Applications of Machine Learning in Wireless Communications (Telecommunications)

Cover
Contents
Foreword
1 Introduction of machine learning
	1.1 Supervised learning
		1.1.1 k-Nearest neighbours method
		1.1.2 Decision tree
			1.1.2.1 Classification and regression tree
			1.1.2.2 Random forest
			1.1.2.3 Gradient boosting decision tree
		1.1.3 Perceptron
			1.1.3.1 Support vector machine
			1.1.3.2 Logistic regression
			1.1.3.3 Multilayer perceptron and deep learning
		1.1.4 Summary of supervised learning
	1.2 Unsupervised learning
		1.2.1 k-Means
		1.2.2 Density-based spatial clustering of applications with noise
		1.2.3 Clustering by fast search and find of density peaks
		1.2.4 Relative core merge clustering algorithm
		1.2.5 Gaussian mixture models and EM algorithm
			1.2.5.1 The EM algorithm
			1.2.5.2 The EM algorithm for GMM
		1.2.6 Principal component analysis
		1.2.7 Autoencoder
		1.2.8 Summary of unsupervised learning
	1.3 Reinforcement learning
		1.3.1 Markov decision process
		1.3.2 Model-based methods
		1.3.3 Model-free methods
			1.3.3.1 Monte Carlo methods
			1.3.3.2 Temporal-difference learning
		1.3.4 Deep reinforcement learning
			1.3.4.1 Value function approximation
			1.3.4.2 Policy gradient methods
		1.3.5 Summary of reinforcement learning
	1.4 Summary
	Acknowledgement
	References
2 Machine-learning-enabled channel modeling
	2.1 Introduction
	2.2 Propagation scenarios classification
		2.2.1 Design of input vector
		2.2.2 Training and adjustment
	2.3 Machine-learning-based MPC clustering
		2.3.1 KPowerMeans-based clustering
			2.3.1.1 Clustering
			2.3.1.2 Validation
			2.3.1.3 Cluster pruning—ShapePrune
			2.3.1.4 Development
		2.3.2 Sparsity-based clustering
		2.3.3 Kernel-power-density-based clustering
		2.3.4 Time-cluster-spatial-lobe ( TCSL)-based clustering
			2.3.4.1 TC clustering
			2.3.4.2 SL clustering
		2.3.5 Target-recognition-based clustering
		2.3.6 Improved subtraction for cluster-centroid initialization
		2.3.7 MR-DMS clustering
			2.3.7.1 Cluster the MPCs
			2.3.7.2 Obtaining the optimum cluster number
	2.4 Automatic MPC tracking algorithms
		2.4.1 MCD-based tracking
		2.4.2 Two-way matching tracking
		2.4.3 Kalman filter-based tracking
		2.4.4 Extended Kalman filter-based parameters estimation and tracking
		2.4.5 Probability-based tracking
	2.5 Deep learning-based channel modeling approach
		2.5.1 BP-based neural network for amplitude modeling
		2.5.2 Development of neural-network-based channel modeling
		2.5.3 RBF-based neural network for wireless channel modeling
		2.5.4 Algorithm improvement based on physical interpretation
	2.6 Conclusion
	References
3 Channel prediction based on machine-learning algorithms
	3.1 Introduction
	3.2 Channel measurements
	3.3 Learning-based reconstruction algorithms
		3.3.1 Batch algorithms
			3.3.1.1 Support vector machine
			3.3.1.2 Neural networks
			3.3.1.3 Matrix completion
		3.3.2 Online algorithms
			3.3.2.1 APSM-based algorithm
			3.3.2.2 Multi-kernel algorithm
	3.4 Optimized sampling
		3.4.1 Active learning
			3.4.1.1 Query by committee
			3.4.1.2 Side information
		3.4.2 Channel prediction results with path-loss measurements
	3.5 Conclusion
	References
4 Machine-learning-based channel estimation
	4.1 Channel model
		4.1.1 Channel input and output
	4.2 Channel estimation in point-to-point systems
		4.2.1 Estimation of frequency-selective channels
	4.3 Deep-learning-based channel estimation
		4.3.1 History of deep learning
		4.3.2 Deep-learning-based channel estimator for orthogonal frequency division multiplexing ( OFDM) systems
		4.3.3 Deep learning for massive MIMO CSI feedback
	4.4 EM-based channel estimator
		4.4.1 Basic principles of EM algorithm
		4.4.2 An example of channel estimation with EM algorithm
	4.5 Conclusion and open problems
	References
5 Signal identification in cognitive radios using machine learning
	5.1 Signal identification in cognitive radios
	5.2 Modulation classification via machine learning
		5.2.1 Modulation classification in multipath fading channels via expectation– maximization
			5.2.1.1 Problem statement
			5.2.1.2 Modulation classification via EM
			EM-based modulation classifier
			5.2.1.3 Numerical results
		5.2.2 Continuous phase modulation classification in fading channels via Baum– Welch algorithm
			5.2.2.1 Problem statement
