Transcript Harvesting-aware Routing

Harvesting-aware Power
Management for
Sensor Networks
Aman Kansal, Jason Hsu,
Vijay Raghunathan & Mani Srivastava
Networked & Embedded Systems Lab
Center for Embedded Networked Sensing
University of California at Los Angeles
Energy Harvesting in Sensor Networks
Energy neutrality a holy grail for sensor networks used in long-term
monitoring applications
Minimize logistical and access costs associated with
replacement of batteries
Wireless sensor nodes with energy harvesting capabilities
Trio/Prometheus
(Solar, Berkeley)
Piezoelectric Windmill
(Wind, UT Arlington)
Commercial Platforms
(Solar/Mechanical, EnOcean)
This Talk
Platform design considerations
Experience in designing and deploying HelioMote,
a solar-powered wireless sensor node platform
Power management techniques
Harvesting-aware energy management of sensor nodes
and sensor networks
HelioMote: A Solar Energy Harvesting
Wireless Sensing Platform
Solar Cells
Overcharge Protection
NiMH Batteries
Undercharge Protection
DC Step-Up Converter
Monitor
Design Challenges
What energy modality?
Single or multiple?
Energy
Transducer
Harvesting
Circuit
Energy
Storage
Energy
Harvesting
& Storage
Manager
CPU
Radio
Sensors
Actuators
Sensor Node
Environmental Energy Sources
Piezoelectric
Highest
power
density
Electromagnetic
Underwater piezo-eel
Photovoltaic
Thermoelectric
Design Challenges
How to maximize energy extraction?
Energy
Transducer
Harvesting
Circuit
Energy
Storage
Energy
Harvesting
& Storage
Manager
CPU
Radio
Sensors
Actuators
Sensor Node
Harvesting Circuit Design
ISC
VOC
SolarWorld’s 3.75” x 2.5” solar panel
Solar panels behave very differently from batteries
Voltage-limited current source with Maximal Power Point (MPP)
Commercial MPP ICs too power hungry
digitally controlled switching regulators that isolate the load and present
desired impedance to the panel
HelioMote opts for low-overhead near-MPP operation by careful choice of panel
and secondary battery
Clamps panel to a battery forcing operation at a battery-dictated voltage
Design Challenges
Is energy buffer needed?
Capacitor or battery?
What battery chemistry?
Energy
Transducer
Harvesting
Circuit
Energy
Storage
Energy
Harvesting
& Storage
Manager
CPU
Radio
Sensors
Actuators
Sensor Node
Energy Storage Technologies
Rechargeable Battery
Ultracapacitor
Specific energy vs. power
Great energy density
Great power density
Efficiency
Moderate
High
Cost
Cheap
Expensive
Recharge cycles
O(100) - O(100000)
Unlimited
Self-discharge
Low
High
Self-discharge
Low
High
Aging
High (Li), Low (Ni)
Moderate
Other issues
Cold weather
Balancing failure
Choice is a Function of Duty Cycle
(a) Direct: better at very low or very high duty cycles
BATT
Solar Panel
Energy Consumer
(Application)
(b) Switched: better at low-to-moderate duty cycles with near-neutral
ambient energy availability
Solar Panel
Ultracapacitor
Energy Consumer
(Application)
Design Challenges
Energy
Transducer
Harvesting
Circuit
Energy
Storage
Energy
Harvesting
& Storage
Manager
How to route energy?
Analog or digital or s/w?
CPU
Radio
Sensors
Actuators
Sensor Node
Energy Storage Management
Independent
Load
CPU
Micropower
Reference
ADC
Switch
Input protection
AC
Switch
Sensor
AC
Micropower
Reference
Digital
Analog
Active all the time
Sleep energy is wasted
CPU (sleep) =
5-50uA
Input protection = 5-20uA
Active energy is huge
ADC = 200-400uA, CPU = 10 mA
Comparators = 3-5uA
References = 1-2uA
Summary of HelioMote Design Choices
Battery: 2 AA NiMH (2400 mAH)
Management: Autonomous,
Analog
Solar Panel: Autonomous,
optimal power point operation
225 mW effective at peak sun
Data Collection: High-accuracy
charge accumulation,
temperature, run-time,
and voltage
Power Characteristics
Voltage: 2.91V regulated
Consumption: 20 mA (active),
0.09 mA (sleep)
Efficiency 80% (active), 50%
(sleep)
Roundtrip battery efficiency:
66%
Self-dischagre: 1% per day
HelioMote in Real-life Deployments
Battery Voltage vs. Time
Current accumulator vs. Time
Snapshot from a 3-month deployment in LA
Many academic and industrial users across several countries
Open-source hardware and software, as well as commercial
ruggedized version
How long will HelioMote last?
