S. Coleri, A. Puri and P. Varaiya

1
Power Efficient System for Sensor Networks

• S. Coleri, A. Puri and P. Varaiya
• UC Berkeley
• Eighth IEEE International Symposium on Computers
  and Communications (ISCC03)
• PEDS Seminar
• Presenter Bob Kinicki

2
Outline

• Introduction to Wireless Sensor Networks
• Previous Work
• The Berkeley System
• Simulation Results
• Conclusions

3
Wireless Sensor Networks

• Sensors small devices with low-power
  transmissions and energy limitations (e.g.,
  battery lifetime concerns)
• The main distinction from traditional wireless
  networks is that the data traffic originates at
  the sensor node and is sent upstream towards
  the access point (AP) that collects the data.
• While the nature of data collection at the sensor
  is likely to be event driven, for robustness, the
  generation of sensor packets should be periodic
  if possible.

4
Power Consumption Components

• Primary source of power consumption is the radio
  transmitting, receiving and listening.
• Key tenet of this paper
• Sensor nodes must only be awake to receive
  packets destined to themselves or to transmit. At
  all other times, the sensors need to sleep to
  conserve power.

5
The Goal

• A system for sensor networks that
• achieves power efficiency in a robust
• and adaptive manner.

6
Previous Work Contention Based

• A separate wake-up radio (channel) to power up
  and down the normal channel
• The key idea is that the wake-up listen mode is
  ultra-low power.
• Uses a wake-up beacon.
• S-MAC (sensor MAC)
• Uses RTS/CTS such that interfering node goes to
  sleep upon overhearing either an RTS or CTS.
• Problems Here??

7
Previous Work Contention Based

• STEM (Sparse Topology and Energy Management)
  trades energy savings for latency through
  listen/sleep modes.
• Uses a separate paging channel.
• Sending node must first poll the target node by
  sending a wake-up message over the paging
  channel.
• Target receiving node would then turn on primary
  radio channel to receive regular transmission.
• This scheme prevents collisions between polling
  and data transmissions.
• This scheme is effective only for sensor
  scenarios where the sensor spends most of its
  time waiting for events to happen!

8
Previous Work TDMA Based

• TDMA schemes eliminate overhearing, collisions
  and idle listening.
• However, proposed TDMA schemes require dealing
  with communication clusters.
• One solution a high power AP that can
  accomplish all the TDMA scheduling.

9
The Berkeley System
AP
AP
AP
sensor
sensor
Multiple hop tree topology
sensor
sensor
sensor
sensor
sensor
sensor
sensor
sensor
10
The Berkeley System
AP
AP
AP
sensor
sensor
AP range
sensor
sensor
sensor
sensor
Sensor range
sensor
sensor
sensor
sensor
11
Sensor Hardware

• UCB Mica motes
• Support adjusting transmission power
• Sensors run on AA batteries that can supply
  2200mAh at 3V.

12
Three Transmission Ranges

• Long used for coordination AP frames and
  reaches all the sensors in one hop.
• Short used to transmit data packets from sensor
  nodes to the AP.
• Key idea choose the lowest possible range that
  still assures network connectivity.
• Medium used in tree construction to learn the
  interferers of each sensor node, namely, nodes
  with signal strength too weak to be decoded but
  strong enough to interfere.

13
Three Communication Phases

• Topology Learning Phase
• Topology Collecting Phase
• Scheduling Phase

14
Topology Learning Phase

• During this phase each node identifies
  interferers, neighbors and parent.
• AP transmits the topology learning packet
• current time, incoming packet time over
  longest range in one hop to all sensor nodes the
  AP will coordinate.
• AP floods the tree construction packet hop
  count over the medium range.

15
Topology Learning Phase

• Random access scheme is used with an interfering
  threshold to decide on neighbors, interferers and
  the parent on the smallest hop path to the AP.

16
Topology Collection Phase

• By the end of this phase, the AP has received
  complete topology information.
• AP transmits the topology collection packet
  current time, incoming packet time over the
  longest range at the announced time.
• Each node transmits its topology packet parent,
  neighbors, interferers. Vague scheme used is
  CSMA with implicit ACK.

17
Scheduling Phase

• Sensor node transmissions are explicitly
  scheduled by AP based on complete topology
  information.
• The AP announces the TDMA schedule by sending the
  time-slotted scheduling packet current time,
  incoming packet time by broadcasting over the
  longest range.
• Scheduling algorithm can vary.
• Using a threshold for percentage of successfully
  scheduled sensor nodes, the idea is to keep the
  system in the scheduling phase until the
  percentage falls below the threshold where upon
  the system will switch to the learning phase.
• High performance comes when the ratio of
  scheduling phases to the other two phases is high.

18
Simulations

• Used TOSSIM, a TinyOS simulator.
• Nodes are randomly distributed in circular area.
• Transmission rate 50 kbps
• 10 Monte Carlo Simulations
• Best possible random access result reached by
  adjusting CSMA listening window sizes and the
  backoff settings.

19
Power Consumption Comparisons

• Assumptions
• Clock interrupt every millisecond (1ms.)
• Sensor sampled once per packet generation period
  (30 seconds).

20
Random Access versus TDMA
Battery Lifetimes Random access 10
days Berkeley TDMA scheme 2 years
21
Random Access versus TDMA

• Listening takes power!
• Random access yields
• retransmissions.
• Overhearing affects
• reception power.

22
Varying Sensor Sampling Rates
The slope is less than one due to the high power
cost associated with clock interrupts.
23
Redundant Sensor Nodes
The important assumption with redundant sensor
nodes and TDMA is that sharing of the scheduled
slot allows redundant not-scheduled nodes to
reduce their clocking rate and then increase it
back during the last part of the packet
generation period.
24
Conclusions

• IF Access Point is not power limited then
  asymmetric transmission power between AP and
  sensor nodes is a good idea.
• Base on ONLY simulations, the Berkeley System
  with TDMA consumes much less power compared to
  random access.
• Redundant sensor groups also has potential to
  save sensor power in the Berkeley System.