Contents
Contributors

Preface

1. Artificial Intelligence Based Approach for Robust Radar Detection in Maritime Scenarios

1.1. Introduction
1.2. Case Study
    1.2.1. Target Model
    1.2.2. Interference Model
1.3. Reference CGLR Detectors
1.4. Intelligent Detectors
    1.4.1. Intelligent Agent Architectures
    1.4.2. Error Function of the Training Process
1.5. Experiments and Results
    1.5.1. Detectors Design
    1.5.2. Real Scenario
1.6. Conclusions
Acknowledgements
References

2. Robust Hierarchical Bayesian Modeling  for Automatic Model Order Determination
 

2.1. Introduction
    2.1.1. Non-Negative Data Representation
    2.1.2. Sparseness in Data Representation
    2.1.3. Non-Negative Matrix Factorization and Model Order Determination
    2.1.3.1. Non-Negative Matrix Factorization
    2.1.3.2. Model Order Determination
2.2. Hierarchical Bayesian Modeling
2.3. Data Model
    2.3.1. Gaussian Distribution
    2.3.2. Poisson Distribution
    2.3.3. Gamma Distribution
2.4. Parameter Model 47
    2.4.1. Half Normal Distribution
    2.4.2. Exponential Distribution
    2.4.3. Tweedie Distribution
2.5. Hyper-parameter Model
    2.5.1. Informative Prior
    2.5.1.1. Gamma Distribution
    2.5.1.2. Tweedie Distribution
    2.5.1.3. Variational Bayesian Distribution
    2.5.2. Non-Informative Prior
    2.5.2.1. Stirling’s Algorithm
    2.5.2.2. Expectation Maximization (EM) Algorithm
2.6. Result from Different Methods
    2.6.1. Fence Dataset
    2.6.2. Swimmer Dataset
    2.6.3. Face Dataset
    2.6.4. Music Dataset
2.7. Conclusion
References

3. Complex Wavelet Additive Synthesis of Audio Signals: Introduction and Applications

3.1. Introduction
3.2. Audio, Time-Frequency Distributions and Wavelets
    3.2.1. Basic Definitions
    3.2.2. The Complex Continuous Wavelet Transform
    3.2.2.1. General Considerations
    3.2.2.2. Extracting Information from the Complex Continuous Wavelet Transform
3.3. Gaussian Complex Filtering
    3.3.1. The Morlet Wavelet Revised
    3.3.2. Mathematical Considerations
    3.3.2.1. The Simplest Case: A Pure Cosine with Constant Amplitude
    3.3.2.2. Pure Cosine with Variable Amplitude
    3.3.2.3. The Sum of n Pure Cosines
    3.3.2.4. The Quadratic Phase FM Signal
    3.3.3. A General Monocomponent Signal: The Linear Chirp Approximation
    3.3.4. The Additive Synthesis Method
3.4. CWAS: Discretization
    3.4.1. Constant Resolution Filter Bank
    3.4.2. Variable Resolution Filter Bank
3.5. The Complex Wavelet Additive Synthesis algorithm
    3.5.1. Calculation of the Wavelet Coefficients
    3.5.2. Wavelet Spectrogram and Scalogram
    3.5.3. Partials and Additive Synthesis Reconstruction
    3.5.4. Partials and Tracking of Partials
3.6. Resynthesis and Extraction of Time-Frequency Characteristics of the Audio Signal
    3.6.1. Audio Model: Time-Frequency Characteristics
    3.6.2. Plotting the Characteristics of the Signal
    3.6.2.1. 2D Representation
    3.6.2.2. 3D Representation
3.7. CWAS Applications
    3.7.1. F0 Estimation
    3.7.1.1. General Considerations
    3.7.1.2. NHRF0
    3.7.1.3. Polyphony Inference: Accumulated Energy
    3.7.1.4. HRF0
    3.7.1.5. Discrete Cepstrum
    3.7.1.6. Estimation Results
    3.7.2. Blind Audio Source Separation
    3.7.2.1. General Considerations
    3.7.2.2. BASS using CWAS
3.8. Conclusions and Future Work
References

