Control issues in cognitive networks. Marko Höyhtyä and Tao Chen CWC-VTT-Gigaseminar 4th December 2008

Transcription

1 Control issues in cognitive networks Marko Höyhtyä and Tao Chen CWC-VTT-Gigaseminar 4th December 2008

2 Outline Cognitive wireless networks Cognitive mesh Topology control Frequency selection Power control 2

3 Cognition cycle Research topics in cognitive radio Spectrum sensing Dynamic spectrum access Coexistence Technical challenges Spectrum sensing Reliability, Sensitivity, and Response time. Coexistence of heterogeneous systems, especially primary users User Driven and secondary users. Multi-dimension resource allocation Signaling to support CR Outside World Cognitive radio (CR) Infer from Context Pre-process Parse Stimuli Observe Orient Establish Priority Immediate New States Infer from Radio Model Normal Urgent Learn Act Normal Generate Alternate Goals Plan Decide (Buttons) Autonomous States Determine Best Allocate Resources Initiate Processes Negotiate Negotiate Protocols 3 Plan Generate Determine Best Best Waveform Known Waveform Adapted from J. Mitola, Cognitive Radio for Flexible Mobile Multimedia Communications, Mobile Networks and Applications, vol. 5, No. 4, pp , 2001 [5]

4 Cognitive wireless networks (CWN) Cognitive radio: learn from the environment and adapt certain radio operating parameters to incoming RF stimuli. (by Simon Haykin [6]) Cognitive wireless networks: learn from network-wide environment and adapt network configuration to incoming RF and network stimuli. Similarity of CR and CWN Use cognitive process, which is goal driven and relies on observations and learning to reach decision. Use software tunable platform. Difference of CR and CWN Scope of controlling goals. Degree of heterogeneity. Degree of freedom. Example of CWN architecture. Proposed by Thomas et al. from Virginia Tech. 4

5 Future Wireless Networks Ubiquitous Communication Among People and Devices Wireless Internet access Nth generation Cellular Wireless Ad Hoc Networks Sensor Networks Wireless Entertainment Smart Homes/Spaces Automated Highways All this and more Future wireless networks will be CWNs! 5

6 A cognitive wireless mesh networks (CogMesh) CR User Primary User m ban nd Sp pectru Licensed Band I CR Ad-Hoc Network without Infrastructure Unlicensed Band Primary User CR Network with Infrastructure CR User Primary User Multi-channel CR User Licensed Band II Coexistence with CR 6

7 Topology control in cognitive mesh network 7

8 Topology control in CogMesh Scenario Secondary users (SU) coexist with primary users (PU). SUs form a CR ad hoc network. Distributed control Self-organization Self-healing SU uses spectrum holes {123} for communications, no common control available. 2 {123} {2,3} Solution 1 Cluster based network formation. {123} Goal: reduce cluster numbers in network Minimal dominating set (MDS) algorithm to control the connection topology and adapt to radio environment changes. {2} 13 {2} Channel list Secondary user Primary user on channel 1 1 8

9 Cluster formation at initial cluster construction (ICC) phase {2,3} 1 2 {2} 13 {2} Cluster head Cluster member Channel list Cluster head Ordinary node 1 Primary user on channel 1 9

10 MDS algorithm to reduce cluster number Reduce cluster number {2,3} {1,2} {123} {1,2} {2,3} {1,2} {1,2} {1} {2,3} {1} {2,3} {2,3} {2,3} {1,2} {1,2} {1,2} {1,2} {1,2} 10 {1,2} 10

11 Simulation Result Before Number of clusters before and after proposed algorithm when spectrum holes change Number of clusters after different of algorithms After 11

12 Control cloud concept Assumption: no common channel available. A control cloud is form by a group of connected nodes who share a common control channel. The objective is to make control clouds as large as possible in order to reduce control overhead. Control channel clouds may grow or shrink according to the available common channels. 12

