1. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP AIC group: Networking Protocols and agent methodology research for Sensor Networks Antonio G. Ruzzelli School of Informatics and Computer Science University College Dublin Dublin, Ireland ruzzelli@ucd.ie www.adaptiveinformation.ie

2. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan 1: Dual channel multiple access

3. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Background Traditional low cost radios for wireless sensors operate with one frequency channel at any given time, e.g Tr1001, CC1000, CC1010 A profusion of MAC protocols focus on energy efficiency over one frequency channels

4. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Unique frequency channel issues MACs like IEEE802.11, SMAC, TRAMA or BMAC suffer from: – High latency (e.g. due to RTS/CTS/ACK in CSMA/CA) –Low flexibility (Difficult to release slots unused in TDMA) –Inefficient usage of the wireless channel (e.g. the ETP problem in CSMA/CA)

5. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Advances in WSNs Novel transceivers can operate with two channels simultaneously with a relative small increase of energy consumption e.g. nRF2401 Supply current one channel in receive 18 mA Supply current two channels in receive23 mA nRF2401 is effectively mounted on the motes developed at the University of Cork (Ireland)

6. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan DCMA/AP: Dual channel multiple access with adaptive preamble Usage of 2 frequency channels –Data channel Cd for data –Control channel Cc for notifications Pros No table of neighbours required No handshake mechanisms like RTS/CTS Reduced idle listening at the receiver Adaptive wake-up node preamble Cons Small increase of current consumption in dual channel reception mode (18ma  23mA) Suitable for: –Nodes working at very low duty cycle –Dual channel transceiver

7. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan The minimum wakeup concept LCCA Nodes alternate long period of inactivity to tiny period of channel assessment; The Least Clear Channel Assessment LCCA is the shortest time period needed for nodes to sense any activity on the channel (~2.5msec in BMAC) LCCA time period is much shorter than the time required for a packet transmission (e.g. 35msec for 5byte transmission with Tr1001) LCCA can reduce node duty cycle to less than 1% Wakeup period : longest period of consecutive node activity when a signal is detected (Sensing time) Wakeup period Ts Sleep period LCCA Time

8. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan DCMA communication mechanism (1) Node are unsynchronized  asynchronous transmission All nodes apply LCCA periodically on the data channel Cd only A node with data to transmit apply LCCA on control channel Cc firstly.

9. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan The transmitter If the channel is clear then the transmitter starts sending the adaptive preamble Pa on Cd At the same time TX keeps on listening to Cc DCMA communication mechanism (2)

10. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan DCMA communication mechanism (3) The receiver During regular CCA, the receiver can sense channel activity on Cd then reply with a TIP packet on Cc TIP =transmission / reception in progress TIP contains (1)the receiver ID, (2) next Rx ID, (3) packet length

11. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan In case of error, the notification is transmitted on Cc The error packet contain the PackID In case of error the packet is rescheduled DCMA communication mechanism (4)

12. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan The exposed terminal problem removal TIP is sent by the receiver  only nodes around the receiver refrain from transmitting The communication mechanism removes the ETP!

13. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Adaptive Preamble mechanism In case of multiple transmission, asynchronous packets help the receiver to obtain the node ID of the later transmitter Tx2. Consequently Tx2 can be enabled by means of a RIP packet The Preamble transmission stops as soon as the RIP packet is received  adaptive! Note: RIP content = TIP content Difference: RIP is used to identify multiple Tx

14. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Opportunistic Crossover mechanism During periodical LCCA, if activity is sensed, nodes switches to Cc to get the TIP/RIP packet TIP/RIP packet contains info about the next scheduled RX node (nextRx) and ongoing packet length The nextRx is in the position to set up a NAV alarm to wake up right after the packet is transmitted. Other nodes set up a double NAV to wake up just before the packet has been forwarded TX RX Next RX Channel Cd Channel Cc Note: Opportunistic crossover needs the next receiver to sense the channel busy (not only the case)

15. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Implementation DCMA/AP has been coded within the OmNet++ based on the object oriented C++.

16. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Preliminary results Decrease of transmission packet delay Increase of network flexibility in terms of access to the channel and node scalability. Some increase of partial overlapping transmissions on the control channel Cc following an increase of packet generation rate In general, initial results follow our expectation that an improved performance could compensate for the increase of energy consumption due to two channel utilization

17. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan The MERLIN architecture: The TDMA/CSMA hybrid approach, the MERLIN protocol as an example

18. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Scheduling tables: V-schedule vs. X- Schedule Frame is divided in 8 slots; Nodes in the same zone transmit simultaneously The X scheduling is obtained by super positioning 2 V-sched one of which upside- down Nodes go into sleep immediately after the transmission

19. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Average end-to-end packet delay X-scheduling vs V-scheduling 0 50 100 150 200 250 300 0.40.60.811.21.41.61.82 Frametime (sec) Network Lifetime (days) X Scheduling V-Scheduling 1 Gateway 100 Nodes rand. Distributed. 800*500 area network Min signal strength(12 m) 50 msg/min sent by 5 rand. nodes Static network Delay calculated in the worst case scenario: 2 sec frametime Operational network lifetime The X scheduling used for applications in which some energy can be traded off for a decrease of latency of messages and for applications in which latency is a tighter constraint; V-scheduling used for low data traffic applications where the need for saving energy is of paramount importance.

20. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan What about an augmented intelligence of decision making? …A multi agent systems to: Planning to take long term decisions (not only if the-else based) Migrate to an area affected by an anomalous event To improve the adaptivity of the networks Decision based on information from different layers Better cope with dynamic changes of the network conditions. To take local decision between neighbouring nodes rather than at the gateway. Hence: – Energy saving –More accurate and faster response to network changes –Increase of preciseness of the action taken

21. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Multi agent system

22. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Our proposed solution: Migrating agent Moderate energy consuming An example:

23. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Disadvantages Accommodate BDI agents is very challenging due to devices computationally limited Debugging agent systems during ongoing applications is very challenging (sensors have only 3 leds provided) Traditionally Multi agent systems (MAS) are java oriented -> JVM needed

24. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Debugging:Agents-nodes mapping at the BS One-to-One –Each node is controlled by one agent that deliberates accordingly –Nodes can be seen as agent perceptors Many-to-One – Many agents map to an individual node –E.g. useful when nodes have several sensory modalities One-to-Many –A single agent map to a group of neighbouring nodes –E.g. useful when decision may be taken by analysing a group of nodes locally placed

25. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Methodology phase 1: Centralised Base station implementation A single agent placed at the BS The agent receives raw data from nodes then analyse them The agent identifies and solve anomalous behaviour of the network or part of it. The agent communicate to the BS what action to take.

26. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Methodology phase 2: Distributed Base station implementation The second phase transforms the centralised solution in a distributed agent-base implementation The key point of this phase is to have a mapping between agents of a MAS and sensor nodes

27. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Methodology phase 3: Distributed agents implementation Agents on the nodes can be modelled through the agents at the BS Hence, agents on the nodes can be easily debugged at the BS The distributed implementation can be achieved by mapping the statements that govern the agents behaviour (such as commitment rules) to the language of the device.

28. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Power management through network coverage:

29. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Integrated Sensor and Routing Coverage Agents Definitions: Sensing Coverage: If any node within the sensed area is covered by at least 1 sensor Redundant node: It can be switched off without affecting the level of coverage provided by the network; Routing Coverage: It exists at least one communication path from any node within the network to the gateway gateway

30. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Opportunistic Power Management Disconnecting sensors based on solely sensing coverage may lead to a disconnected network; for example: Interpolation for sensing Coverage can be used when nodes does not have a well defined sensing radius e.g sensing temperature A Time-Zones network division for Routing coverage can be used when nodes does not have a well defined transmitting radius e.g channel irregularities Redundant based on routing and sensing coverage Hibernating redundant sensors, must be decided based on both routing and sensing connectivity! u Zone n+1 Zone n-1 Zone n Redundant based on sensor coverage Gateway Disconnected Transmission radius Integrated Sensor and Routing Coverage Agents

31. UNIVERSITY COLLEGE DUBLINDUBLIN CITY UNIVERSITY SMI || NCSR || CDVP Ruzzelli, O’Hare, Jurdak, Tynan Questions and comments are welcome Thank you for your kind attention!