This is the talk I gave at SECON 2008, the 5th annual IEEE Communications Society Conference on Sensor, Mesh and Ad hoc Communications and Networks.

1. Gateway Design for Data
Gathering Sensor Networks
Presented by Raluca Musaloiu-E.
Johns Hopkins University
That’s Razvan Andreas
Me Musaloiu-E. Terzis

2. Easier to deploy a WSN than years ago.
TinyOS
MANTIS OS
MICAz
Contiki
SOS
RETOS
Tmote Sky

3. Life Under
Your Feet
WSNs for studying the
effects of abiotic factors
on soil animals.
http://lifeunderyourfeet.org

4. Koala
System for reliably
extracting bulk data from
duty-cycled nodes.

5. Current solutions for WSN
gateways are
not energy efficient
and bulky.

6. Architecture

7. Back-End
Server
Internet
Commands
Data
Requests
WSN
WSN

8. CPU 400 MHz Intel PXA255
Memory 64 MB SDRAM Stargate
32 MB Flash
CF and PCMCIA connectors
51-pin expansion connector
Low power consumption < 500 mA
http://www.xbow.com

9. CPU Low-power PIC, 32768 Hz
PowerNut
Supply voltage 7-30 V
Power 180 uA (worst case)
consumption
(downstream
device shutoff) 58-100 uA (typical)
http://www.jkmicro.com

10. Normal operation cycle
Establish the Download Download data from the motes Sleep
Boot uplink connection commands
Upload previous data
Fig. 2. The gateway’s normal operation cycle.
s state and the PowerNut itself consumes very little energy, This command specifies the number of seconds that
s solution is more efficient than putting the gateway to gateway should go to sleep at the end of the curr
ep-sleep mode. On the other hand, the gateway must go activity period. The gateway uses this value to set
ough its boot process every time power is restored. This time period for deep-sleep or to configure the amo
ot sequence, including loading the operating system, can of time the PowerNut power controller will disconn
ke tens of seconds. Section IV explores the trade-off between the power from the gateway itself.
wering off the gateway and using its deep-sleep mode. 3) REPORT <count>
This command specifies the number of operation cyc

11. Commands
TIME <curent time> REPORT <count>
SLEEP <seconds> SEND_LOG
UPDATE <file_location> wsn|gw

12. WSN interface
Koala Deluge Typhoon
download data network programming

13. Long haul connectivity

14. Evaluation

15. Deep-sleep versus Power-off mode
1
2 Energy efficiency of long haul radios
Lifetime estimation
3

16. 1
Deep-sleep versus Power-off mode
Decision is based on inactivity time.

17. Network up
Network down
Deep-sleep
500 mA
306 mA
247 mA
375 mA
144 mA
250 mA
123 mA
125 mA
102 mA
63 mA
83 mA
6 mA 23 mA 42 mA
0 mA
No cards Wi-Fi 3G No cards
No daughter-board With daughter-board

18. Average current drawn during the booting sequence
300 mA
225 mA
150 mA 280 mA 268 mA
203 mA
176 mA
75 mA
0 mA
No cards Wi-Fi 3G DB, no cards

19. Power-off mode is preferred if time > 3.9 min for 3G
(5.2 min for Wi-Fi, 16.5 min with no cards).
100
3G - deep-sleep
Wi-Fi - deep-sleep
no cards - deep-sleep
80 3G - power-off
Wi-Fi - power-off
no cards - power-off
Average Current (mA)
A
60
40
B C
20
0
0 500 1000 1500 2000
Inactivity Time (s)

20. 2
Energy efficiency of long haul radios
Measure the time to transfer a file.

21. 1.2 mAh to transfer 1 MB file in deep-sleep with Wi-Fi
(16 mAh with 3G).
18
Wi-Fi - deep-sleep
3G - deep-sleep
16 Wi-Fi - power-off
3G - power-off
14
Energy Consumption (mAh)
12
3G
10
8
6
Wi-Fi
4
2
0
100KB 250KB 500KB 750KB 1M

22. Energy consumption to transfer 1 MB file in power-off mode.
700 700
600 B C E 600 B
Current (mA)
Current (mA)
500 D 500
400
300 A
3G 400
300 A
200 200
100 100
0 0
0 20 40 60 80 100 120 140 160 180 0 20
Time (s)
Fig. 7. Current drawn during the transfer of a 1 MB file using the 3G radio Fig. 8. Current dr
in the power-off configuration. radio in the power-o
700
600 B C D E
Current (mA)
500
loading, PPP dial-up connection is set in section C, followed
A
400
300 Wi-Fi 55
in section D by the actual transfer, and a short section E in
200 50
100
which the connection is terminated. Figure 8 presents the same
0 45
0 20 40 60 80 100 120 140 160 180
stages when the Wi-Fi card is Time (s)
used. 40
ime [minutes]
Fig. 8. Current drawn during the transfer of a 1 MB file using the Wi-Fi 35
radioLifetime Estimation
C. in the power-off configuration.
30
We can now estimate the lifetime of a battery-operated

23. 3
Lifetime estimation
Estimate the life of a battery-operated gateway.

