Ultra-Low Energy Wireless

Sensor and Monitor Networks

Jan M. Rabaey
http://www.eecs.berkeley.edu/~jan
In cooperation with Profs B. Brodersen, A. Sangiovanni-Vincentelli,
K. Ramchandran, P. Wright, and the PicoRadio group.

The New Internet

The Berkeley Endeavour project

The Vision
Localizers
Tiny devices, chirping their impulse codes at one
another, using time of flight and distributed algorithms
to accurately locate each participating device. Several
thousands of them form the positioning grid Together
they were a form of low-level network, providing
information on the orientation, positioning and the
relative positioning of the electronic jets
It is quite self-sufficient. Just pulse them with
microwaves, maybe a dozen times a second
Pham Trinli, Thousands of years from now
Vernor Vinge,
A Deepness in the Sky, 1999

PicoRadios
Meso-scale low-cost (< 0.5 $) sensor-computationcommunication nodes for ubiquitous wireless
data acquisition that minimize power/energy dissipation
Minimize energy (<5 nJ/(correct) bit) for energy-limited source
Minimize power (< 100 mW) for power-limited source (enabling
energy scavenging)

By using the following strategies

self-configuring networks
fluid trade-off between communication and computation
Integrated System-on-a-Chip, using aggressive low-energy
architectures and circuits

Target date: 2004

The Smart Home

Dense network of
sensor and monitor nodes

Security
Environment monitoring and control
Object tagging
Identification

Wireless in the Home

Source: IEEE Spectrum,

December 99

Industrial Building Environment

Management

Task/ambient conditioning systems allow thermal

condition in small, localized zones (e.g. work-stations)
to be individually controlled by building occupants
micro-climates within a building
Requires dense network of sensor/monitor nodes
Wireless infrastructure provides flexibility in
composition and topology

The Interactive Museum

Wireless
node

Offices

Entrance

Exhibits

Cafe

The Enhanced User Interface

The virtual keyboard

(Kris Pister, UCB)

Other Applications:
Disaster mitigation, traffic management and control
Integrated patient monitoring, diagnostics, and drug administration
Automated manufacturing and intelligent assembly
Toys, etc

System Requirements and Constraints

Large numbers of nodes between 0.05 and 1
nodes/m2
Cheap (<0.5$) and small ( < 1 cm3)
Limited operation range of network maximum
50-100 m
Low data rates per node 1-10 bits/sec average
up to 10 kbit/sec in rare local connections to potentially support
non-latency critical voice channel

Low mobility (at least 90% of the nodes stationary)

Crucial Design Parameter:
Spatial capacity (or density) 100-200 bits/sec/m2

Fact or Fiction?
Today

Estimated: 2002-2003

Integrated radio
+ sensor ona-chip

How to get there?

Network
level

Functional & Performance

Requirements

Network Architecture

Performance analysis

Constraints

Node
level

Functional & Performance

Requirements

Node Architecture

Performance analysis

Think Energy!

Opportunities
Exploit the application properties
Sensor data is correlated in time and space
Sensor networks are query-based
Sensing without precise localization seldom makes
sense
Duty cycle of sensor nodes is very small

And use node architectures that excel in the

common case
Stream-based data flow processing for baseband
Concurrent Finite State Machines for protocol stack

Power & energy dissipation

in ad-hoc wireless networks

Plink

bits
a b dist
Pstandby
sec
actual _ bits
g
e a b dist
Pstandby
sec
g

Eactual _ bit e a b dist

actual _ bits sec

Pstandby

* These equations assume perfect power control

a: computation energy for

transceiving a single bit
b: transmission cost factor
for a single bit
g: path-loss exponent (2..4)
e: overhead (in extra bits
needed for transmission of
a single bit)
Pstandby: standby power (eg.,
due to need to keep
receiver on)

Saving Power at the Application Layer

Plink

bits
a b dist
Pstandby
sec
actual _ bits
g
e a b dist
Pstandby
sec
g

a: computation energy for

transceiving a single bit
b: transmission cost factor
for a single bit
g: path-loss exponent (2..4)
e: overhead (in extra bits
needed for transmission of
a single bit)
Pstandby: standby power (eg.,
due to need to keep
receiver on)

actual_bits/sec: source coding

Opportunity: sensor data is correlated in time and space
Trading off communication for computation

Distributed Source Coding (Ramchandran)

Sensor networks present major spatial data
correlation, but exploitation requires major intranode communication

Good theoretical question:

How much performance loss if such communication unavailable?

