1
Efficient Error Recovery Using Network Coding in
Underwater Sensor Networks
Zheng Guo, Bing Wang, and Jun-Hong Cui
Presented By Daniel Gyllstrom
2
Goal

• Monitor Underwater Environments

3
A Problem

• Communication error prone
• Characteristic of communication lines
• Node Mobility
• Long propagation delays

4
Goal

• Reduce communication errors under energy
  consumption constraints

5
Problem Setting
6
Existing Approaches

• Error Correction Techniques
• Autonomic Repeat reQuest (ARQ)
• Timeouts
• ACKs
• What TCP does
• --------------gt long delays

7
Existing Approaches

• Error Correction Techniques
• Forward Error Correction (FEC)
• Proactively add bits to avoid retransmission
• End to End
• Hop by Hop
• --------------gt How much redundancy?

8
Primer Multipath Routing
path 1
destination
source
path 2
9
Primer Network Coding
A 1001 B 0110 , AB 1111
AB 1111 B 0110
, (AB) B A 1001
10
Primer Network Coding
relay 1
Y1 a1X1 a2X2 Y2 ß1X1 ß2X2
X1, X2
Y1, Y2
sink
source
Y3 a3X1 a4X2 Y4 ß3X1 ß4X2
X1, X2
Y3, Y4
relay 2
11
Why Network Coding?

• H2O sensor nodes have sufficient computational
  power
• Broadcast nature of communication gt multipath
  makes sense
• Paper shows good source of error recovery

12
Network Coding (Source)

• Packets divided into groups
• K packets X1, X2 XK
• Linearly combine the K packets --gt K packets
  (where K K)
• Y1, Y2 XK where and
  gij are coefficients
• selected randomly from finite field
• (gi1, , giK) encoding vector carried w/
  packet
• notice K K --gt redundancy

13
Network Coding (Relay)

• Buffer incoming packets for some period
• Linearly combine packets to form several groups

14
Network Coding (Relay)

• Consider relay r
• Receive packets X1, X2 XM where (fi1, ,
  fiK) is encoding vector for Xi
• Outgoing Packets M total where M min(M,K)
• Y1, Y2 YM where and hij
  picked randomly from field (gi1, , giK)
• gij is linear combo of (incoming encoding vector
  - f) (random finite field - h)

15
Network Coding (Sink)

• Receive linearly independent encoding vectors
• Solve linear equations to recover original packets

16
Adjustments

• Need certain of nodes in relay set. How to
  ensure this?
• Adjust transmission range to include correct of
  downstream nodes

17
Adjustments

• How much redundancy?
• Less downstream neighbors ? more redundancy
• More downstream neighbors ? less redundancy

18
Metrics of Interest

• R successful delivery ratio
• T normalized energy consumption
• T (Total transmissions)/R

19
Analysis

• Purpose
• Compare algorithms - later
• how to set redundancy parameter adjust
  transmission range
• Assumptions
• Relay sets do not intersect
• Node receives packets from all nodes in prev.
  relay set
• Discard packets from same relay set

20
Analysis

• Results
• K generation size 3
• K packets output 5
• N relay set size 3

21
Simulation

• Setup

1 km
1 km
sink
1 km
source
22
Simulation

• Setup (contd)
• Vector Based Forwarding
• Routing pipe from source to sink
• Nodes inside pipe forward packets
• Packet 50 bytes
• K generation size 3
• K src output this many packets 5

23
Simulation

• Node distribution
• Grid random deployment
• Uniform random deployment

24
Simulation Grid Deployment

• 200m x 200m x 200m grids
• 2 nodes per grid
• Set transmission power to 300m pipe radius
  (150m) to cover 3 to 4 nodes

25
Simulation Grid Deployment
26
Simulation Grid Deployment
27
Simulation Grid Deployment
28
Simulation Grid Deployment
29
Simulation Uniform Rand
Vary transmission range (100m 400m) radius
150m
30
Simulation Uniform Rand
Vary K transmission range 300m
31
Discussion

• Is this easier to configure than FEC?
• FEC can use ACKS to determine loss rate
• Can redundancy and transmission range be adjusted
  on the fly?
• Algorithm highly dependent on topology, is this
  ok?
• How to deploy grid network?
• Why not include feedback to configure network
  coding