Cognitive Green Communications: From Concept to Practice

Transcription

1 Cognitive Green Communications: From Concept to Practice Honggang ZHANG International Chair - CominLabs Université Européenne de Bretagne (UEB) & Supélec/IETR Supélec SCEE Seminar March 21, 2013 Rennes, France 1

2 Outline Part I the Concept: Energy-efficient Cognitive Green Radio Communications Part II the Practice: Cognitive Green Communications for Achieving Energy Saving within Cellular Mobile Networks ACKNOWLEDGEMENT: This presentation is supported by the International Chair Program, CominLabs Excellence Center, Université Européenne de Bretagne (UEB) and SUPELEC/IETR. (GREAT: Green Cognitive Radio for Energy-Aware wireless communication Technologies evolution) Also, thanks to Prof. Jacques Palicot (SUPELEC), Dr. Tao Chen (VTT), Dr. Xianfu Chen (VTT), Mr. Rongpeng Li (ZJU), and Mr. Xuan Zhou (ZJU) for their supporting materials. 2

3 UEB - Université Européenne de Bretagne 3

4 Global Warming The Most Dangerous Threat? 4

5 Terrible Climate Change: Trans-Arctic Shipping Routes Navigable 21st-midcentury Source: Laurence C. Smith and Scott R. Stephenson, New Trans-Arctic Shipping Routes Navigable by Midcentury, PNAS, January

6 Data Explosion - Exponential Traffic Growth 6

7 Data Explosion - Exponential Traffic Growth (2) Source: 7

8 Part I: Green Communications Paradigm Change from Coverage- & Capacity- Driven to Energy-Efficiency Driven Era 8

9 Energy Crisis and Challenges Source: Prof. T. Aoyama, Keio University, ISCIT 2010 Keynote Speech. 9

10 Energy Crisis and Challenges (2) ICT Sector Commitments to Targets and Deadlines for CO 2 and Greenhouse Gas Emissions and Energy Efficiency/Consumption (European Commission 2009/03/12) 10

11 Energy Crisis and Challenges (3) ICT Sector Commitments to Targets and Deadlines for CO 2 and Greenhouse Gas Emissions and Energy Efficiency/Consumption (European Commission 2009/03/12) 11

12 Architecture of Telecommunication Networks 12

13 Mobile Telecommunications Networks Power Consumption Breakdown Energy consumption composition in Vodafone (Source: Vodafone) 13

14 Energy Consumption in Radio Access Networks 14

15 Energy Consumption Reference Model for Base Station Source: Tao Chen, et al., Network Energy Saving Technologies for Green Wireless Access Networks IEEE Wireless Communications Magazine,

16 Energy Consumption Reference Model for Base Station (2) Note: Values in italic are power consumption figures in GSM system. 16

17 Network-wide Energy Saving Strategies & Techniques Increasing bandwidth can also save energy, depending on context Source: Tao Chen, et al., Network Energy Saving Technologies for Green Wireless Access Networks IEEE Wireless Communications Magazine,

18 Cognitive Green Communications Intelligence with Adaptation, Balancing & Optimization for Network Energy Saving 18

19 Features & Key Functionalities of Cognitive Radio (Cognitive Cycle) Source: Gurkan Gur and Fatih Alagoz, Green Wireless Communications via Cognitive Dimension: An Overview, IEEE Network, March

20 Embedded Intelligence in a General Cognitive Radio Transceiver Cognitive Radio Node Application Layer Network Layer MAC Layer PHY Layer Source: Xianfu Chen, Zhifeng Zhao, and Honggang Zhang, Stochastic Power Adaptation with Multi-agent Reinforcement Learning for Cognitive Wireless Mesh Networks, IEEE Transactions on Mobile Computing, Q Xianfu Chen, Zhifeng Zhao, Honggang Zhang, and Tao Chen, Conjectural variations in multi-agent reinforcement learning for energy-efficient cognitive wireless mesh networks, in Proceedings of IEEE WCNC 2012, Paris, France, Apr

21 Why Reinforcement Learning? Machine Learning 21

22 Basics of Reinforcement Learning Policy Reward Value Model of environment Policy: What to do Reward: What is good Value: What is good because it predicts reward Model: What follows what 22

23 Processes of Reinforcement Learning 23

24 Workflow of Energy Saving Mechanism Enabled by Cognitive Process/Cycle Source: Oliver Blume, et al. Energy Savings in Mobile Networks Based on Adaptation to Traffic Statistics, Bell Labs Technical Journal 15(2), (2010). 24

25 Once upon a Time What was Cognitive Radio, Really? Joe Mitola s Cognitive Radio (1999) Simon Haykin s Cognitive Radio (2005) Cognitive Radio (G. Gur and F. Alagoz, 2011) DySPAN s Cognitive Radio (2007) 25

26 Once upon a Time What was Cognitive Radio, Really? (2) 26

27 Part II: The Practice Energy Saving for Greener Cellular Mobile Networks Tidal Effect of Cellular Networks Traffic Flow & Loads 27

28 Representative Patterns of Traffic Loads during One Day (Cellular Networks) 28

29 Representative Patterns of Traffic Loads during 3 Weeks (Cellular Networks) Normalized load of three different cell sectors over 3 weeks. The moving average of each cell over one second has been plotted. The cells show high load (Top), varying load (Middle), and low load (Bottom). Source: Daniel Willkomm et al., Primary User Behavior in Cellular Networks and Implications for Dynamic Spectrum Access. 29

30 Representative Patterns of Traffic Load during 5 Days (Core Networks/Internet) E-commerce website: 292 production web servers over 5 days. (Traffic varies by day/weekend, power doesn t.) 30