Answer: no performance loss! (under certain

conditions) based on non-constructive
information-theoretic argument (Slepian-Wolf).

Engineering question:
How do we develop a practical and constructive framework
to optimize global network usage?

Saving Power at the Network Layer

Plink

bits
a b dist
Pstandby
sec
actual _ bits
g
e a b dist
Pstandby
sec
g

a: computation energy for

transmission of a single bit
b: transmission cost factor
for a single bit
g: path-loss exponent (2..4)
e: overhead (in extra bits
needed for transmission of
a single bit)
Pstandby: standby power (eg.,
due to need to keep
receiver on)

dist: network partitioning using multi-hop

e: cost of network discovery and maintenance
Opportunities:
exploit application (i.e. sensoring) properties
merge with localization

Communicating over Long Distances

Multi-hop Networks
Source
Dest

Example:

1 hop over 50 m
1.25 nJ/bit

5 hops of 10 m each

log(b/a)

5 2 pJ/bit = 10 pJ/bit

Multi-hop reduces transmission energy by 125!

(assuming path loss exponent of 4)

Optimal number of hops needed for

free space path loss.

But network discovery

and maintenance overhead

Low-Energy Network Strategies

Reactive Routing good for rarely used routes

Proactive Routing good for frequently used routes

Need solution that is more adequate for problem at hand:

class (contents)-based and location (geographic)-based
addressing.
Routing Overhead (bytes)
4000
3500
3000
2500
2000
1500
1000
500
0

Routing Overhead (bytes) Normalized

DSDV
AODV
20

33

56

Number of Nodes

(discovering one route)

16000
14000
12000
10000
8000
6000
4000
2000
0

20

33

56

Number of Nodes

(discovering n routes)

Example: Directed Diffusion (Estrin)

Application-aware communication primitives
expressed in terms of named data (not in terms of the nodes
generating or requesting data)
Example: Give me temperature information

Nodes diffuse the interest towards producers via a

sequence of local interactions
This process sets up gradients in the network which
channel the delivery of data
Reinforcement and negative reinforcement used to
converge to efficient distribution
Intermediate nodes opportunistically fuse interests,
aggregate, correlate or cache data
Other options: swarm intelligence

Distributed Positioning
Average Position Error (% of grid dimensions)

10 nodes, 25 waves, anchor weight=5, 30 iterations, 30% anchors, NO gradients

10
8
6
4
2
0
60
50

40

40
30

20
Range Error (%)

20
0

10
0

Initial Position Error (%)

Merging locationing with networking leads to pruning

of overhead messaging

Saving Power at the Media Access Layer (!)

Plink

bits
a b dist
Pstandby
sec
actual _ bits
g
e a b dist
Pstandby
sec
g

a: computation energy for

transceiving a single bit
b: transmission cost factor
for a single bit
g: path-loss exponent (2..4)
e: overhead (in extra bits
needed for transmission of
a single bit)
Pstandby: standby power (eg.,
due to need to keep
receiver on)

Pstandby: Power-management of inactive nodes

e: Cost of collisions and retransmissions (interference)
Opportunities: distribution of communication in time and frequency
rendez-vous scheduling in local neighborhood
usage of multiple virtual channels reduces interference

Where Does Energy Go?

