Environmental Energy Harvesting

1. Presented by: Chaitanya K. SambharaPaper by: Aman Kansal and Mani B Srivastava University of California Los Angeles - Instructor : Dr Yingshu Li Environmental Energy Harvesting Framework for Sensor Networks

2. introduction

3. Introduction Sensor networks are energy constrained. Sensor networks can increase their lifetime by extracting energy from their environment. Significant improvement can be achieved if task allocation is aligned by spatio-temporal characteristics of energy availability.

4. Issues we currently face • To provide maximum autonomy to the system with minimum or no human intervention. • Many times the size of the battery exceeds system components in many sensor nodes.

5. Solutions • To provide the network with capabilities to automatically feed itself from its environment. • For example : Solar power, microbial fuel cells, vibrations and acoustic noise.

6. The main idea • The energy availability is not homogeneous at all nodes. • Can we adapt task distribution among nodes to the detailed characteristics of environmental energy availability?

7. Key contributions to this idea • Current energy aware task allotment strategies base their decision on residual energy at each node. • They do consider but not exploit the fact that replenishing capability may be present at some nodes. • Spatio-temporal properties of energy supply should be exploited for task allotment.

8. Harvesting Problem • The problem of extracting maximum work out of a given energy environment is called “harvesting problem”. • A distributed framework to solve harvesting problem is called Environmental Energy Harvesting Framework (EEHF). • No special proxies should be needed to be installed.

9. Tasks of EEHF • Adaptively learn energy properties of environment and renewal property at each node through local measurements. • Make the brief information available for energy aware task assignment load balancing, leader elections and energy aware communication.

10. Where EEHF comes handy? • The usability of a network does not exclusively depend on the total energy available in all its nodes. • It also matters how the energy is distributed among the nodes. • What if a particular section of network dies?

11. Design Challenges • Workload in the network may not follow replenishment cycles. • If the residual battery keeps decreasing. • Knowledge of residual energy alone is not sufficient to know how much extra energy can be consumed.

12. Design Issues • Learning the environment • Sharing network-wide information.

13. Framework Design • The following parameters are used to track the characteristics of energy availability: • T: time epoch over which availability prediction is made • Em: mean energy expected in subsequent T duration • Ecm: energy consumed in every T interval on tasks not in control • of the scheduler • η: prediction confidence, a number between 0 and 1 • Φ : information regarding when the next recharging opportunity is • expected within next T time • Eb: current battery remaining • B: maximum battery size, beyond which environmentally available • energy cannot be stored • The challenge is now reduced to finding a single cost C at each node based on the above parameters: • C = f(T,Em,Ecm, η, Φ,Eb,B)

14. Interactions between various EEHF algorithms • Sensor networks may consist of hundreds of nodes, making manual configuration impossible • Enables IP nodes to become communication ready without user involvement.

15. Parameterize • This block combines the parameters learnt by the above blocks and the remaining battery Eb into one cost metric. • Thus, an effective battery, E, and the cost C can be calculated as: E = w1(Em - Ecm) + w2Eb Where w1 and w2 are assigned so as to give a much higher weight to replenishable energy.

16. Parameterize • C = 1/E • Also C can be defined as C = 1/(1- G), where G is the amount of battery already used

17. Scalability Friendly Information Exchange • In-network Averaging: • A load balancing scheduler may want to assign loads in proportion to the effective batteries at the nodes. • The nodes can then volunteer to accept a workload proportional to L = (E -Eav) / (Emax -Eav). • Distributed Scheduler : • Certain tasking algorithms learn the local costs on their own as required. • Such an approach can only be used when the scheduler itself is distributed.

18. Example Application: Routing • We compare the performance of EEHF and a residual energy based scheme, for routing. • The simulated sensor network has N = 100 nodes in a 100m 100m region, with nodes having a radio range of 20m. • The nodes are assumed randomly placed in the region for which light intensity variations have been collected, distributing a fraction p in well-lit regions and (1-p) in dark regions

19. when p = 0.25

20. When p = 0.50

21. When p = 0.75

22. Conclusion • Problem of energy harvesting was introduced here. • EEHF takes the first step towards development of methods to completely align the task distribution in the network with energy availability. • Future work includes determining the maximum work that can be extracted from a given energy environment

23. Questions ?