Towards Cognitive Autonomous Networks
John Wiley & Sons Inc (Verlag)
978-1-119-58638-8 (ISBN)
Editors Stephen S. Mwanje is a Senior Research Engineer in the Cognitive Network Management research team at Nokia Bell Labs in Munich, Germany. He leads the team's research on the system aspects related to the development and application of Cognition in network management, with a special focus on the management of the radio access networks. Prior to Bell Labs, he worked for 8 years in network operations, planning and deployment for radio access, microwave and fiber optic systems. Christian Mannweiler is the Head of the Core Network Architecture & Security Research department at Nokia Bell Labs in Munich, Germany. His research focuses on network automation for 5G mobile communications systems, architectures for cloudified networks, and integration of cellular and private industrial networks. Prior to working at Bell Labs, he has been a Senior Researcher at the German Research Center for Artificial Intelligence (DFKI GmbH) in Kaiserslautern, Germany.
List of Contributors xix
Foreword I xxi
Foreword II xxv
Preface xxvii
1 The Need for Cognitive Autonomy in Communication Networks 1
Stephen S. Mwanje, Christian Mannweiler and Henning Sanneck
1.1 Complexity in Communication Networks 2
1.1.1 The Network as a Graph 2
1.1.2 Planes, Layers, and Cross-Functional Design 4
1.1.3 New Network Technology – 5G 6
1.1.4 Processes, Algorithms, and Automation 9
1.1.5 Network State Changes and Transitions 9
1.1.6 Multi-RAT Deployments 10
1.2 Cognition in Network Management Automation 11
1.2.1 Business, Service and Network Management Systems 11
1.2.2 The FCAPS Framework 13
1.2.3 Classes/Areas of NMA Use Cases 15
1.2.4 SON – The First Generation of NMA in Mobile Networks 17
1.2.5 Cognitive Network Management – Second Generation NMA 18
1.2.6 The Promise of Cognitive Autonomy 18
1.3 Taxonomy for Cognitive Autonomous Networks 19
1.3.1 Automation, Autonomy, Self-Organization, and Cognition 19
1.3.2 Data Analytics, Machine Learning, and AI 21
1.3.3 Network Autonomous Capabilities 22
1.3.4 Levels of Network Automation 23
1.3.5 Content Outline 25
References 27
2 Evolution of Mobile Communication Networks 29
Christian Mannweiler, Cinzia Sartori, Bernhard Wegmann, Hannu Flinck, Andreas Maeder, Jürgen Goerge and Rudolf Winkelmann
2.1 Voice and Low-Volume Data Communications 30
2.1.1 Service Evolution – From Voice to Mobile Internet 31
2.1.2 2G and 3G System Architecture 33
2.1.3 GERAN – 2G RAN 35
2.1.4 UTRAN – 3G RAN 36
2.2 Mobile Broadband Communications 38
2.2.1 Mobile Broadband Services and System Requirements 38
2.2.2 4G System Architecture 39
2.2.3 E-UTRAN – 4G RAN 40
2.3 Network Evolution – Towards Cloud-Native Networks 42
2.3.1 System-Level Technology Enablers 42
2.3.2 Challenges and Constraints Towards Cloud-Native Networks 46
2.3.3 Implementation Aspects of Cloud-Native Networks 47
2.4 Multi-Service Mobile Communications 49
2.4.1 Multi-Tenant Networks for Vertical Industries 50
2.4.2 5G System Architecture 51
2.4.3 Service-Based Architecture in the 5G Core 54
2.4.4 5G RAN 56
2.4.5 5G New Radio 59
2.4.6 5G Mobile Network Deployment Options 63
2.5 Evolution of Transport Networks 69
2.5.1 Architecture of Transport Networks 69
2.5.2 Transport Network Technologies 70