When idle: Channel Monitoring
Collision and Retransmission
Signaling overhead (header, control pkts)

Mostly-Sleepy MAC Layer Protocols

Computational energy for receiving a bit is larger
than the computational energy to transmit a bit
(receiver has to discriminate and synchronize)
Most MAC protocols assume that the receiver is
always on and listening!
Activity in sensor networks is low and random
Careful scheduling of activity pays off big time, but
has to be performed in distributed fashion

PicoMAC
Spread Spectrum Multi-Channel Scheme
To Reduce Collision Rate
To Reduce Signaling Overhead (Shrink Address Space)

Truly Reactive Messaging

Power Down the Whole Data Radio
Reduce Monitoring Energy Consumption by 103 Times
Wakeup Radio will Power Up Data Radio for Data Reception

Saving Power at the Physical Layer

bits
Pstandby
sec
actual _ bits
g
e a b dist
Pstandby
sec

Plink a b dist g

a: computation energy for

transceiving a single bit
b: transmission cost factor
for a single bit
g: path-loss exponent (2..4)
e: overhead (in extra bits
needed for transmission of
a single bit)
Pstandby: standby power (eg.,
due to need to keep
receiver on)

a: Choice of data rate, modulation, physical channel access

e: Cost of framing, channel coding, synchronization, CRC
Opportunities: radios with fast acquisition (and probably less perfect channel)

The Mostly Digital Radio

Analog

Digital

cos[2(2GHz)t]
RF input
(fc = 2GHz)

I (50MS/s)
A/D
Digital
Baseband
Receiver

RF filter

LNA

A/D
Q (50MS/s)

chip boundary
sin[2(2GHz)t]

Integration/Power Tradeoff
Example:

High carrier frequency allows small integrated

passive elements, but increases power consumption
Carrier Frequency Vs. Power Consumption

Why?

10

Increase Id to boost
device ft
Increased power
consumption in PLLs

0.1

10
Power Consumption (mW)

100

0.1
1000

Carrier Frequency (GHz)

1.2V Receiver
Price Label Radio
WINS
Wireless Hearing Aid
Super-Regenerative
Pager
Philips Pager - UAA2080
Bluetooth
BWRC D.C. Radio

PicoRadio Implementation Issues

Dynamic nature of PicoRadio networks requires
adaptive, flexible solution
Traditional programmable platforms cannot meet
the stringent low-power requirements
3 orders of magnitude in energy efficiency between custom
and programmable solutions

Configurable (parameterizable) architectures

combine energy efficiency with limited flexibility
System-on-a-chip approach enables integration
of heterogeneous and mixed mode modules on
same die
Predicted improvements: factor 10 each year!

The Energy-Flexibility Gap

Energy Efficiency
MOPS/mW (or MIPS/mW)

1000

Dedicated
HW
100
10

Reconfigurable
Processor/Logic
ASIPs
DSPs

Pleiades
10-80 MOPS/mW

2 V DSP: 3 MOPS/mW

Embedded Processors

SA110
0.4 MIPS/mW

0.1
Flexibility (Coverage)

(Re)configurable Computing:
Merging Efficiency and Versatility
Spatially programmed connection of processing elements.
Hardware customized to
specifics of problem.
Direct map of problem
specific dataflow, control.

Circuits adapted as
problem requirements
change.

Architecture Comparison
LMS Correlator at 1.67 MSymbols Data Rate
Complexity: 300 Mmult/sec and 357 Macc/sec

16 Mmacs/mW!
Note: TMS implementation requires 36 parallel processors to meet data rate validity questionable

Intercom TDMA MAC

Implementation alternatives
ASIC

FPGA

ARM8

Power 0.26mW
2.1mW
114mW
Energy 10.2pJ/op 81.4pJ/op n*457pJ/op
ASIC: 1V, 0.25 mm CMOS process
FPGA: 1.5 V 0.25 mm CMOS low-energy FPGA
ARM8: 1 V 25 MHz processor; n = 13,000
Ratio: 1 - 8 - >> 400