31 Base Stations Traffic Loads Measurement Campaigns in Zhejiang (China) Traffic records from 9 MSCs and SGSNs with about 7000 BSs with coverage of 780 km 2 Both GSM and UMTSBSs traffic from January to December in 2012, serving about 3 million subscribers Source: Xuan Zhou, Zhifeng Zhao, Rongpeng Li, Yifan Zhou, and Honggang Zhang, The Predictability of Cellular Networks Traffic, IEEE ISCIT2012, October

32 Measured Traffic Loads Variation Patterns (One Week) 32

33 Typical Examples of Measured Base Stations Traffic Loads in Zhejiang (China) Source: Rongpeng Li, Zhifeng Zhao, Yan Wei, Xuan Zhou, and Honggang Zhang, GM-PAB: a grid-based energy saving scheme with predicted traffic load guidance for cellular networks, in Proceedings of IEEE ICC 2012, Ottawa, Canada, Jun

34 Sensing and Prediction of Cellular Networks Traffic Flows & Loads BS1 link 1 BS3 route 2 link 3 y1,t x2,t x3,t route 3 Router route 1 X Missing x x x 1,1 2,1 3,1 link 2 x x x 1,2 2,2 3,2 BS2 Anomaly x x x 1,3 2,3 3,3 x x x 1,4 1,4 1,4 x x x 1,5 2,5 3,5 Future x x x 1,6 2,6 3,6 Interpolation: fill in the missing data from incomplete and/or indirect measurements of the Traffic Matrices 34

35 Sensing and Prediction of Cellular Networks Traffic Flows &Loads (2) Source: Rongpeng Li, Zhifeng Zhao, Xuan Zhou, and Honggang Zhang, Energy savings scheme in radio access networks via compressive sensing-based traffic load prediction, European Transactions on Emerging Telecommunications Technologies (ETT), Nov

36 Network Energy Saving through BS Switching on/off (Sleep Mode) 36

37 Block Diagram of Reinforcement Learning - The learning system and the environment are both inside the feedback loop 37

38 Reinforcement Learning: Actor-Critic Approach 38

39 Stochastic BS Switching Operation with Actor-Critic Learning Source Rongpeng Li, Zhifeng Zhao, Xian Chen, and Honggang Zhang, Energy Saving through a Learning Framework in Greener Cellular Radio Access Networks, in Proceedings of IEEE Globecom 2012, Anaheim, USA, Dec

40 Stochastic BS Switching Operation with Actor-Critic Learning (2) 40

41 Base Stations Traffic Load State Vector 41

42 Traffic Loads and BS Power Consumption Model 42

43 Actor-Critic Learning: Markov Decision Process 43

44 Actor-Critic Learning: Markov Decision Process (2) 44

45 Actor-Critic Learning: Markov Decision Process (3) 45

46 Actor-Critic Learning Scheme for BS Power Saving 46

47 Numerical Analysis Parameter description Value Simulation area 1.5km * 1.5km Maximum transmission power Macro BS 20W Micro BS 1W Maximum operational power Macro BS 865W Micro BS 38W Height Macro BS 32m Micro BS 12.5m Intra-cell interference factor 0.01 Channel bandwidth 1.25MHz File requests Arrival rate File size 100kbyte Constant power percentage 47

48 Energy Saving by Actor-Critic Learning (BS Switching & Sleep Mode) Performance comparison between Actor-Critic learning framework (LF) based energy saving scheme and the state-of-the-art (SOTA) scheme (JSAC, Sept. 2012) under various static/variant traffic arrival rates. Source Rongpeng Li, Zhifeng Zhao, Xian Chen, and Honggang Zhang, Energy Saving through a Learning Framework in Greener Cellular Radio Access Networks, in Proceedings of IEEE Globecom 2012, Anaheim, USA, Dec

49 Basics and Advantages of Transfer Learning 49

50 Stochastic BS Switching Operation with Transfer Reinforcement Learning Xuan Zhou, Zhifeng Zhao, R. Li, Y. Zhou, J. Palicot, and Honggang Zhang, TACT: A Transfer Actor-Critic Learning Framework for Energy Saving in Cellular Radio Access Networks, arxiv: , November

51 Basics and Features of Transfer Reinforcement Learning 51

52 Features of Transfer Actor-Critic Learning 52

53 TACT : The Transfer Learning Framework for Energy Saving Scheme 53

54 Numerical Analysis 54

55 Energy Saving by Transfer Actor-Critic Learning (BS Switching & Sleep Mode) Performance comparison among classical AC scheme, TACT scheme and SOTA scheme under various homogeneous traffic arrival rates when the transfer rate θ = 0.1 Performance impact of the transfer rate factor θ to the TACT scheme when λ = Xuan Zhou, Zhifeng Zhao, R. Li, Y. Zhou, J. Palicot, and Honggang Zhang, TACT: A Transfer Actor-Critic Learning Framework for Energy Saving in Cellular Radio Access Networks, arxiv: , November

56 Summary & Conclusion Environmental-friendly Green Communications: A paradigm change from traditional coverage- & capacity-driven to energy-efficiency driven communications and networks (Smart, sustainable, and self-harmonized greener ICT). Cognitive Green Radio Communications: Besides spectrum and energy, intelligence is the THIRD kind of resource, but without limitation of scarcity. Learning and decision making algorithms under green constraint can play a significant role in enabling energy- and spectralefficient greener future communications. Effective energy saving can be realized by using various learning approaches in mobile cellular networks. Cognitive Green Communications: From Concept to Reality! 56

57 Thanks! 57