24. N number of WSN nodes 25
alpha storage threshold 0.25
M mote storage capacity 1 MB
B1 data generation rate 18 bytes/min
B2 data generation rate 180 bytes/min

25. 1. Deep-sleep 23 mA Wi-Fi
Energy consumed
in 42 mA 3G
sleep-mode. Power-off 180 uA

26. ower-off configuration. radio in the power-off configuration.
g, PPP dial-up connection is set in section C, followed 55
ion D by the actual transfer, and a short section E in 50
the connection is terminated. Figure 8 presents the same 45
when the Wi-Fi card is used. 40
Radio-on time [minutes]
etime Estimation 35
30
can now estimate the lifetime of a battery-operated
y for each of the four configurations (deep-sleep/power- 25
de with Wi-Fi/3G radio). 20
do so, we consider a WSN of N = 25 motes, each with 15
e capacity of M bytes. Each mote generates measure-
2.
10
at a rate of B bytes/min. We assume that there are no 5
y requirements in delivering the measurements to the
Energy consumed
nd server so the only constraint is to offload the data
0
0 50 100 150 200 250 300 350 400 450 500 550
the motes overflow their local storage. Therefore, the Data-size [kilobytes]
to retrieve
y must retrieve the motes’ measurements after each of Fig. 9. Total time the gateway is active downloading data as a function of
KoalatheUltra-Low Power Data RetrievalwithWireless Sensor Networks
- per-mote download size for a network in 25 nodes. These download
otes has collected α · M , (α ≤ 1) bytes of data. For times were reported in [25].
WSN data.
le, when α = 0.25, the gateway must collect 256 KB
Razvan Musaloiu-E, Chieh-Jan Mike Liang, Andreas Terzis, IPSN ‘08
a from each mote (6.25 MB in total), approximately
10 days when the data generation rate B = 18 bytes/min
proximately every day if B =180 bytes/min. Table IV data uploaded to the back-end server. Note that in the power-
es a list of all the model’s parameters. 102 mA Wi-Fi
off configuration this energy includes the cost of booting the
stimate the gateway’s expected lifetime, we must com- gateway (cf. Sec.IV-B).
s total energy consumption over time. This quantity is We combine the three factors to derive the results shown in
124 mA 3G
m of three factors: (1) the energy consumption while the Figures 10 and 11. These figures show the cumulative energy
y is sleeping (either in deep-sleep or power-off mode), consumption as a function of time in deep-sleep and power-off
consumption when the gateway is actively collecting modes for both long-haul radios, when the gateway downloads
(interface down)
ata, and (3) the consumption while the gateway uploads data every one and every ten days respectively. Based on
o the back-end server. We estimate each of these three these consumption rates, we estimate gateway lifetimes in two
next. scenarios. First when the gateway is powered by standard AA-
cell lithium batteries with a capacity of 3,000 mAh and second

27. 18
Wi-Fi - deep-sleep
3G - deep-sleep
16 Wi-Fi - power-off
3G - power-off
14
Energy Consumption (mAh)
12
10
3. 8
6
Energy consumed 4
to
2
0
upload data.
100KB 250KB 500KB 750KB 1M
Use a linear function of the
amount of data uploaded.

28. ti on
a
tim
Es
20000
Deep-sleep
Energy Consumption (mAh)
15000
Power-off
10000
5000
3G - deep-sleep
Wi-Fi - deep-sleep
3G - power-off
Wi-Fi - power-off
0
0 50 100 150 200 250 300
19,000 mAh Days

29. 2,000 days
1,860 days
1,500 days
300 days
1,071 days
300 days
1,000 days
225 days
150 days 137 days
500 days
75 days
31 days
16 days
18 days 34 days 0 days
0 days Deep-sleep Power-off
Deep-sleep Power-off
3G
18 bytes/min Wi-Fi 180 bytes/min

30. Alternatives

31. Hardware
Flashlite 186 RCM4500W RabbitCore Verdex XM4
JK microsystems Rabbit Semiconductors Gumstix
Mod2570 Wildfire 5282 Imote2
Nerburner Intec Automation CrossBow

32. Software
VxWorks

33. Software framework for WSN gateways.
Battery-operated mote gateway.
Investigate software and hardware alternatives.

34. Mike
Acknowledgments
LUYF Crew, Robert Szewczyk
National Science Foundation, JHU/APL, Microsoft Research, Moore Foundation, Seaver Institute

35. Thanks

36. Questions