Idea: Exploit model of computation: concurrent finite state machines,

communicating through message passing

Envisioned PicoNode Platform

Reconfigurable
State Machines

Embedded uP

FPGA
Dedicated
DSP

Reconfigurable
DataPath

Small footprint directdownconversion R/F

frontend
Digital baseband
processing
implemented on
combination of fixed
and configurable
datapath structures
Protocol stack
implemented on
combination
FPGA/reconfigurable
state machines
Embedded
microprocessor running
at absolute minimal
rates

PicoNode I

Motorola StarTac
Cellular Battery (3.6V)
Serial Port
Window

Casing Cover

Connectors for
sensor boards

Pico Radio
Test Bed

Flexible platform for

experimentation on
networking and
protocol strategies
Size: 3x4x2
Power dissipation <
1 W (peak)
Multiple radio
modules: Bluetooth,
Proxim,
Collection of sensor
and monitor cards

PicoNode II (two-chip)
Custom
analog
circuitry

Mixed
analog/
digital

Fixed
logic

Programmable
logic
Embedded
Processor
(Xtensa)

Software
running on
processor

Memory
Protocol
Sub-system

Interconnect Network (Sonics Silicon Backplane)

ADC
Analog RF
DAC
Chip 1
Direct down-conversion front-end
(Yee et al)

Digital
Baseband
Baseband
processing
Processing

Fixed
Protocol Stack

Programmable
Protocol Stack
(FPGA)

Chip 2

The Holy Grail: Energy Scavenging

Power (Energy) Density
Batteries (Zinc-Air)
Batteries(Lithium ion)

Source of Estimates

1050 -1560 mWh/cm (1.4 V)

Published data from manufacturers

300 mWh/cm (3 - 4 V)

Published data from manufacturers

15 mW/cm - direct sun

Solar (Outdoors)

0.15mW/cm - cloudy day.

Published data and testing.

.006 mW/cm - my desk

Solar (Indoor)
Vibrations

0.57 mW/cm - 12 in. under a 60W bulb

Testing

0.001 - 0.1 mW/cm

Simulations and Testing

3E-6 mW/cm at 75 Db sound level

Acoustic Noise
Passive Human
Powered

9.6E-4 mW/cm2 at 100 Db sound level

1.8 mW (Shoe inserts >> 1 cm )

Published Study.

Thermal Conversion

0.0018 mW - 10 deg. C gradient

Published Study.

Direct Calculations from Acoustic Theory

80 mW/cm

Nuclear Reaction

1E6 mWh/cm
3
300 - 500 mW/cm

Fuel Cells

~4000 mWh/cm

SOURCE:
P. Wright & S. Randy
UC ME Dept.

Published Data.
Published Data.

Example: MEMS Variable Capacitor

Out of the plane, variable gap capacitor

springs

50m
500m

Proof mass

Integrated Manufacturing Lab

Up to 10 mW of power demonstrated

PicoRadio Design Challenges

6
7

PicoNode Architecture Design

10
18
17

11

19
20

12
16
13
15

14

Positioning
Network
Architecture

Power (Energy) Density

FPGA

Embedded uP

Batteries
(Zinc-Air) 1050 -1560 mWh/cm3
Batteries
(rechargeable Lithium) 300 mWh/cm3 (3 - 4 V)
15 mW/cm2 - direct sun
Solar
1mW/cm2 - ave. over 24 hrs.
Vibrations
0.05 - 0.5 mW/cm3
Inertial Human Power
Acoustic Noise
Non-Inertial Human
Power
Nuclear Reaction
One Time Chemical
Reaction

Dedicated FSM

3E-6 mW/cm2 at 75 Db
9.6E-4 mW/cm2 at 100 Db

Dedicated
DSP

1.8 mW (Shoe inserts)

80 mW/cm3
1E6m Wh/cm3

Reconfigurable
DataPath

Performance Analysis

Fluid Flow
Fuel Cells

300 - 500 mW/cm3

~4000 mWh/cm3

Energy Constraints
PicoNode

hopsoptimal ceil dist Total 10g

Offices
Entrance

Exhibits

Use Cases

Cafe

Conceptual Modeling