Battery Lifetime Estimation and Optimization for Underwater Sensor Networks

1
Battery Lifetime Estimation and Optimization for
Underwater Sensor Networks

• Raja Jurdak
• Cristina Vidiera Lopes
• Pierre Baldi

Presented by Stanley J Barr sbarr_at_cs.uml.edu
2
Introduction
3
Underwater Acoustic Comms

• Easy Deployment
• Better performance than RF technologies
• Successfully used in military apps for years
• Military uses have led to civilian interest
• Physical indicators (Salinity, pressure, temp,)
• Chem/Bio Indicators (bacteria, chem levels, )
• UCI wants to develop a UWSN for monitoring
  environmental indicators

4
Problem Statement
One of the major considerations for the
development of such a network is the power
consumption at each node. This work is motivated
by the practical need to estimate the battery
life of sensor nodes, which has implications on
the usefulness, topology and range of the
network.
5
Goal of this research

• Provide an estimation method for UWSN battery
  lifetime
• Propose topology dependent optimizations for
  power consumption
• Evaluate the benefits of the optimizations for a
  typical shallow UWSN

6
Network Battery Lifetime Estimation Method

• Introduction

7
Design issues in shallow UWSN

• Spectrum allocation
• Limited available acoustic spectrum
• Factors related to topology
• Inter-node distance
• Number of forwarding nodes
• Shallow water environment
• Distinct multi-path issues
• Noise factors have distinct patterns
• Examples Winds and Shipping Activities

Design Choices that address these challenges
affect battery lifetime of a network, which is
our metric.
8
Network Battery Lifetime Estimation Method

• Network Design Parameters

9
Illustrative Network Topology
10
Points about Figure One

• Illustrative network topology
• Multi hop
• Centralized topology
• Trees rooted at base station
• Data flow always towards the base station

11
Node design assumptions

• Xmit and rcv are main sources of power
  consumption
• Sensing and processing negligible
• Omni-directional hydrophones

12
Factor One Xmit frequence (f)

• Use FDM as a multiple access technique
• Pro - FDM Allows nodes to simultaneously xmit in
  a density sensor populated region
• Con - Signal loss is function of freq. and dist.,
  so higher freqs consume more power per dist.
  unit
• Maximum freq. (f) determines
• Worst case for battery lifetime
• Power consumption of the network

13
Factor Two Update Interval (R)

• Update interval is how often a sensor reports its
  collected data
• There are tradeoffs with increasing the update
  interval to produce fewer updates
• Pro Uses less power
• Con Means lower accuracy of sensed data
• For UCIs purpose of capturing environmental
  variations setting R 20 minutes provides
  sufficient resolution

14
Factor Three Inter-node Distance (d)

• We consider a multi-hop network
• Nodes forward messages through neighbors to reach
  base station
• Nodes only have to transmit one hop
• Pro Multi-hop extends range of network
• Con Increases power consumption on nodes that
  forward data for others
• Distance between nodes becomes important

15
Factor FourCluster Size (M)

• Forwarding packets is energy expensive
• We partition in the network in clusters
• Clustered nodes only forward packets for one
  another
• A cluster can be further broken into tiers
• Tier number denotes distance from Base Station
• Thus lower tiers required to forward more data
• Ms choice
• Depends on data sampling granularity
• Establishes a tradeoff between
• Power consumption for transmission of large
  distances
• Power overhead of forwarding data

16
A clustered and tiered UWSN
17
Network design parameters summary

• Factor One Xmit frequency (f)
• Factor Two Update Interval (R)
• Factor Three Inter-node Distance (d)
• Factor Four Cluster Size (M)

18
Underwater Acoustic Fundamentals
19
Passive Sonar Equation
SNRSL-TL-NLDI

• SNR Characterizes signal to noise ratio of an
  emitted underwater signal at receiver. For out
  application we consider a target of SNR 15 dB
  at the receiver.

All quantities are dB re mPa, where the reference
value 1 mPa 0.67 10-22 Watts/cm2
20
Passive Sonar Equation
SNRSL-TL-NLDI

• SL Source Level, the amount of sound radiated by
  a sound source. Its definition is the intensity
  of the sound at a distance of 1 meter from the
  source.

21
Passive Sonar Equation
SNRSL-TL-NLDI

• TL is the transmission loss.

22
Passive Sonar Equation
SNRSL-TL-NLDI

• NL is the noise level. Example factors include
  waves, shipping traffic, wind level, biological
  noise, seaquakes and volcanic activity. These
  are location dependent.
• For our application we assume NL 70 dB

23
Passive Sonar Equation
SNRSL-TL-NLDI

• DI Is the directivity index. As we are
  considering omni-directional hydrophones our DI
  0.

24
Passive Sonar Equation
SLTL85

• After incorporating our assumptions
• we are reduced to the above equation.

25
Transmission Loss

• The following formula provides TL for
  cylindrically spread signals.
• Inter-node distance (d) in meters and (a) dB per
  Km is the frequency dependent medium absorption
  constant.
• The above table uses empirically derived values
  for medium absorption in shallow seawater at
  temperatures between 4 and 20 degrees C. The
  transmission frequency (f) is in KHz.