13 Use swarm intelligence for control cloud formation A population of simple agents interacting locally with one another and with their environment to perform complex tasks. Use the principle of division of labor Parallel optimization method Examples: ant colonies, bird flocking, animal herding, bacterial growth, and fish schooling SI in communications Routing AntNet AntHocNet Spectrum hole detection Particle Swarm Optimization 13

14 Swarm intelligence algorithm for control cloud A node chooses its control channel according to quality of available channels and choices of neighbors. Each node broadcasts HELLO messages to its neighbors on control channel. The channel lists and statistic states are included in HELLO messages. The receiving of HELLO act as pheromone in SI to affect the decision of the node on its control channel. The objective is to let neighbor nodes select a common channel with good quality as their common control channel. {2,3} 1 2 {2} {2} 1 3 {2,3} 1 2 {2} {2} 1 3 {2,3} 1 2 {2} {2}

15 Performance comparison 15

16 Frequency selection 16

17 Control challenge The biggest challenge in cognitive networks is designing clever algorithms that will take all needed information that are available including location of CR nodes, sensing information, traffic patterns of different users, database information of nations and regulations etc. and make decisions about where in the spectrum to operate at any given moment and how much power to use in that band. DSM module RF stimuli Spectrum sensing Power control PU system parameters Database Channel selection 17

18 Learning in frequency selection Cognitive radio should be more than only an opportunistic radio, i.e., radio taking immediate advantage of spectrum opportunities Ability to learn from experiences makes the operation more efficient compared to the case where only information available only at the design time is possible Learning and prediction helps cognitive radio to find out frequency channels offering longest idle times for secondary use 18

19 System model A CR stores sensing information to the database It classifies the traffic patterns of different channels and selects the prediction method for each channel based on classification When a CR has to switch channel, it selects an available one offering the longest idle time into use Channel history 1) Spectrum sensing 6) Data transmission 2) Traffic pattern classification Channel state flag Switch channel yes 3) Prediction method decision 4) Idle time prediction 5) Switching decision no 19

20 Intelligent channel selection Sensing of primary channels is a periodic sampling process to determine the state (ON or OFF) of the channels at every sampling instant Traffic patterns are basically divided into stochastic and deterministic ones Classification of patterns is made based on the periodicity it information Rules for prediction based on measurement studies, analysis, verification with simulations 20

21 Results With exponential traffic, intelligent t selection can reduce the amount of switches with 40 % Weibull and Pareto distributed traffic give same kind of results With deterministic traffic the gain is really high, amount of switches can be one third compared to random selection. 21

22 Power control 22

23 Power control CR uses sensing to obtain information about local spectrum use, sensitivity of sensor together with primary transmission power defines the sensing range rs Transmission power of the CR defines both the communication range rc and the interference range ri of it. Maximum power limit for secondary transmission can be estimated based on PU parameters and sensitivity of the sensor d i PU tx r s P su L F (r s d c ) + N + N F 6 db. d c PU rx r i SU rx r c SU tx 23

24 Adaptive transmission power control Adaptive inverse power control algorithms Maintaining required QoS with minimum transmission power (not exceeding the limit) to minimize interference Applicable to centralized architecture, also possible in clustered network We have developed adaptive filtered-x LMS (FxLMS) power control method that is close to optimal Truncation can be used in a system/application that is not delay-sensitive to further improve the performance xˆ [ k] h[k] n[k] x[k] h[k] ˆ[ ] 24

25 Results Secondary power limit increases with increasing primary transmission power Truncated method offers more energy efficient transmission and decreases the created interference allowing the less sensitive sensing. Transmission power limit for secondary user 25 Secondary tran nsmission power [db Bm] Transmitted SNR values for different power control methods Method Average transmitted SNR full inversion 27± 2 - db db truncated inversion 20.1 db 25.7 db Maximum transmitted SNR Primary transmission power [dbm] 25

26 Conclusions Control in cognitive networks is challenging and different from traditional networks due to dynamic environment We studied three different topics Topology control Control clouds for common control channel problem Clustering for network formation Power control Power limits Algorithms Frequency selection based on classification and prediction 26

27 Thank you! Any questions? Contact information: 27