Tragedy of the Coulombs

Transcription

1 Tragedy of the Coulombs Federating Energy Storage for Tiny, Intermittently-Powered Sensors Josiah Hester Lanny Sitanayah Jacob Sorber SenSys 15 November 2, 215

2 Sensing 2

3 3 This Talk 1. Batteryless sensing challenges 2. Problems with centralized energy 3. Federating energy

4 4 Vision Sophisticated sensing on batteryless devices

5 5 Batteryless Devices UMass Moo

6 5 Batteryless Devices UMass Moo

7 5 Batteryless Devices UMass Moo

8 5 Batteryless Devices UMass Moo

9 6 Batteryless Devices As small as possible Minimal energy storage (Cap) UMich Moo Harvest energy (RF, Solar, Glucose) Run when you can Frequent failures Erratic supply

10 7 Applications Infrastructure Monitoring Pipelines Bridges Roads Wildlife Tracking Small animal Implant once Building Monitoring Occupancy Energy Waste Wearables Clothing Jewelry

11 Tragedy of the Commons 8

12 Tragedy of the Commons 8

13 Tragedy of the Commons 8

14 Tragedy of the Commons 8

15 9 Greenhouse Monitoring Water waste is a problem Overwatering typical Coarse data on plants Dense sensing is a solution Fine grained plant information Cheap, non-invasive, long lived

16 1 Greenhouse Monitoring Radio Harvesting Computation Storage Sensing

17 11 Batteryless Sensing 6 Supply Voltage (V) MCU Humidity Sensor Leaf Wetness Radio

18 11 Batteryless Sensing 6 Supply Voltage (V) MCU Humidity Sensor Leaf Usable Range Wetness Radio

19 11 Batteryless Sensing 6 Supply Voltage (V) MCU Humidity Sensor Leaf Usable Range Wetness Radio Voltage requirements vary

20 12 Batteryless Sensing Sensor 4 Sensor Voltage Time (s)

21 12 Batteryless Sensing Sensor 4 Sensor Voltage Attempt Send Pkt Time (s)

22 12 Batteryless Sensing Sensor 4 Sensor Voltage Attempt Send Pkt Reset Time (s)

23 12 Batteryless Sensing Sensor 4 Sensor Voltage Attempt Send Pkt Reset Task coupling increases failure Time (s)

24 13 Batteryless Sensing Sensor Sensor Voltage Could have sent Overestimate Actual Time (s)

25 13 Batteryless Sensing Sensor Sensor Voltage Could have sent Overestimate Actual Task coupling decreases utility Time (s)

26 14 Scheduling Tasks Energy Cost (mj) Greenhouse Monitoring 1s Computation 1x Leaf Rds 1x Humidity Rds Send 1 Packet Task

27 15 Scheduling Tasks.5 Energy Stored (mj) Send 1 Pkt Sample 1x Leaf Sample 1x Humidity 1s of Computation Time (s)

28 16 Scheduling Tasks.5 Energy Stored (mj) Time (s)

29 16 Scheduling Tasks.5 Energy Stored (mj) Start Threshold 1.3 seconds charge time Time (s)

30 17 Scheduling Tasks Capacitor Voltage Large Small Start Threshold Time (s)

31 17 Scheduling Tasks Capacitor Voltage Large Small Start Threshold Time (s) 1.2s charge

32 17 Scheduling Tasks Capacitor Voltage Large Small Start Threshold s charge Time (s) 1.2s charge

33 18 Batteryless Sensing Batteryless sensing is hard Tasks are coupled, causing death Execution is not predictable Low energy tasks wait on high energy tasks Because energy storage is centralized

34 18 Batteryless Sensing Batteryless sensing is hard Tasks are coupled, causing death Execution is not predictable Low energy tasks wait on high energy tasks Because energy storage is centralized What if we federate energy storage?

35 19 Federated Energy Benefits useful work starts sooner fewer power failures simpler application decisions relaxes voltage coupling increases energy harvested

36 2 UFoP Multiple capacitors One for microcontroller One for each peripheral Static rate and priority + Charge Controller Sensor Ready? Radio Ready? Energy Harvesting Core Sensor Radio -

37 21 Start work sooner UFoP MCU Radio MCU Volts Volts Ready Time (s) Ready Time (s)

38 22 Decrease power failures Centralized Capacitor Voltage Attempt Send Pkt MCU Reset Time (s)

39 23 Decrease power failures Federated Capacitor Voltage Attempt Send Pkt Radio Fail Radio Cap MCU Cap Time (s)

40 24 Simplify app decisions Centralized Capacitor Voltage Compute Send Pkt Time (s)

41 25 Simplify app decisions Capacitor Voltage Compute Federated Send Pkt Time (s) Radio Cap MCU Cap Compute

42 26 Relax voltage constraints 6 Supply Voltage (V) MCU Humidity Sensor Leaf Wetness Radio

43 26 Relax voltage constraints 6 Supply Voltage (V) MCU Humidity Sensor Centralized Range Leaf Wetness Radio

44 26 Relax voltage constraints 6 Supply Voltage (V) MCU Humidity Sensor Centralized Range Leaf Wetness Radio Federated Range

45 27 Harvest more energy Load

46 27 Harvest more energy Power Load

47 27 Harvest more energy Power volts Load

48 28 Harvest more energy Current (I) Volts (V) Power, P=IV

49 29 Harvest more energy Federated Volts Centralized Volts Time (s)

50 29 Harvest more energy Federated Volts Centralized Volts Time (s)

51 29 Harvest more energy Federated Volts Centralized Volts Reduces voltage volatility Time (s)

52 3 Implementation Energy Harvesting Charging Control Solar or Thermal or RF or Kinetic UFoP Controller (Custom PCB) DC Power 1st stage Capacitor Peripheral Control Cap Cap MCU (MSP43) Peripherals Sensor Radio Current Flow Control Signals

53 31 Results Evaluation Availability Resiliency Energy harvested Deployment Greenhouse monitoring

54 Methodology 32

55 32 Methodology RF Low RF High

56 32 Methodology RF Low RF High

57 32 Methodology RF Low RF High Solar

58 33 Availability volts mcu radio sensor volts Centralized Federated 1.5 mcu radio sensor time(s) First Stage Capacitor Radio Capacitor Sensor Capacitor More computational availability

59 34 Resiliency UFoP Centralized 1% 3 7.5% % % 7.5 % Radio Transmit Fails Low Voltage Events More resilient than without federating

60 35 Energy Harvested 12 UFoP Centralized Energy (mj) Solar RF High Energy RF Low Energy More energy harvested

61 36 Deployment 1% availablity 75% 5% 25% storage Centralized MCU Centralized Radio UFoP MCU UFoP Radio % Afternoon Evening Morning

62 37 Future Work Dynamic UFoP Dynamic priority, capacity, thresholds Resiliency Energy harvested Language Manage energy and time

63 Summary Federating energy storage dramatically improves the utility of batteryless sensors. persist.cs.clemson.edu This research is based upon work supported by the National Science Foundation under grants ACI and CNS Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation 38