			5.2.2.2 Classification of CPM via BW
			5.2.2.3 Numerical results
	5.3 Specific emitter identification via machine learning
		5.3.1 System model
			5.3.1.1 Single-hop scenario
			5.3.1.2 Relaying scenario
		5.3.2 Feature extraction
			5.3.2.1 Hilbert– Huang transform
			5.3.2.2 Entropy and first-and second-order moments-based algorithm
			5.3.2.3 Correlation-based algorithm
			5.3.2.4 Fisher's discriminant ratio-based algorithm
		5.3.3 Identification procedure via SVM
		5.3.4 Numerical results
		5.3.5 Conclusions
	References
6 Compressive sensing for wireless sensor networks
	6.1 Sparse signal representation
		6.1.1 Signal representation
		6.1.2 Representation error
	6.2 CS and signal recovery
		6.2.1 CS model
		6.2.2 Conditions for the equivalent sensing matrix
			6.2.2.1 Null space property
			6.2.2.2 Restricted isometry property
			6.2.2.3 Mutual coherence
		6.2.3 Numerical algorithms for sparse recovery
			6.2.3.1 Convex optimization algorithms
			6.2.3.2 Greedy pursuit algorithms
	6.3 Optimized sensing matrix design for CS
		6.3.1 Elad's method
		6.3.2 Duarte-Carvajalino and Sapiro's method
		6.3.3 Xu et al.' s method
		6.3.4 Chen et al.' s method
	6.4 CS-based WSNs
		6.4.1 Robust data transmission
		6.4.2 Compressive data gathering
			6.4.2.1 WSNs with single hop communications
			6.4.2.2 WSNs with multi-hop communications
		6.4.3 Sparse events detection
		6.4.4 Reduced-dimension multiple access
		6.4.5 Localization
	6.5 Summary
	References
7 Reinforcement learning-based channel sharing in wireless vehicular networks
	7.1 Introduction
		7.1.1 Motivation
		7.1.2 Chapter organization
	7.2 Connected vehicles architecture
		7.2.1 Electronic control units
		7.2.2 Automotive sensors
		7.2.3 Intra-vehicle communications
		7.2.4 Vehicular ad hoc networks
		7.2.5 Network domains
		7.2.6 Types of communication
	7.3 Dedicated short range communication
		7.3.1 IEEE 802.11p
		7.3.2 WAVE Short Message Protocol
		7.3.3 Control channel behaviour
		7.3.4 Message types
	7.4 The IEEE 802.11p medium access control
		7.4.1 Distributed coordination function
		7.4.2 Basic access mechanism
		7.4.3 Binary exponential backoff
		7.4.4 RTS/ CTS handshake
		7.4.5 DCF for broadcasting
		7.4.6 Enhanced distributed channel access
	7.5 Network traffic congestion in wireless vehicular networks
		7.5.1 Transmission power control
		7.5.2 Transmission rate control
		7.5.3 Adaptive backoff algorithms
	7.6 Reinforcement learning-based channel access control
		7.6.1 Review of learning channel access control protocols
		7.6.2 Markov decision processes
		7.6.3 Q-learning
	7.7 Q-learning MAC protocol
		7.7.1 The action selection dilemma
		7.7.2 Convergence requirements
		7.7.3 A priori approximate controller
		7.7.4 Online controller augmentation
		7.7.5 Implementation details
	7.8 VANET simulation modelling
		7.8.1 Network simulator
		7.8.2 Mobility simulator
		7.8.3 Implementation
	7.9 Protocol performance
		7.9.1 Simulation setup
		7.9.2 Effect of increased network density
		7.9.3 Effect of data rate
		7.9.4 Effect of multi-hop
	7.10 Conclusion
	References
8 Machine-learning-based perceptual video coding in wireless multimedia communications
	8.1 Background
	8.2 Literature review on perceptual video coding
		8.2.1 Perceptual models
			8.2.1.1 Manual identification
			8.2.1.2 Automatic identification
		8.2.2 Incorporation in video coding
			8.2.2.1 Model-based approaches
			8.2.2.2 Learning-based approaches
	8.3 Minimizing perceptual distortion with the RTE method
		8.3.1 Rate control implementation on HEVC-MSP
		8.3.2 Optimization formulation on perceptual distortion
		8.3.3 RTE method for solving the optimization formulation
		8.3.4 Bit reallocation for maintaining optimization
	8.4 Computational complexity analysis
		8.4.1 Theoretical analysis
		8.4.2 Numerical analysis
	8.5 Experimental results on single image coding
		8.5.1 Test and parameter settings
		8.5.2 Assessment on rate–distortion performance
		8.5.3 Assessment of BD-rate savings
		8.5.4 Assessment of control accuracy
		8.5.5 Generalization test
	8.6 Experimental results on video coding
		8.6.1 Experiment
			8.6.1.1 Settings
			8.6.1.2 Evaluation on R–D performance
			8.6.1.3 Evaluation on RC accuracy
	8.7 Conclusion
	References
9 Machine-learning-based saliency detection and its video decoding application in wireless multimedia communications