2.6 Management of Communication Networks 72
2.6.1 Basic Principles of Network Management 72
2.6.2 Network Management Architectures 76
2.6.3 The Role of Information Models in Network Management 79
2.6.4 Dimensions of Describing Interfaces 80
2.6.5 Network Information Models 82
2.6.6 Limitations of Common Information Models 85
2.7 Conclusion – Cognitive Autonomy in 5G and Beyond 87
2.7.1 Management of Individual 5G Network Features 87
2.7.2 End-to-End Operation of 5G Networks 88
2.7.3 Novel Operational Stakeholders in 5G System Operations 88
References 89
3 Self-Organization in Pre-5G Communication Networks 93
Muhammad Naseer-ul-Islam, Janne Ali-Tolppa, Stephen S. Mwanje and Guillaume Decarreau
3.1 Automating Network Operations 94
3.1.1 Traditional Network Operations 94
3.1.2 SON-Based Network Operations 95
3.1.3 SON Automation Areas and Use Cases 97
3.2 Network Deployment and Self-Configuration 98
3.2.1 Plug and Play 98
3.2.2 Automatic Neighbour Relations (ANR) 101
3.2.3 LTE Physical Cell Identity (PCI) Assignment 103
3.3 Self-Optimization 108
3.3.1 Mobility Load Balancing (MLB) 108
3.3.2 Mobility Robustness Optimization (MRO) 111
3.3.3 Energy Saving Management 115
3.3.4 Coverage and Capacity Optimization (CCO) 117
3.3.5 Random Access Channel (RACH) Optimization 120
3.3.6 Inter-Cell Interference Coordination (ICIC) 122
3.4 Self-Healing 124
3.4.1 The General Self-Healing Process 125
3.4.2 Cell Degradation Detection 125
3.4.3 Cell Degradation Diagnosis 127
3.4.4 Cell Outage Compensation 128
3.5 Support Function for SON Operation 129
3.5.1 SON Coordination 129
3.5.2 Minimization of Drive Test (MDT) 133
3.6 5G SON Support and Trends in 3GPP 136
3.6.1 Critical 5G RAN Features 136
3.6.2 SON Standardization for 5G 137
3.7 Concluding Remarks 140
References 141
4 Modelling Cognitive Decision Making 145
Stephen S. Mwanje and Henning Sanneck
4.1 Inspirations from Bio-Inspired Autonomy 146
4.1.1 Distributed, Efficient Equilibria 146
4.1.2 Distributed, Effective Management 147
4.1.3 Robustness Amidst Self-Organization 147
4.1.4 Adaptability 147
4.1.5 Natural Stochasticity 148
4.1.6 From Simplicity Emerges Complexity 148
4.2 Self-Organization as Visible Cognitive Automation 148
4.2.1 Attempts at Definition 149
4.2.2 Bio-Chemical Examples of Self-Organizing Systems 149
4.2.3 Human Social-Economic Examples of Self-Organizing Systems 151
4.2.4 Features of Self-Organization – As Evidenced by Ant Foraging 152
4.2.5 Self-Organization or Cognitive Autonomy? – The Case of Ants 154
4.3 Human Cognition 154
4.3.1 Basic Cognitive Processes 155
4.3.2 Higher, Complex Cognitive Processes 156
4.3.3 Cognitive Processes in Learning 158
4.4 Modelling Cognition: A Perception-Reasoning Pipeline 159
4.4.1 Conceptualization 160
4.4.2 Contextualization 160
4.4.3 Organization 161
4.4.4 Inference 161
4.4.5 Memory Operations 162
4.4.6 Concurrent Processing and Actioning 162
4.4.7 Attention and the Higher Processes 163
4.4.8 Comparing Models of Cognition 164
4.5 Implications for Network Management Automation 167
4.5.1 Complexity of the PRP Processes 167
4.5.2 How Cognitive Is SON? 168
4.5.3 Expectations from Cognitive Autonomous Networks 168
4.6 Conclusions 169
References 170
5 Classic Artificial Intelligence: Tools for Autonomous Reasoning 173
Stephen Mwanje, Marton Kajo, Benedek Schultz, Kimmo Hatonen and Ilaria Malanchini