4. A Review of the Recent Cepstrum-based Methods for Acoustic Feedback Cancellation

4.1. Introduction
4.2. Cesptral Analysis of the SR System
4.3. Cesptral Analysis of the AFC System
    4.3.1. Cesptral Analysis of the Microphone Signal
    4.3.2. Cesptral Analysis of the Error Signal
4.4. AFC Methods Based on Cepstral Analysis
    4.4.1. AFC-CM
    4.4.2. AFC-CE
4.5. Simulation Configuration
    4.5.1. Simulated Environment
    4.5.1.1. Feedback Path
    4.5.1.2. Forward Path
    4.5.1.3. Delay Filter
    4.5.2. Evaluation Metrics
    4.5.2.1. Maximum Stable Gain
    4.5.2.2. Misalignment
    4.5.2.3. Frequency-weighted Log-spectral Signal Distortion
    4.5.2.4. W-PESQ
    4.5.3. Speech Database
4.6. Simulation Results
    4.6.1. Results for Long Acoustic Feedback Path
    4.6.2. Results for Short Acoustic Feedback Path
4.7. Conclusions
References

5. Covariance Analysis of Periodically Correlated Random Processes for Unknown Non-stationarity Period

5.1. Introduction
5.2. Period Estimation as a Problem of Searching for Hidden Periodicities
5.3. PCRP-model of Hidden Periodicities
5.4. Coherent Covariance Analysis
    5.4.1. The Estimation of the Mean Function Period
    5.4.2. Estimation of Mean Function and Its Fourier Coefficients
    5.4.3. Estimation of Covariance and Correlation Function Period
    5.4.4. Estimation of Covariance Function and Its Fourier Coefficients
    5.4.5. The Verification of the Developed Approach
    5.4.5.1. The Analysis of Simulated Data
    5.4.5.2. The Covariance Analysis of Vibration of the Fast Moving Spindle
5.5. Component Covariance Analysis
    5.5.1. Period Estimation of Deterministic Component
    5.5.1.1. Cosine and Sine Transforms of Realization
    5.5.1.2. Component Period Estimation
    5.5.2. Covariance Functionals
    5.5.2.1. Cosine and Sine Covariance Functionals
    5.5.2.2. Component Covariance Functional
    5.5.3. Correlation Functionals
    5.5.3.1. Cosine and Sine Correlation Functionals
    5.5.3.2. Component Correlation Functional
    5.5.4. Moment Functions Estimators
    5.5.4.1. Mean Function Estimator
    5.5.4.2. The Covariance Function Estimator
    5.5.5. Period Estimation for Simulated and Real Data
    5.5.5.1. Simulated Data Analysis
    5.5.5.2. Application to Vibration Signal Analysis
5.6. Comparison with Other Results
5.7. Conclusions
References
 

6. Learning Colours from Textures by Effective Representation of Images

6.1. Introduction
6.2. Early Learning-based Colourisation Methods
6.3. Learning Colours from Texture on Non-linear Manifolds
    6.3.1. Colourisation Problem
    6.3.2. Manifold of Patch Populations
    6.3.3. Sparsity Constraints
    6.3.4. Semantic Groups
    6.3.5. Prediction Algorithm
    6.3.6. Implementation Details
    6.3.7. Experiment Results
6.4. Transferring Colours by Deep Neural Networks
    6.4.1. Fully Convolutional and Diluting Neural Networks
    6.4.2. Colouration by FCN
6.5. End Notes
References

7. The Physical Layer of Low Power Wide Area Networks: Strategies, Information Theory’s Limit and Existing Solutions

7.1. Introduction
7.2. Constraints of the Physical Layer
    7.2.1. Power Amplifier and Constant Envelope Modulations
    7.2.2. Achieving Long Range
    7.2.3. Sensitivity of the Receiver
7.3. Strategies
    7.3.1. Reduce the Bandwidth
    7.3.2. Reduce the Required Signal-to-Noise Ratio
    7.3.2.1. Reduce the Spectral Efficiency
    7.3.2.2. Reduce the Spectral Efficiency Along with the Required E_b/N_0
    7.3.2.3. Reduce the Level of Required E_b/N_0
    7.3.3. Comparison to the Information Theory’s Limit
7.4. Existing Solutions
    7.4.1. Proprietary Solutions
    7.4.1.1. Sigfox
    7.4.1.2. LoRa
    7.4.2. Standardized Technologies
    7.4.2.1. 802.15.4k
    7.4.2.2. Narrow-Band IoT
    7.4.3. Turbo-FSK
    7.4.3.1. Transmitter
    7.4.3.2. Receiver
    7.4.4. Performance
7.5. Conclusion
References

