1. Sensor Networks
Lecture 8

2. LEACH Clustering
LEACH: (Low Energy Adaptive Clustering Hierarchy) rotates cluster heads to balance energy consumption
Each cluster head performs its duty for a period of time
Each sensor makes an independent decision on whether to become a cluster head and if yes broadcasts advertisement packets

3. LEACH Clustering (cont.)
Each sensor that is not a cluster head listens to advertisements and selects the closest cluster head
Once a cluster head knows the membership, a schedule is created for the transmission from sensors in the cluster to the cluster head to avoid collision (e.g., based on TDMA)
The cluster head can send a single packet to the base station (directly) over long distance to save energy consumption
No assurance of optimal cluster distributions

4. HEED Clustering
HEED (Hybrid Energy-Efficient Distributed clustering) uses the residual energy info for cluster head election to prolong sensor network lifetime
Probability of a sensor becoming a cluster head is:
Clusters are elected in iterations:
A sensor announces its intention to become a cluster head, along with a cost measure indicating communication cost if it were elected a cluster head
A non-CH sensor picks a candidate with the lowest cost
A non-CH sensor not covered doubles its CHprob in iterations until CHprob is 1, in which case the sensor elects itself to the cluster head

5. PEGASIS: Power-Efficient Gathering in Sensor Information Systems
A chain of sensors is formed for data transmission (could be formulated by the base station)
Finding the optimal chain is NP-complete
Sensor readings are aggregated hop by hop until a single packet is delivered to the base station: effective when aggregation is possible
Advantages: No long-distance data transmission; no overhead of maintaining cluster heads
Disadvantages:
Significant overhead: Can use tree instead
Disproportionate energy depletion (for sensors near the base station): Can rotate parent nodes in the tree

6. Aggregation/Duplicate Suppression
Aggregation of information in a tree structure
In-network information processing such as max, min, avg
Duplication Suppression:
On forwarding messages, sensor nodes whose values match those of other sensor nodes can simply annotate the message
Or just remain silent, on overhearing identical (or “similar enough”) values

7. Querying a Sensor Network
Can have sensor nodes periodically transmit sensor readings
More likely: Ask the sensor network a question and receive an answer
Issues:
Getting the request out to the nodes
Getting responses back from sensor nodes who have answers
Routing:
Directed Diffusion Routing
Geographic Forwarding (such as Geocasting)

8. Query-Oriented Routing
For query-oriented routing: Queries are disseminated from the base station to the sensor nodes in a feature zone
Sensor readings are sent by sensors to the base station in a reverse flooding order
Sensor nodes that receive multiple copies of the same message suppress forwarding

9. Query: Asking a Question

10. Response to Base Station: Initial

11. Directed Diffusion Routing
Direction: From source (sensors) to sink (base station)
Positive/negative feedback is used to encourage/discourage sensor nodes for forwarding messages toward the base station
Feedback can be based on delay in receiving data
Positive is sent to the first and negative is sent to others
A node will forward with low frequency unless it receives positive feedback
This feedback propagates throughout the sensor network to suppress multiple transmissions
Eventually message forwarding converges to the use of a single path with data aggregation for energy saving from the source to the base station

12. Responses, After Some Guidance
Use directed diffusion based on positive/negative feedback to guide response message forwarding

13. Directed Diffusion Routing Cont.
Pros
On demand route setup
Each node does aggregation and caching, thus good energy efficiency and low delay
Cons
Query-driven, not a good choice for continuous data delivery
Extra overhead for data matching and queries

14. Geographic Routing [Ref. 11]
For dense sensor networks such that a sensor is available in the direction of routing
Location of destination is sufficient to determine the routing orientation
Research issue:
selecting paths with a long lifetime for delivering messages between sensors, or from sensors to a base station without excessively consuming energy
Determining paths that avoid “holes” – determining the boundary or perimeter of a hole through local information exchanges periodically to trade energy consumption (for hole detection) vs. routing efficiency

15. Geographic Forwarding

16. References
Chapters 8-11, F. Adelstein, S.K.S. Gupta, G.G. Richard III and L. Schwiebert, Fundamentals of Mobile and Pervasive Computing, McGraw Hill, 2005.
Other References:
10. X. Yu, “Distributed cache updating for the dynamic source routing protocol,” IEEE Transactions on Mobile Computing, Vol. 5, No. 6, pp. 2006, pp
11. S. Wu and K.S. Candan, “Power-Aware Single and Multipath Geographic Routing in Sensor Networks,” Ad Hoc Networks, Vol. 5, 2007, pp. 974–997.