5.1 Classical AI: Expectations and Limitations 174
5.1.1 Caveat: The Common-Sense Knowledge Problem 174
5.1.2 Search and Planning for Intelligent Decision Making 175
5.1.3 The Symbolic AI Framework 176
5.2 Expert Systems 177
5.2.1 System Components 177
5.2.2 Cognitive Capabilities and Application of Expert Systems 177
5.2.3 Rule-Based Handover-Events Root Cause Analysis 178
5.2.4 Limitations of Expert Systems 179
5.3 Closed-Loop Control Systems 180
5.3.1 The Controller 180
5.3.2 Cognitive Capabilities and Application of Closed-Loop Control 181
5.3.3 Example: Handover Optimization Loop 181
5.4 Case-Based Reasoning 182
5.4.1 The CBR Execution Cycle 183
5.4.2 Cognitive Capabilities and Applications of CBR Systems 184
5.4.3 CBR Example for RAN Energy Savings Management 185
5.4.4 Limitations of CBR Systems 185
5.5 Fuzzy Inference Systems 186
5.5.1 Fuzzy Sets and Membership Functions 186
5.5.2 Fuzzy Logic and Fuzzy Rules 187
5.5.3 Fuzzy Interference System Components 188
5.5.4 Cognitive Capabilities and Applications of FIS 189
5.5.5 Example Application: Selecting Handover Margins 190
5.6 Bayesian Networks 192
5.6.1 Definitions 193
5.6.2 Example Application: Diagnosis in Mobile Networks 193
5.6.3 Selecting and Training Bayesian Networks 194
5.6.4 Cognitive Capabilities and Applications of Bayesian Networks 195
5.7 Time Series Forecasting 196
5.7.1 Time Series Modelling 196
5.7.2 Auto Regressive and Moving Average Models 198
5.7.3 Cognitive Capabilities and Applications of Time Series Models 198
5.8 Conclusion 199
References 199
6 Machine Learning: Tools for End-to-End Cognition 203
Stephen Mwanje, Marton Kajoa and Benedek Schultz
6.1 Learning from Data 204
6.1.1 Definitions 205
6.1.2 Training Using Numerical Optimization 207
6.1.3 Over- and Underfitting, Regularization 209
6.1.4 Supervised Learning in Practice – Regression 211
6.1.5 Supervised Learning in Practice – Classification 212
6.1.6 Unsupervised Learning in Practice – Dimensionality Reduction 213
6.1.7 Unsupervised Learning in Practice – Clustering Using K-Means 215
6.1.8 Cognitive Capabilities and Limitations of Machine Learning 216
6.1.9 Example Application: Temporal-Spatial Load Profiling 218
6.2 Neural Networks 219
6.2.1 Neurons and Activation Functions 220
6.2.2 Neural Network Computational Model 221
6.2.3 Training Through Gradient Descent and Backpropagation 222
6.2.4 Overfitting and Regularization 224
6.2.5 Cognitive Capabilities of Neural Networks 226
6.2.6 Application Areas in Communication Networks 226
6.3 A Dip into Deep Neural Networks 227
6.3.1 Deep Learning 227
6.3.2 The Vanishing Gradients Problem 228
6.3.3 Drivers, Enablers, and Computational Constraints 229
6.3.4 Convolutional Networks for Image Recognition 231
6.3.5 Recurrent Neural Networks for Sequence Processing 235
6.3.6 Combining LSTMs with Convolutional Networks 237
6.3.7 Autoencoders for Data Compression and Cleaning 238
6.3.8 Cognitive Capabilities and Application of Deep Neural Networks 240
6.4 Reinforcement Learning 241
6.4.1 Learning Through Exploration 241
6.4.2 RL Challenges and Framework 242
6.4.3 Value Functions 243
6.4.4 Model-Based Learning Through Value and Policy Iteration 244
6.4.5 Q-Learning Through Dynamic Programming 245
6.4.6 Linear Function Approximation 246
6.4.7 Generalized Approximators and Deep Q-Learning 247
