Internet of Things Cognitive Radio Technologies

Transcription

1 Internet of Things Cognitive Radio Technologies Torino, 29 aprile 2010 Roberto GARELLO, Politecnico di Torino, Italy

2 Speaker: Roberto GARELLO, Ph.D. Associate Professor in Communication Engineering Dipartimento di Elettronica Politecnico di Torino, Italy sito web: 2

3 European Commission Report, May 2009 [1] 3

4 4

5 5

6 Contents Introduction to cognitive radio Spectrum sensing Sensing network structure: fusion centre/fully distributed Resource allocation/cross layer design Standard 6

7 Introduction to Cognitive Radio new systems increasing bandwidth need static frequency allocation Spectrum congestion idea = opportunistic usage of some frequency bands 7

8 Cognitive Radio paradigm opportunistic usage Primary users (licensed) can transmit whenever they need Secondary users can transmit only when the primary users are silent 8

9 Primary users are transmitting Secondary users cannot transmit Primary user Primary user Secondary users Primary users are not transmitting Secondary users can transmit Primary user Primary user Secondary users 9

10 COGNITIVE RADIO Must: Sense the frequency bands and understand the channel status [SPECTRUM SENSING] Dynamically adapt its radio parameters for o maximize its throughput o produce zero (very limited) interference to primary users [RESOURCE ALLOCATION] 10

11 SPECTRUM SENSING In its simplest form: must understand if a channel is free or busy More features: multi-dimensional sensing: determine frequency, temporal, geographical occupancy modulation detection: infer modulation type & parameters, channel access, protocol 11

12 SPECTRUM SENSING: typical performance measures False alarm probability The primary users are not transmitting Primary user Primary user Secondary users The secondary user spectrum sensing makes a mistake and reveals its presence The secondary users could transmit but they don t They are wasting bandwidth 12

13 SPECTRUM SENSING: typical performance measures Missed detection probability The primary users are transmitting Primary user Primary user Secondary users The secondary user spectrum sensing makes a mistake and does not reveal its presence The secondary users should not transmit but they do They are producing interference 13

14 Multi-dimensional spectrum sensing Frequency/time: The spectrum bandwidth is divided in smaller channels Channel occupation is not continuous 14

15 Multi-dimensional spectrum sensing Geographical: the spectrum can be available to a subset of the cognitive networks (example: users at larger distance) 15

16 SPECTRUM SENSING: requirements Good performance o Low False alarm probability: maximize secondary users throughput o Low Missed detection probability: minimize primary users interference Very reactive (must take a decision in a limited amount of time) Efficient secondary user network structure (central fusion centre, all/some/few cognitive users, completely distributed) Good integration with upper layers (sensing/allocation schedule for a single node, sensing distribution between nodes) 16

17 SPECTRUM SENSING: algorithms Non-parametric detection: (energy detection/eigenvalue-based detection) no assumption on primary signals Search for known patterns Cyclo-stationarity sensing Matched filter completely known primary signals 17

18 Energy detection H 0 ONLY NOISE y(n) = v(n) H 1 SIGNAL + NOISE y(n) = s(n)+v(n) N samples for slot y = (y(1) y(n) y(n)) Test statistics T = E 2 σ E N n= 1 2 = y( n) Estimated energy 2 σ Estimated noise variance 18

19 Energy detection: algorithm decision Set a threshold λ if T E σ λ E = < λ σ = if T 2 2 DECISION = signal is present DECISION = signal is absent 19

20 Energy detection: performance FALSE-ALARM PROBABILITY: signal is absent but detector erroneously detects it H 0 ONLY NOISE y(n) = v(n) T λ λ P = fa P( T λ H0) = Q N 1 2 σ Q function decreases with its argument P fa decreases if threshold λ increases 20

21 Energy detection: performance MISSED-DETECTION: signal is present but detector erroneously does not detect it H 1 SIGNAL + NOISE y(n) = s(n)+v(n) T < λ λ 1 SNR) 2 σ Pmd = P( T < λ H1) = 1 Q N 1+ 2SNR Q function decreases with its argument P md decreases if threshold λ decreases 21

22 Energy detection: ROC performance ROC (Receiver Operating Characteristic): P md vs. P fa Taken from [3] 22

23 Eigenvalue-based detection With respect to energy detection, this class of detectors is characterized by: Better performance Higher complexity Samples are collected by K independent receivers (or antennas), forming a matrix Y y11... y1 N = yk1... y kn 23