8. Deep Belief Networks and Its Applications  in Evaluation of Vehicle Interior Noise

8.1. Introduction
8.2. Related Works
    8.2.1. Psychoacoustics-based Interior Sound Quality Prediction
    8.2.2. Machine-learning-based Interior Sound Quality Prediction
8.3. Deep Belief Networks
    8.3.1. Restricted Boltzmann Machine Architecture
    8.3.2. Deep Belief Network Architecture
    8.3.3. Development of Deep Belief Networks to Model Continuous Data
8.4. Recording and Evaluating the Vehicle Interior Noise
    8.4.1. Development of the Vehicle Interior Noise Database
    8.4.2. Subjective Evaluation of the Vehicle Interior Noise
8.5. Interior Noise Feature Extraction
    8.5.1. Psychoacoustic-based Feature Extraction
    8.5.2. Energy-based Feature Extraction
    8.5.3. Feature Comparison and Selection
8.6. Development of the CRBM-DBN-based Sound Quality  Prediction Model
    8.6.1. Performance Measurements
    8.6.2. Designing the Architecture of CRBM-DBN
    8.6.3. Model Verification and Comparison
8.7. Conclusion
Acknowledgments
References

9. Selection of Diagnostic Symptoms  for Technical Condition Assessment  and Prognosis

9.1. Introduction
9.2. Singular Value Decomposition (SVD) Method
    9.2.1. Theoretical Outline
    9.2.2. Examples
9.3. Information Content Measure (ICM) Method
    9.3.1. Background
    9.3.2. The Question of Stationarity
    9.3.3. The Question of Abrupt Changes
    9.3.4. Representativeness Factor
9.4. Example
9.5. Conclusions
Acknowledgements
References

10. Nonlinear Acoustic Echo Cancellation  Using a Dynamic Micro-speaker Model

10.1. Introduction
10.2. Loudspeaker Nonlinearities
10.3. Quasi-linear Model of the Micro-speaker
10.4. Model Identification of a System Characterised by Nonlinear Damping
10.5. Experimental Identification of a Micro-speaker
10.6. Conclusions
Acknowledgements
References

11. A New Damage Detection Method for Bridge Condition Assessment in Structural Health Monitoring (SHM)

11.1. Introduction
    11.1.1. An Idea of the System State Representation Methodology (SRM)
    11.1.2. Advances of SRM Compare with the Traditional Methods and Research Scopes
    11.1.3. Data Management
    11.1.4. Nonlinear Simulation and Feature Extracting
    11.1.5. Condition Assessment
11.2. SRM Concept and Algorithms
    11.2.1. Definition of the State of a System
    11.2.2. System State Defining
    11.2.3. System State Approximating
    11.2.4. Optimal State Representation Based on a Kernel Function
11.3. SRM Assessment Method
    11.3.1. Probability-based Assessment Methods
    11.3.1.1. Statistic Test Methods
    11.3.1.2. System State Indices
    11.3.1.3. Hypothesis Probability with Two Divided-levels
    11.3.1.4. Approximate Probability with Two Divided-levels
    11.3.1.5. Approximate Probability with Multi-Levels
    11.3.1.6. Assessment Based on State Probability
    11.3.1.7. Assessment Based on Information Entropy
    11.3.2. A Simple Simulation Experimental Result
    11.3.3. Summary
11.4. SRM Tool 1: Compatible Gradient Algorithm for Large Scale Linear Problem
    11.4.1. A Simple Gradient Algorithm for Linear Problem
    11.4.2. Compatible Gradient Algorithm
    11.4.3. A Simple Example
    11.4.4. Summary
    11.4.5. Monotonic Descent Conjugates Gradient (MDCG) Method
11.5. SRM Tool 2: Frequency Slice Wavelet Transform Theory  and Methods
    11.5.1. Introduction
    11.5.2. Frequency Slice Wavelet Transform Analysis
    11.5.2.1. Frequency Slice Wavelet Transform
    11.5.2.2. Bandwidth-to-Frequency Ratio Property
    11.5.2.3. Scale Selection
    11.5.2.4. Reconstructive Procedure Independency Property
    11.5.2.5. FSF Design Property
    11.5.2.6. Filter Property
    11.5.2.7. Discrete FSF
    11.5.2.8. Discrete FSWT
    11.5.3. FSWT Signal Process Flow
    11.5.4. Comparing with Classical Transforms
    11.5.5. Applications
    11.5.5.1. Signal Filter in Time and Frequency Domain
    11.5.5.2. Separation of Close Modals
    11.5.5.3. Resolution and Symmetry
    11.5.5.4. Energy Leakage Problem
    11.5.5.5. Scale Optimization
    11.5.5.6. Dynamic Scale
    11.5.5.7. A Real Life Signal
    11.5.6. Summary