6.4.8 Policy Gradient and Actor-Critic Methods 248
6.4.9 Cognitive Capabilities and Application of Reinforcement Learning 252
6.5 Conclusions 253
References 253
7 Cognitive Autonomy for Network Configuration 255
Stephen S. Mwanje, Rashid Mijumbi and Lars Christoph Schmelz
7.1 Context Awareness for Auto-Configuration 256
7.1.1 Environment, Network, and Function Contexts 257
7.1.2 NAF Context-Aware Configuration 259
7.1.3 Objective Model 260
7.1.4 Context Model – Context Regions and Classes 263
7.1.5 Deriving the Context Model 265
7.1.6 Deriving Network and Function Configuration Policies 266
7.2 Multi-Layer Co-Channel PCI Auto-Configuration 267
7.2.1 Automating PCI Assignment in LTE and 5G Radio 268
7.2.2 PCI Assignment Objectives 269
7.2.3 Blind PCI Auto Configuration 270
7.2.4 Initial Blind Assignment 271
7.2.5 Learning Pico-Macro NRs 272
7.2.6 Predicting Macro-Macro NRs 272
7.2.7 PCI Update/Optimization and New Cells Configuration 273
7.2.8 Performance Expectations 273
7.3 Energy Saving Management in Multi-Layer RANs 274
7.3.1 The HetNet Energy Saving Management Challenge 275
7.3.2 Power Saving Groups 276
7.3.3 Cell Switch-On Switch-Off Order 277
7.3.4 PSG Load and ESM Triggering 278
7.3.5 Static Cell Activation and Deactivation Sequence 279
7.3.6 Reference-Cell-Based ESM 280
7.3.7 ESM with Multiple Reference Cells 281
7.3.8 Distributed Cell Activation and Deactivation 283
7.3.9 Improving ESM Solutions Through Cognition 285
7.4 Dynamic Baselines for Real-Time Network Control 285
7.4.1 DARN System Design 286
7.4.2 Data Pre-Processing 288
7.4.3 Prediction 288
7.4.4 Decomposition 289
7.4.5 Learning Augmentation 290
7.4.5.1 Knowledge Base 291
7.4.5.2 Alarm Generation 292
7.4.5.3 Metric Clustering 293
7.4.6 Evaluation 294
7.5 Conclusions 297
References 298
8 Cognitive Autonomy for Network-Optimization 301
Stephen S. Mwanje, Mohammad Naseer Ul-Islam and Qi Liao
8.1 Self-Optimization in Communication Networks 302
8.1.1 Characterization of Self-Optimization 302
8.1.2 Open- and Closed-Loop Self-Optimization 304
8.1.3 Reactive and Proactive Self-Optimization 305
8.1.4 Model-Based and Statistical Learning Self-Optimization 306
8.2 Q-Learning Framework for Self-Optimization 306
8.2.1 Self-Optimization as a Learning Loop 307
8.2.2 Homogeneous Multi-Agent Q-Learning 308
8.2.3 The Heterogeneous Multi-Agent Q-Learning SO Framework 309
8.2.4 Fuzzy Q-Learning 310
8.3 QL for Mobility Robustness Optimization 314
8.3.1 HO Performance and Parameters Sensitivity 314
8.3.2 Q-Learning Based MRO (QMRO) 315
8.3.3 Parameter Search Strategy 317
8.3.4 Optimization Algorithm 318
8.3.5 Evaluation 318
8.4 Fuzzy Q-Learning for Tilt Optimization 322
8.4.1 Fuzzy Q-Learning Controller (FQLC) Components 322
8.4.2 The FQLC Algorithm 324
8.4.3 Homogeneous Multi-Agent Learning Strategies 325
8.4.4 Coverage and Capacity Optimization 327
8.4.5 Self-Healing and eNB Deployment 327
8.5 Interference-Aware Flexible Resource Assignment in 5G 329
8.5.1 Muting in Wireless Networks 330
8.5.2 Notations, Definitions, and Preliminaries 331
8.5.3 System Model and Problem Formulation 332
8.5.4 Optimal Resource Allocation and Performance Limits 334
8.5.5 Successive Approximation of Fixed Point (SAFP) 335
8.5.6 Partial Resource Muting 335
8.5.7 Evaluation 337
8.6 Summary and Open Challenges 340
