-σ, σ] pdf of Gaussian Distribution of Incheon QPSK Data(Coding Rate wise)... 156
Figure 6.4. Confusion Matrix for Modulation Classification of Parameters'... 160
Figure 6.5. Scatter Plot for (a) Modulation Classification Data with [-1.2σ, 1.2σ],... 162
Figure 6.6. Scatter plot of [SNR, CB] vs. Data Rate:... 170
Figure 6.7. Interpolated Scatter plot of [CB, MED, RMS Delay Spread] vs. Data Rate. 171
Appendix Figure 1. Confusion matrix for (a) modulation classification with three-parameter... 180