	9.1 Introduction
	9.2 Related work on video-saliency detection
		9.2.1 Heuristic video-saliency detection
		9.2.2 Data-driven video-saliency detection
	9.3 Database and analysis
		9.3.1 Database of eye tracking on raw videos
		9.3.2 Analysis on our eye-tracking database
		9.3.3 Observations from our eye-tracking database
	9.4 HEVC features for saliency detection
		9.4.1 Basic HEVC features
		9.4.2 Temporal difference features in HEVC domain
		9.4.3 Spatial difference features in HEVC domain
	9.5 Machine-learning-based video-saliency detection
		9.5.1 Training algorithm
		9.5.2 Saliency detection
	9.6 Experimental results
		9.6.1 Setting on encoding and training
		9.6.2 Analysis on parameter selection
		9.6.3 Evaluation on our database
		9.6.4 Evaluation on other databases
		9.6.5 Evaluation on other work conditions
		9.6.6 Effectiveness of single features and learning algorithm
	9.7 Conclusion
	References
10 Deep learning for indoor localization based on bimodal CSI data
	10.1 Introduction
	10.2 Deep learning for indoor localization
		10.2.1 Autoencoder neural network
		10.2.2 Convolutional neural network
		10.2.3 Long short-term memory
	10.3 Preliminaries and hypotheses
		10.3.1 Channel state information preliminaries
		10.3.2 Distribution of amplitude and phase
		10.3.3 Hypotheses
			10.3.3.1 Hypothesis 1
			10.3.3.2 Hypothesis 2
			10.3.3.3 Hypothesis 3
	10.4 The BiLoc system
		10.4.1 BiLoc system architecture
		10.4.2 Off-line training for bimodal fingerprint database
		10.4.3 Online data fusion for position estimation
	10.5 Experimental study
		10.5.1 Test configuration
		10.5.2 Accuracy of location estimation
		10.5.3 2.4 versus 5 GHz
		10.5.4 Impact of parameter
	10.6 Future directions and challenges
		10.6.1 New deep-learning methods for indoor localization
		10.6.2 Sensor fusion for indoor localization using deep learning
		10.6.3 Secure indoor localization using deep learning
	10.7 Conclusions
	Acknowledgments
	References
11 Reinforcement-learning-based wireless resource allocation
	11.1 Basics of stochastic approximation
		11.1.1 Iterative algorithm
		11.1.2 Stochastic fixed-point problem
	11.2 Markov decision process: basic theory and applications
		11.2.1 Basic components of MDP
		11.2.2 Finite-horizon MDP
			11.2.2.1 Case study: multi-carrier power allocation via finite-horizon MDP
		11.2.3 Infinite-horizon MDP with discounted cost
			11.2.3.1 Case study: multi-carrier power allocation with random packet arrival
		11.2.4 Infinite-horizon MDP with average cost
			11.2.4.1 Case study: multi-carrier power allocation with average cost
	11.3 Reinforcement learning
		11.3.1 Online solution via stochastic approximation
			11.3.1.1 Case study: multi-carrier power allocation without channel statistics
		11.3.2 Q-learning
			11.3.2.1 Case study: multi-carrier power allocation via Q-learning
	11.4 Summary and discussion
	References
12 learning-based power control in small-cell networks
	12.1 Introduction
	12.2 System model
		12.2.1 System description
		12.2.2 Effective capacity
		12.2.3 Problem formulation
	12.3 Noncooperative game theoretic solution
	12.4 Q-learning algorithm
	12.4 Q-learning algorithm
		12.4.1 Stackelberg game framework
		12.4.2 Q-learning
		12.4.3 Q-learning procedure
			12.4.3.1 Sparsely deployed scenario
			12.4.3.2 Densely deployed scenario
			12.4.3.3 Distributed Q-learning algorithm
		12.4.4 The proposed BDb-WFQA based on NPCG
	12.5 Simulation and analysis
		12.5.1 Simulation for Q-learning based on Stackelberg game
		12.5.2 Simulation for BDb-WFQA algorithm
	12.6 Conclusion
	References
13 Data-driven vehicular mobility modeling and prediction
	13.1 Introduction
	13.2 Related work
	13.3 Model
		13.3.1 Data sets and preprocessing
		13.3.2 Model motivation
		13.3.3 Queue modeling
	13.4 Performance derivation
		13.4.1 Vehicular distribution
		13.4.2 Average sojourn time
		13.4.3 Average mobility length
	13.5 Model validation
		13.5.1 Time selection and area partition
			13.5.1.1 Area partition
			13.5.1.2 Observation period selection
		13.5.2 Arrival rate validation
		13.5.3 Vehicular distribution
		13.5.4 Average sojourn time and mobility length
	13.6 Applications of networking
		13.6.1 RSU capacity decision
		13.6.2 V2I and V2V combined performance analysis
	13.7 Conclusions
	References
Index
Back Cover