References 341
9 Cognitive Autonomy for Network Self-Healing 345
Janne Ali-Tolppa, Marton Kajo, Borislava Gajic, Ilaria Malanchini, Benedek Schultz and Qi Liao
9.1 Resilience and Self-Healing 346
9.1.1 Resilience by Design 347
9.1.2 Holistic Self-Healing 348
9.2 Overview on Cognitive Self-Healing 349
9.2.1 The Basic Building Blocks of Self-Healing 350
9.2.2 Profiling and Anomaly Detection 351
9.2.3 Diagnosis 353
9.2.4 Remediation Action 354
9.2.5 Advanced Self-Healing Concepts 354
9.2.6 Feature Reduction and Context Selection for Anomaly Detection 356
9.3 Anomaly Detection in Radio Access Networks 358
9.3.1 Use Cases 359
9.3.2 An Overview of the RAN Anomaly Detection Process 360
9.3.3 Profiling the Normal Behaviour 361
9.3.4 The New Normal – Adapting to Changes 362
9.3.5 Anomaly-Level Calculation 364
9.3.6 Anomaly Event Detection 365
9.4 Diagnosis and Remediation in Radio Access Networks 366
9.4.1 Symptom Collection 367
9.4.2 Diagnosis 367
9.4.3 Augmented Diagnosis 369
9.4.4 Deploying Corrective Actions 371
9.5 Knowledge Sharing in Cognitive Self-Healing 371
9.5.1 Information Sharing in Mobile Networks 371
9.5.2 Transfer Learning and Self-Healing for Mobile Networks 373
9.5.3 Applying Transfer Learning to Self-Healing 374
9.5.4 Prognostic Cross-Domain Anomaly Detection and Diagnosis 374
9.5.5 Cognitive Slice Lifecycle Management 375
9.5.6 Diagnosis Knowledge Cloud 376
9.5.7 Diagnosis Cloud Components 377
9.5.8 Diagnosis Cloud Evaluation 378
9.6 The Future of Self-Healing in Cognitive Mobile Networks 379
9.6.1 Predictive and Preventive Self-Healing 379
9.6.2 Predicting the Black Swan – Ludic Fallacy and Self-Healing 380
References 382
10 Cognitive Autonomy in Cross-Domain Network Analytics 385
Szabolcs Nováczki, Péter Szilágyi and Csaba Vulkán
10.1 System State Modelling for Cognitive Automation 386
10.1.1 Cognitive Context-Aware Assessment and Actioning 386
10.1.2 State Modelling and Abstraction 387
10.1.3 Deriving the System-State Model 389
10.1.4 Symptom Attribution and Interpretation 392
10.1.5 Remediation and Self-Monitoring of Actions 394
10.2 Real-Time User-Plane Analytics 396
10.2.1 Levels of User Behaviour and Traffic Patterns 396
10.2.2 Monitoring and Insight Collection 398
10.2.3 Sources of U-Plane Insight 400
10.2.4 Insight Analytics from Correlated Measurements 401
10.2.5 Insight Analytics from Packet Patterns 402
10.3 Real-Time Customer Experience Management 405
10.3.1 Intent Contextualization and QoE Policy Automation 406
10.3.2 QoE Descriptors and QoE Target Definition 408
10.3.3 QoE Enforcement 410
10.4 Mobile Backhaul Automation 411
10.4.1 The Opportunities of MBH Automation 412
10.4.2 Architecture of the Automated MBH Management 413
10.4.3 MBH Automation Use Cases 416
10.5 Summary 417
References 418
11 System Aspects for Cognitive Autonomous Networks 419
Stephen S. Mwanje, Janne Ali-Tolppa and Ilaria Malanchini
11.1 The SON Network Management Automation System 420
11.1.1 SON Framework for Network Management Automation 420
11.1.2 SON as Closed-Loop Control 421
11.1.3 SON Operation – The Rule-Based Multi-Agent Control 422
11.2 NMA Systems as Multi-Agent Systems 423
11.2.1 Single-Agent System (SAS) Decomposition 423
11.2.2 Single Coordinator or Multi-Agent Team Learning 424
11.2.3 Team Modelling 425
11.2.4 Concurrent Games/Concurrent Learning 425