26
Transmission Power

• SL Amount of sound radiated
• It Amount of sound power transmitted through
  a unit area
• Pt Transmitter power required to achieve It 1m
  from source at a water depth of (H) meters.
• We now have a method to determine transmission
  power requirements for inter-node comms between
  two nodes with distance (d), using frequency (f),
  to achieve our desired SNR of 15 dB at the
  receiver

27
Data Delivery Assumptions

• Size of data packet 1000 bit
• Each channel has 1 KHz bandwidth
• Bit rate for each node is 1 Kbit/sec
• Packet transmission is 1 second
• Pt is for a contention less environment
• Generic stop and wait MAC
• ACK size 200 bits
• Assume 10 packet loss each way
• Roundtrip success rate 81
• Packets must be sent 1/.81 or 1.23 to guarantee
  success
• Receive power 1/5 of transmit power

28
Data Delivery Equations

• The average power in watts needed to send a
  single frame Pframe is
• N data packets forwarded either way
• First two terms account for data packets
• Last two terms account for acks
• Bottleneck for network life occurs at tier 1
• In a grid topology Nmax (M/2)2
• In a chain topology Nmax M

29
Network Lifetime and Power Consumption (PCTR)

• PCTR Ratio of overall power consumption to
  throughput
• Its a power cost for transmitting bits across a
  network
• Every update each node in a cluster of size (M),
  sends its packets and forwards its neighbors
  packets
• Yielding the average above
• Note This is a tier independent average

30
Network Lifetime and Power consumption
Total active time at tier 1 in seconds
Assume we have 3 9 volt with 1.2 Amp batteries
at each node in VAhour
The total active time of a transceiver is the
ratio of Total energy to power consumed in 1
frame, in hours.
Transceiver is only active for a portion of the
update Interval R in seconds and the lifetime is
expressed in days. Note Earlier R was
expressed in minutes.
31
Topology-Dependent Optimizations

• Up till now all equations have been tier
  independent
• Now we consider two tier dependent enhancements
• Tier dependent frequency assignment
• TL increases at higher frequencies
• Implies nodes using higher frequencies must use
  more power
• Implies lower numbered tiers should have lower
  frequencies to save power
• Tier dependent distance assignment
• TL increases at increased inter-node distances
• Increasing inter-node distance with each tier can
  reduce power loads at lower tiers

32
Required ModificationsFor Tier dependent
Optimizations

• Goal one is to reduce the overall power
  consumption per frame in the network
• Replace all occurrences of N at tier i with N M
  i 1
• At each tier we compute Pframe, Pmax, and Ttotal
• We also modify PCTR to reflect the distinction
  among tiers
• Goal Two of Tier dependent assignments is to move
  the bottleneck away from the base station
• By using individual tier values for Pframe and
  Ttotal we shift the bottleneck to tier i

33
Case Study

• Deployment has max depth (H) of 10m
• Inter-node distance (d) 50m to 1km
• Update period (R) 20 minutes
• Required maintenance interval 100 days

34
PCTRas a function of d and f
Note 250m denotes a visible change in behavior.
35
Battery Lifetime as a function of d and f
Note distance and frequency affect battery life.
36
Tier Dependent Assignments

• CFB - Constant Frequency Band
• Tiers 1-50 get frequencies from 1-50KHz
• Tiers 50 and higher get 50KHz
• VFB - Variable Frequency Bands
• Tiers 1-50 allocated as in CFB for clusters where
  M lt 51
• When M gt 50 Divide spectrum into M/ 50
• CID - Constant inter-node distance
• Inter-node distance of tier i is 50i for i lt 20
  and 1km otherwise
• VID - Variable inter-node distance
• Tiers 1-20 allocated as in CID for clusters where
  M lt 20
• Increase inter-node distance by 1/ M Km

37
Bottlenecks Tiers vs. Cluster Size
38
Power Consumption vs. Cluster Size
39
Battery Lifetime vs. Cluster Size
40
Grid Topology

• Same equations different Nmax and N
• In a SxS grid Nmax S and N S(S1)/2
• Grid topology requires uniform inter-node
  distances, so we only consider only CFB and VFB
  in our tier dependent and independent models.

41
PCTR vs. Cluster Size
Note Maximas occur at perfect squares.
42
Battery Life vs. Cluster Size
Note In basis method life drops sharply when
adding a new node creates a new tier. Battery
life is in years which is much better than chain
topology.
43
Maximum Range Alternatives
44
Method Comparison For Chain Topologies

• Tier dependent inter-node methods
• For situations which need far more fine grained
  monitoring local to the base station
• VID - Provides longest chain life
• Deploying larger nets is expensive
• based on density and amount of sensors
• CID is only better that tier independent
• But looser requirements on node placement
• Less nodes required then VID
• Tier dependent frequency methods
• Good when you require uniform granularity in
  sampling
• CFB Better than basic method with little added
  complexity
• VDB - Achieves longest ranges,
• but adds signal processing as it requires same
  channel rate using smaller frequency bandwidth