An Architecture for a Power-Aware Distributed Microsensor Node · An Architecture for a Power-Aware...

An Architecture for a Power-Aware Distributed Microsensor Node

Rex Min, Manish Bhardwaj, Seong-Hwan Cho, Amit Sinha, Eugene Shih, Alice Wang, and

Anantha Chandrakasan

Massachusetts Institute of Technology

October 12, 2000

Distributed Microsensor Networks

An alternative to macrosensorso Small, ubiquitous, easily deployed nodes

oCollaborative data gathering, ad hoc networking for fault-tolerance

o Battery replacement not an option:

Smart Dust PicoRadio WINS

n Some prototype nodes under active research:

How to achieve months/years of operation from a single battery?

Operational Characteristics

n Event driveno Low duty cycles

n Low bandwidtho bits/sec to kbits/sec

n High Spatial Densityo 0.1 nodes/m2 to 20 nodes/m2

n Short transmission distanceso 5-10m typical (< 100m)

n High operational diversity

...from the environmento Event arrival rate/type

o Ambient noise

o Signal statistics

...from network roleso Sensor

o Relay

o Data aggregator

:

...from user demandso Tolerable latency

o Result SNR

o Pr(Detection)

Characterizing Diversity

Scenario

Ene

rgy

Esystem2

di

Esystem1

n Scenario: the space of possible operating points

n Scenario distribution: quantifies diversity as a distribution di

n Energy curve: Esystem characterized for each operating point

Power-aware systems have a low Esystemdi product

A Power-Aware Sensor Node

MIT µAMPS: Adaptive Multi-Domain Power-Aware SensorsA sensor node that demonstrates power-aware methodologies

Scenario

Eperfect

Ene

rgy

Esystemdi

ACTIVE IDLE ACTIVE

n Component-by-component optimization

n Reduction of worst-case power dissipation

n Graceful energy scalability across a diversity of operating conditions

n Energy-quality trade-offs

n Collaboration across levels of the system hierarchy

Low-Power Design Power-Aware Methodologies

Power-Aware Node Architecture

Leakage current Workload variation

Bias currentStart-up time

Standby currentLow duty cycle

Capacity variations

Efficiency variations

Desired result quality variations

Available energyVoltage scheduling

RadioSA-1100A/DSeismic Sensor

Acoustic Sensor

ROMRAM

DC-DC Conversion

Battery

Power

Protocols

Algorithms

µOS

Node Prototype

sensor/processor boardradio baseband

miniaturized DVS control(Nathan Ickes)

n Version 1 prototype with COTS components

n Advanced iterations will feature custom chipsets

Power-Aware Methodologies

n Idle mode: deepest possible shutdown with minimum overheado Leakage current control

oRadio duty cycles

o Sleep state assignment

n Active mode: scalability in energy consumption exploits node’s operational diversityoDynamic voltage scaling

o Energy-quality scalable algorithms

oVariable-strength error correction

oCurrent profiles for maximum battery capacity

Idle Mode: Leakage Current Control

n Leakage dominates switching energy for low duty cycles

n A major concern for event-driven sensor operation

Measurements from SA-1100

Eswitch = CtotVDD2

Eleak = VDD (I0 e ) tVDDn’VT

Radio Considerations

n Startup energy can dominate transmission energy for short distances and packetsn Demands power-aware network protocolsn Favors buffering if latency is tolerable

101

102

103

104

10510

-8

10-7

10-6

10-5

Packet Size

Ene

rgy/

bit (

J)

1 Mbps data rate0 dBm Tx power20 mW electronics

EFIXED = CVDD2

Active Mode: Dynamic Voltage Scaling

0.2 0.4 0.8 1.0

0.2

0.4

0.6

0.8

1.0

Normalized Workload

Nor

mal

ized

Ene

rgy

Fixed Supply

VariableSupply

0

0 0.6

ACTIVE IDLE

Fixed Power Supply

ACTIVE

EVAR = C (VDD/2)2 = EFIXED / 4

Variable Power Supply

12

12

DVS Implementation

SA-1100

Control

µOS

VoutController

Power

3.6V

Voltage request, 0.9 - 1.5 V

5

1.6V limiter

5

digitally adjustable DC-DC converter powers SA-1100 core

µOS selects appropriate clock frequency based on workload and latency constraints

SA-1100 requests a voltage appropriate for its clock frequency

Energy Characterization of SA-1100

Eswitch = CtotVDD2 Eleak = VDD (I0 e ) t

VDDn’VT

n Energy per operation at full processor utilization

fixed voltage operation

variablevoltageoperation

0 50 100 150 2000

0.2

0.4

0.6

0.8

1

Impulse Response Length

Nor

mal

ized

Ene

rgy

VariableVoltage

FixedVoltage

59MHz

73MHz

89MHz

206MHz

Application: Energy-Scalable Digital Filter

n Processor workload proportional to filter length

n µOS adjusts clock frequency (and voltage) with workload

w1[n]

wM[n]

w2[n]

s1[n]

y[n]s2[n]

sM[n]

Σ

Sensor Network

Beamforming Node

Algorithmic Considerations: Beamforming

n Data aggregation from multiple sensors into a single, high-SNR resultoData redundancy removed => less network transmission

o Energy-scalable: vary the number of input signals to beamformer

oHow is quality affected?

65

43

21

50m

10m

target

B

A

Number of sensors

Qua

lity

(mat

ched

filte

r res

pons

e)

A

B

Beamforming Example

more energy consumed

LMS Beamforming in order (choosing sensors 1...N) leads to energy-quality variations with source location

Example scenario with sensor cluster, target, and interference source

interference

Number of sensors

Qua

lity

(mat

ched

filte

r res

pons

e)

A

B

Number of sensors

Qua

lity

(mat

ched

filte

r res

pons

e)

A

B

AfterBefore

n Most significant first transformation improves energy-quality characteristicsoQuicksort signals by their SNR; beamform with strongest signals

o Low overhead transformation: 0.44% overhead for 2-sensor (worst) case

Power-Aware Transformation

Conclusions

n Power-awareness means...oGraceful energy-quality scalability at all levels

oHardware-software collaboration to save energy

oAccounting for the unique power dissipation characteristics of the target application

n For a long lifetime, microsensor nodes must be power-aware