11.3 Post-Action Verification of Automation Functions Effects 426
11.3.1 Scope Generation 427
11.3.2 Performance Assessment 428
11.3.3 Degradation Detection, Scoring and Diagnosis 429
11.3.4 Deploying Corrective Actions – The Deployment Plan 431
11.3.5 Resolving False Verification Collisions 433
11.4 Optimistic Concurrency Control Using Verification 436
11.4.1 Optimistic Concurrency Control in Distributed Systems 436
11.4.2 Optimistic Concurrency Control in SON Coordination 437
11.4.3 Extending the Coordination Transaction with Verification 437
11.5 A Framework for Cognitive Automation in Networks 440
11.5.1 Leveraging CFs in the Functional Decomposition of CAN Systems 440
11.5.2 Network Objectives and Context 442
11.5.3 Decision Applications (DApps) 443
11.5.4 Coordination and Control 444
11.5.4.1 Configuration Management Engine (CME) 444
11.5.4.2 Coordination Engine (CE) 445
11.5.5 Interfacing Among Functions 446
11.6 Synchronized Cooperative Learning in CANs 446
11.6.1 The SCL Principle 448
11.6.2 Managing Concurrency: Spatial-Temporal Scheduling (STS) 449
11.6.3 Aggregating Peer Information 451
11.6.4 SCL for MRO-MLB Conflicts 452
11.7 Inter-Function Coopetition – A Game Theoretic Opportunity 456
11.7.1 A Distributed Intelligence Challenge 457
11.7.2 Game Theory and Bayesian Games 458
11.7.3 Learning in Bayesian Games 461
11.7.4 CF Coordination as Learning Over Bayesian Games 463
11.8 Summary and Open Challenges 464
11.8.1 System Supervision 464
11.8.2 The New Paradigm 465
11.8.3 Old Problems with New Faces? 466
References 466
12 Towards Actualizing Network Autonomy 469
Stephen S. Mwanje, Jürgen Goerge, Janne Ali-Tolppa, Kimmo Hatonen, Harald Bender, Csaba Rotter, Ilaria Malanchini and Henning Sanneck
12.1 Cognitive Autonomous Networks – The Vision 470
12.1.1 Cognitive Techniques in Network Automation 471
12.1.2 Success Factors in Implementing CAN Projects 475
12.1.3 Implications on KPI Design and Event Logging 476
12.1.4 Network Function Centralization and Federation 477
12.1.5 CAN Outlook on Architecture and Technology Evolution 478
12.1.6 CAN Outlook on NM System Evolution 483
12.2 Modelling Networks: The System View 486
12.2.1 System Description of a Mobile Network 486
12.2.2 Describing Performance 488
12.2.3 Implications on Automation 489
12.2.4 Control Strategies 490
12.2.5 Two-Dimensional Continuum of Control 495
12.2.6 Levels of Policy Abstraction 497
12.2.7 Implications on Optimization 500
12.2.8 The Promise of Intent-Based Network Control 502
12.3 The Development – Operations Interface in CANs 506
12.3.1 The DevOps Paradigm 506
12.3.2 Requirements for Successful Adoption of DevOps 508
12.3.3 Benefits of DevOps for CAN 509
12.4 CAN as Data Intensive Network Operations 510
12.4.1 Network Data: A New Network Asset 510
12.4.2 From Network Management to Data Management 511
12.4.3 Managing Failure in CANs 512
References 514
Index 517
| Erscheinungsdatum | 23.10.2020 |
|---|---|
| Verlagsort | New York |
| Sprache | englisch |
| Maße | 152 x 229 mm |
| Gewicht | 680 g |
| Themenwelt | Technik ► Elektrotechnik / Energietechnik |
| Technik ► Nachrichtentechnik | |
| ISBN-10 | 1-119-58638-0 / 1119586380 |
| ISBN-13 | 978-1-119-58638-8 / 9781119586388 |
| Zustand | Neuware |
| Haben Sie eine Frage zum Produkt? |
aus dem Bereich