NASA surface meteorology and solar energy data for
Los Angeles (34 N, 118 W) for December
Average daily insolation (horizontal): 2.60 kWH / m2
Worst case NO-SUN days over 14 day period is 4.99 days
Solar panel provides 585mWH (2106J) per day
Panel directly powers Heliomote for 2 hours a day
Energy is partially drawn from battery the rest of the time
Two scenarios analyzed
Node receives unobstructed sunlight throughout day
Node is in shade for 50% of the time
Perpetual operation feasible?
Results of Analysis: HelioMote in
LA Winter
Duty
cycle
Power
(mW)
50% of Day in Shade
Surplus
Energy (%)
Discharge
Depth (%)
Unobstructed Node
Lifetime
(years)
Surplus
Energy (%)
Discharge
Depth (%)
Lifetime
(years)
1%
1.24
1.84
1.49
25 years
5.20
1.47
25 yrs
5%
4.15
0.65
2.66
23 years
4.02
2.58
23 yrs
7.5%
5.96
3.29
3.28
22 yrs
10%
7.78
2.55
3.97
21 yrs
1.08
5.36
19 yrs
15%
11.41
20%
15.05
Energy from panel insufficient to
provide perpetual operation
Energy from panel insufficient to
provide perpetual operation
Lifetime = min (time to first outage, battery degradation to 80%)
Even with obstructions, sustained operation at 7% duty cycle is feasible (18% without
obstructions)
Experimental numbers show sustained operation at ~ 60% duty cycle in LA summer
and ~ 20% during LA winter
Energy supply is 3X higher in Summer (7.25 kWH/m2)
Realistic Notion of Perpetuity
Component failures and degradation
Battery: 5-20 years
Ultracapacitor: 2-20 years
Solar panel: 2 – 10 years
Thin-film: 2-10 years
Crystalline: ~20 years
http://www.boatus.com/boattech/SolarPanels.htm
Environmental issues
Dust and debris accumulates on surface and
blocks light (forcing premature servicing, so
just change the battery)
Seasonal changes affect light availability at a
given point
Vegetation growth over time
Debris and Vegetation greatly reduce
solar panel efficiency
So, realistically, lifetime beyond 10-20 years
is wishful!
Solar panel shows sign of rust after
2 months of deployment
Design Challenges
Energy
Transducer
Harvesting
Circuit
Energy
Storage
Energy
Harvesting
& Storage
Manager
How to schedule node
operations?
CPU
Radio
Sensors
Actuators
Sensor Node
Management of Energy Harvesting
Variation in harvesting opportunities
E.g. harvested energy is a function of node location,
time-of-day, aging, duration of energy storage etc.
How to extract maximum performance?
How to achieve energy neutral operation?
Isn’t Residual Battery Energy
Awareness Enough?