24 Eigenvalue-based detection Given the K x K sample covariance matrix R = YY h Its K eigenvalues are computed: λ... 1 λk The most popular test is the GLRT (Generalized Likelihood Ratio Test): T = λ K i= 1 1 λ i which strongly improves the Energy Detection performance 24

25 Non-parametric detection algorithms: performance comparison 25

26 Spectrum sensing network structure Device-centric (a single device senses and takes the decision) Cooperative sensing (K device sense and merge their information to take the decision) o Fusion center o Fully distributed 26

27 Fusion center All the sensing devices transmit some information (local decision or more information) to a central privileged device The fusion center merge the information, takes the information and communicate it to the network 27

28 Fully distributed sensing Each sensing user sends some information to its neighbors Sensing information is propagated through the network After some iterations the algorithm can be used to -compute multiple, location-dependent, sensing estimation relative to each single node - make a global decision about the presence of primary users in the overall area 28

29 Example: factor graph modelling Fully distributed sensing by factor graph 29

30 Belief propagation (on loopy networks) Network message passing 30

31 Fully distributed sensing by factor graph After few iteration, each node is helped by its neighbors to converge to the true sensing estimation 31

32 Interesting feature: Fully distributed sensing by factor graph Some nodes can obtain the estimation without sensing, by using the neighbors information 32

33 Resource allocation: cross layer design Towards a unified framework for designing: Spectrum sensing parameters MAC (medium access control) with the aim of optimizing the secondary network average throughput 33

34 Resource allocation: cross layer design Opportunistic spectrum usage requires to take decisions on some key events. Some examples: - When a channel is declared free and secondary users begin to transmit over it, is the spectrum sensing performed in parallel to data transmission? - How often must the secondary users negotiate their transmission parameters? - Can multiple channels be used together? - What happens when Secondary Users data transfer is interrupted by Primary Users? It is being buffered for further transmission It tries to switch to a free channel - Is it better to divide the secondary network in clusters? How to do this? 34

35 Enabling technology Software radio Represents an optimal solution : re-configurable, re-programmable, adaptive, reactive, multi-standard, etc. Node computational capacity and power consumption Key issue: some operations can be rather complex Nodes synchronization, and coordination Low data-rate control channel? 35

36 International standards Cognitive radio WRAN WiFi k Bluetooth 36

37 Cognitive Radio Networks (802.22) First international cognitive radio standard : WRAN (Wireless Regional Area Networks) USA TV bands (54/862 MHz) Stringent requirements (P fa < 0.1, P md < 0.1) Positioning information based on some GPS-equipped base stations can be used to improve geographical information 37

38 WLAN (802.11k) WLAN devices usually connect to the Access Point with the strongest signal level k: The Access Point senses each channel and collects sense information from users Non utilization of each band is estimated This information is used for channel allocation If an Access Point is already loaded at his maximum capacity (also taking into account interference), new users are assigned to underutilized AP Overall system throughput increases 38

39 Bluetooth Adaptive Frequency Hopping ISM band: frequency hopping to reduce interference with other devices Adaptive Frequency Hopping: Channel quality is sensed Bad channels are avoided 39

40 Example of application: WiFi vs. WSN [5] The concept of adaptive resource allocation can already be applied to existing systems Spectrum sensing + optimal frequency allocation Example: WSN, Bluetooth P TX =20 dbm P TX =0 dbm The WiFi and WSN channels partially overlap in the 2.4 GHz ISM Asymetric situation: WiFi interferers over WSN 40

41 Some references [1] Maarten Botterman European Commission, Internet of Things: an early reality of the Future Internet, Workshop Report, May 2009 [2] Ekram Hossain, Vijay Bhargava, Cognitive Wireless Communication Networks, Springer [3] Tevfik Yucek and Huseyin Arslan, A Survey of Spectrum Sensing Algorithms for Cognitive Radio Applications, IEEE Communications surveys and tutorials, vol. 11, no. 1, first quarter [4] Jihoon Park, Przemysław Pawełczak, and Danijela Cabric, Performance of Joint Spectrum Sensing and MAC Algorithms for Multichannel Opportunistic Spectrum Access Ad Hoc Networks, submitted to IEEE Transactions, [5] Federico Penna, Roberto Garello, Maurizio A. Spirito, Distributed Inference of Channel Occupation Probabilities in Cognitive Networks via Message Passing, Dyspan 2010, Singapore, April [6] Istituto Superiore Mario Boella and Politecnico di Torino, Impact of Wi-Fi Traffic on the IEEE Channels Occupation in Indoor Environments, ICEAA