11.6. SRM Feature Extracting Methods and Applications
    11.6.1. Basic Idea of System Feature
    11.6.2. Approximate Feature Vectors Problem
    11.6.3. Extracting Damping Feature
    11.6.4. Feature Extract Algorithm
    11.6.5. Overall Logics of SRM Method
    11.6.6. Application to a Laboratory Bridge Monitoring System and Discussions
    11.6.6.1. Outline of SRM-based Damage Detection
    11.6.6.2. Overview of Bridge Model Girder Experiment
    11.6.6.3. Experimental Results and Discussions
11.7. Concluding Remarks
References

12. On Line Diagnosis in Induction Motors and Load

12.1. Introduction
12.2. Internal Fault Detection in Induction Motors
    12.2.1. Motor Current Signature Analysis
    12.2.2. Extended Park’s Vector Approach
    12.2.3. Towards an Integrated Fault Diagnostic System
    12.2.4. Experimental Results
    12.2.4.1. The Experimental Prototype
    12.2.4.2. Case Study 1
    12.2.4.3. Case Study 2
    12.2.4.4. Case Study 3
    12.2.4.5. General Comments
12.3. Mechanical Faults Detection beyond the Motor
    12.3.1. General Comments
    12.3.2. Misaligned Drives Model
    12.3.3. Fault Detection Method
    12.3.4. Experimental Results
12.4. Gear Box Fault Detection
12.5. Conclusions
Acknowledgements
References

13. Vocal Folds Dynamics by Means of Optical Flow Techniques: A Review of the Methods

13.1. Introduction
    13.1.1. Voice Assessment and Endoscopy
    13.1.2. Scientific and Clinical Applications of Laryngeal High-Speed Videoendoscopy
    13.1.3. From a Big Amount of Data to a Synthesized View
13.2. Optical-Flow Computation
    13.2.1. Theoretical Background
    13.2.2. Optical Flow Evaluation
    13.2.3. Optical Flow in LHSV
13.3. Application of OF Computation to LHSV Image Analysis
    13.3.1. Database
    13.3.2. Image Processing Implementation
    13.3.3. Optical Flow Facilitative Playbacks
    13.3.3.1. Optical Flow Kymogram (OFKG)
    13.3.3.2. Optical Flow Glottovibrogram (OFGVG)
    13.3.3.3. Glottal Optical Flow Waveform (GOFW)
    13.3.3.4. Vocal Folds Displacements Trajectories
    13.3.4. Reliability Assessment of Optical Flow Playbacks
13.4. Assessment of OF-Based Playbacks
    13.4.1. Displacements Trajectories Comparison
    13.4.2. Comparison between OFGVG, GVG and |dxGVG|
    13.4.3. Glottal Velocity: Derivative of Glottal Area Waveform and Glottal Optical FlowWaveform
    13.4.4. Optical Flow Using Local Dynamics along One Line: Digital Kymogram and Optical Flow Kymogram
13.5. Conclusions
Acknowledgements
References

Index