Node A
Eb
Path 1
Es per day, all before 12N
Destination
Source
Node B
Path 2
Es per day, all after 12N
Eb
Scenario:
1. Routing costs Er per hour
2. One hour of routing before 12N, and one hour after 12N
3. Roundtrip battery efficiency 
Residual Battery at 12N
Node A
Eb+(Es -Er)
Path 1
Destination
Source
Node B
Harvesting-aware Routing
Eb
Node A
Eb+Es
Destination
Source
Node B
Path 2
Eb-Er
Battery-aware Routing
Residual Battery at End-of-day
Node A
Eb+(Es -Er)
Destination
Source
Node B
Path 2
Harvesting-aware Routing
Eb+(Es -Er)
Node A
Eb+Es-Er
Path 1
Destination
Source
Node B
Eb +E -Er
Battery-aware Routing
Harvesting-aware Power Management
Goal is not power minimization but energy neutrality
Indefinitely long lifetime, limited only by h/w longevity
Subject to performance constraints and optimization
Unknown spatiotemporal profile of harvested energy
At a node: adapt temporal profile of workload
In a n/w: adapt spatial profile of workload (across nodes)
Learn Ambient
Energy
Characteristics
Learn
Consumption
Statistics
Duty
Cycling
Predict Future
Energy
Opportunity
Resource
Scheduling
Routing
Topology
Control
Understanding Energy Neutrality:
A Harvesting Theory
Condition for energy neutrality with a battery with
roundtrip efficiency  and leakage leak is
T
  Ps (t)  Pc (t)  dt 
0
T
 P (t)  P (t)
c
s

dt 
0
T

leak
dt  B0  0
T  [0,)
0
Modeling bursty energy source Ps(t) and consumer Pc(t)
T
T
 P (t)   T  
s
0
1
1
 P (t)   T  
s
1
T
2
0
Sufficient conditions for energy neutrality
2  1  leak


B0   2   3
B  B0
 P (t)   T  
c
0
2
3
At a Node:
Harvesting-aware Duty Cycling
Duty cycling between active and low-power states for power scaling
Approach
System utility function as a function of D
Time slots T with duty cycle calculated for a window of Nw slots
TxNw = a natural energy neutral period such as 1 day
At start of window predict harvested energy level for next TxNw
slots using history and external weather predictions
Calculate D for Nw slots for max U subject to energy neutrality
Revise duty cycle allocations based on actual observed Ps(t)
Application Utility vs. Duty Cycle
Stored vs. Direct Solar Energy Usage
Practical Dynamic Duty Cycle
Adaptation
12
70
72-day deployment @ LA
50
Current (mA)
Prediction errors
10
abs error (mA)
60
40
30
20
8
6
4
2
10
0
0
10
20
30
40
50
60
70
0
0
5
10
Day
15
Time(H)
20
25
Solar Energy Utilization(%)
100
90
80
Optimal
Adaptive
Simple
Optimal
Oracle, LP solution
Naive
70
Constant over a day
based on predicted total
energy
60
50
40
0.4
Dynamic
0.5
0.6
0.7
0.8
Battery roundtrip efficiency ()
0.9
1
Adaptive control based
on error and duty cycle
limits
Across a Network:
Harvesting-aware Routing
Link Cost = 1/E
Learn Local Energy
Characteristics
1/E(A))
A
1/E(B))
Predict
Future
Energy
Opportunity
B Distributed
Decision
for
Scheduling
Duty
Cycle
Routing
• Basic
scheme: Measure average energy
Learn
Consumption
•received
Use a distributed
routing
algorithm
that
per
day,
E
(n)
current
Statistics
assigns
routes
based
on
this link
cost
• Combine
this
metric
linearly
with
residual
battery,
Eday
Topology
res:
•
Enhanced
scheme:
learn
pattern
within
• e.g Bellman Ford
routing
T
Autoregressive
filter:
Control
E = [w1 w2] [Eav Eres]
where
[w1 w2]=is aE
a weight
Eav(n+1)
(n) +
currentvector
(1-a)Eav(n)
• Potential enhancement: predict consumption and
replace Eav by (Eav-Econsumption,av )
Harvesting-aware Routing Performance
morning
Afternoon
Battery Aware
Harvesting Aware
Simulation using light traces from James Reserve
Energy snapshots
Summary
Energy harvesting emerging as a viable technology for sensor
network deployments
Experience with first generation of platforms though significant
platform issues remain
Efficiency, aging & biofouling, multimodal harvesting
Challenges in providing performance and lifetime assurance under
highly-variable ambient energy availability
Harvesting theory for fundamental insights
Practical node and network level methods
For more info, visit http://nesl.ee.ucla.edu/projects/heliomote
Acknowledgements
Collaborators: Jonathan Friedman, Sadaf Zahedi
Research support: CENS, DARPA, NSF, ONR
Backup Slides
Impact on Solar Panel Efficiency
Ultracapacitor Direct to Solar Panel
(40mA panel - 20mA load)
Capacitor
Voltage
Capacitor/Panel Voltage
3
2.5
2
Radio Operation
Threshold
1.5
1
Normalized
Wasted Energy
0.5
0
-600
-400
-200
0
200
400
600
800
1000
1200
Time (seconds)
Capacitor induced voltage clamping lasting for 18
minutes leads to 36% waste of solar panel energy
Environmental
Energy Availability (J/Day)
The Bottom Line
90,720,000 delta energy points analyzed
Region where
Switched Ultracap
Architecture Extends
System Run-time
Application Duty Cycle