17. Fault Tolerance and Reliability
Sensor nodes are more susceptible to failure because of direct exposure to the environment and energy depletion
Failure and fault recovery are basic assumptions: incorporate redundancy to cope with failure
Performing consensus in a cluster for high reliability of measurement
Clustering based on sensing responsibility
Static vs. dynamic grouping
Dynamic grouping does not need to maintain state information and is more accurate (near the event) but incurs overhead in forming the group and reaching consensus

18. Searching for Agreement: Static Grouping

19. Searching for Agreement: Dynamic Grouping

20. MAC Layer Protocols
IEEE scheduling protocols are not suitable for wireless sensor networks because:
With RTS/CTS (Request to Send / Clear to Send) , collision can still occur because of hidden/expose terminal problems
Listening to traffic to avoid collision requires the nodes to stay on
TDMA is more suitable (requiring clock synchronization)
A number of reservation mini-slots can be used to reserve each of the transmission slots
Sensors can indicate whether or not they wish to transmit a message during the scheduling time segment
Nodes that are not planning to send or receive a packet need to stay on only during the reservation time slot to see if other sensors are sending a packet to them
Collisions are avoided, except for small reservation packets

21. Tradeoff between Energy Efficiency and Reliability/Performance
An important design issue
Improved reliability vs. energy consumption
Aggregating sensor readings vs. loss of information
Energy-efficient protocols often involve increased delay, loss of accuracy, reduced reliability and/or other performance penalty
Direct sensor-BS transmission vs. sensor-CH-BS
Sensor readings with redundancy
Achieving application requirements while prolonging lifetime is a major challenge

22. Fault Tolerant Data Propagation
Reference: [12] listed at the end
Use path redundancy to cope with sensor “reading” faults
One path (no redundancy)
Multiple paths to return sensor readings and a majority voting of the first three readings returned is performed to cope with faults
For example, use Time To Live (TTL) to indicate how many hops a sensor reading message is to be propagated, thereby creating multiple paths to propagate the sensor reading message from source to sink

23. Fault Tolerant Data Propagation
Source: node A
Sink: node I
When TTL = 3
hops, there are
7 paths from A to
When TTL=4 hops,
there are 21 paths

24. Fault Tolerant Data Propagation
An example
Source: node E
Sink: node I
p: link fault probability (causing
reading error)
q: node fault probability (causing
TTL=1: Reliability is 1-p
TTL=2: what is the reliability?
Three possible paths: E->I, E->H->I,
E->F->I, with fault probability of p A A
System fails when two out of three
paths fail, so reliability is
1-pA2-2pA(1-A)-(1-p)A2 where A=1-
(1-q)(1-p)2 =2p+q-2pq-p2+p2q
The more the path redundancy, the
higher the reliability at the expense
of more energy consumption

25. Energy Efficiency Metric: Mean Time to Failure (MTTF)
Time till the first node dies (not useful)
Time half of the sensor nodes die (too arbitrary)
Time when the sensor network can no longer perform its intended function (yeah!)
Difficult to define precisely
Designing protocols so that
All the sensors die at roughly the same time
Sensors die in random locations instead of in specific locations

26. Balancing Energy Consumption
Clustering – is it always good?
Triangular routing: sensors -> cluster head -> base station
Overhead in selecting and rotating among sensors to be cluster heads
Good only if message aggregation is feasible; otherwise directly sending sensing readings to the base station may end up saving energy more

27. Energy-Efficient Clustering
Reference: [13] listed at the end
Two key parameters:
p: probability of a sensor becoming a cluster head
k: number of hops covered by a cluster
Find optimal (p, k) that would minimize the energy consumed

28. Energy-Efficient Clustering: Formulation
Sensors are distributed following a homogeneous spatial Poisson process with intensity → in a square area of size 4a2
Per-hop distance is r
Energy model: each sensor uses 1 unit of energy to transmit or receive 1 unit of data
The information processing center is in the middle of the area
Idea: Define a function for the energy used and find (p, k) that would minimize the energy used

29. On Optimal Path and Source Redundancy in Sensor Networks
Reference: [14] listed at the end
Analyze the effect of redundancy on MTTF and determine the optimal path and source redundancy level to maximize MTTF while satisfying reliability (Rreq) and timeliness (Treq) QoS requirements in WSNs.
Develop a hop-by-hop data delivery mechanism utilizing source and path redundancy with the goal to satisfy QoS requirements while maximizing the lifetime of the sensor system
Query: must return a sensor reading to the PC within the real-time deadline.

30. Cluster based WSN architecture

31. Hop-by-Hop Data Delivery Protocol

32. Hop-by-hop Data Delivery Protocol
Based on localized geographic routing
Path redundancy: Form m paths from a source CH to the PC:
m SNs in hop one relay the data through broadcasting
only one SN relays the data in each of the subsequent hops in each path
Source redundancy: Each of the ms SNs to communicate with the source CH through a distinct path:
only one SN relays the data through broadcast in each of the subsequent hops in each path

33. Probability Model
System MTTF - Total number of queries the system can answer before it fails due to energy depletion, sensor faults, or channel error
Rq - Reliability of a query as a result of applying the hop-by-hop data delivery mechanism with m paths for path level redundancy and ms sensors for source level redundancy