Data Link Layer Architecture for Wireless Sensor Networks

1
Data Link Layer Architecture for Wireless Sensor
Networks

• Charlie Zhong
• Qualifying Exam
• October 8, 2001

2
My Contributions

• A systematic approach to design low energy data
  link layer (DLL) for wireless sensor networks
  based on well defined energy metrics and
  tradeoffs between subsystems

3
Outline

• Background
• A systematic approach to achieve lowest overall
  power consumption
• Future work

4
Applications for Sensor Networks
Energy consumption control
Disaster mitigation
Traffic management
5
Wireless Sensor Networks Characteristics

• Large number of nodes
• Low mobility
• Low data rate
• QoS is primarily about coverage, not delay or
  guarantee of delivery
• Need most easy maintenance, long operation time
  and minimal human intervention
• Power is the key!

6
A Multi-hop Network
Controller
Sensors
Actuators
7
Data Link Layer Functions

• Transfers data between network and physical
  layers
• Maintains neighborhood info
• Power control, error control and access control
• Computes location

Controller
Sensors
Actuators
8
Data Link Layer Requirements

• Supports required functions
• Communications with required reliability
• Location as part of information
• Power efficient
• Distributed
• Requires no global synchronization
• Scalable
• Robust
• Easy setup and maintenance

9
Challenges

• Multiple metrics to consider (e.g.TX power,
  reliability and connectivity)
• Contradictory requirements and goals of different
  subsystems
• No methodology in place to compare different
  designs
• Need a design method and analysis

10
Existing Work

• There is a lot of effort in low power DLL
  designs. Some examples are
• Media Access Control (MAC) only
• 802.11 power management (not for ad-hoc net.)
• UCLA distributed scheduling
• Joint optimization of two subsystems
• GTE topology control
• Tu-Berlin combined tuning of RF power and MAC

11
Their problems

• They just optimize power consumption for a part
  of a system (e.g. MAC only)
• Their design methodologies are in ad-hoc fashion
  when a subsystem is optimized, the other
  subsystems become sub optimal
• Examples of methodologies analytical (simplified
  MATLAB models) or simulative (NS, OPNET and SDL)
• They offer no quantitative comparison

12
A Systematic Approach

• Identify the goal for optimization
• Perform functional break down
• Understand the complicated inter-dependency
  between the design of different subsystems
• Quantify design metrics, know the tradeoffs
• Compare different algorithms
• Predict the direction for improvements

Lowest overall power consumption!
13
Advantages of This Approach

• Considers the impact of all metrics
• Creates an architecture where subsystems
  cooperate
• Provides quantitative comparison
• Achieves joint optimum, rather than individual
  optimum in power consumption

14
A Systematic Approach

• Identify the goal for optimization
• Perform functional break down
• Understand the complicated inter-dependency
  between the design of different subsystems
• Quantify design metrics, know the tradeoffs
• Compare different algorithms
• Predict the direction for improvements

Lowest overall power consumption!
15
Optimization Metric for Entire Sensor Network

• Cost local battery source
• Value desired functions
• Goal the time the desired functions can be
  maintained should be as long as possible for
  fixed energy cost

Network Life
16
Network Life

• Definition the time network stays
• functioning for given power supply

Power consumption distribution
Application
Network Life
Redundancy
Topology/density
Network life is measurable
17
How to Maximize Network Life an Example
Application

• Application knows how important an end-end
  session is
• Network specifies required number of neighbors
• Data link sets transmit power level based on
  required number of neighbors and link-level
  reliability

Priority
Reliability
Transport
Link-level reliability
Network
Link metrics
NBs
Data Link
TX power
Battery drain
Physical
18
Parameterized Protocol Stacks

• Partition a task into manageable pieces
• Configurable interfaces between layers and
  between subsystems in DLL
• Close-loop interactions for adaptive control
• Layers/subsystems cooperate to maximize network
  life

19
A Systematic Approach

• Identify the goal for optimization
• Perform functional break down
• Understand the complicated inter-dependency
  between the design of different subsystems
• Quantify design metrics, know the tradeoffs
• Compare different algorithms
• Predict the direction for improvements

Lowest overall power consumption!
20
DLL Functional Break Down

• Iterative process
• Clean interface

Unified Modeling Language
21
Example Neighbor List Subsystem

• It creates and maintains neighbor list
• Use case diagram shows its relationship with
  others

22
A Systematic Approach

• Identify the goal for optimization
• Perform functional break down
• Understand the complicated inter-dependency
  between the design of different subsystems
• Quantify design metrics, know the tradeoffs
• Compare different algorithms
• Predict the direction for improvements

Lowest overall power consumption!
23
Data Link Layer Tradeoffs
Reliability
Redundancy
Network
Transport
Connectivity
Traffic density
Link-level reliability
NBs
BER
NBs
interferers
Data link data rate
Collision rate
Power Control
MAC
Interference
TX power
channels
Data rate
Physical layer
24
A Systematic Approach

• Identify the goal for optimization
• Perform functional break down
• Understand the complicated inter-dependency
  between the design of different subsystems
• Quantify design metrics, know the tradeoffs
• Compare different algorithms
• Predict the direction for improvements

Use power control subsystem as an example
25
Power Control Subsystem What it does

• Controls the transmit power
• Adaptively controls topology for desired
  connectivity
• Compensates topology changes incurred by mobility
  and dead nodes
• Controls a nodes neighborhood

26
Interactions with Other Subsystems
Collisions
Load balancing
Connectivity
Spatial reuse
Retransmissions
Performance degradation
Battery drain
Error performance
27
Design Metrics
Simplified model
BER threshold
Number of neighbors
Modulation
Collision rate
Power Control
Coding
Interference
Transmit power PT
PR received power at radius r n path loss
exponent r Radius d0 close-in reference
distance CReceiver sensitivity p BER
threshold D node density d number of
neighbors pkt packet error rate N max of
retransmissions M of bits/packet Reliability
prob. of packet loss after N retransmissions
pktN1
28
Relationship between Radius and Number of
Neighbors
OMNET simulation 125 nodes, randomly placed in
3000X3000X3000 space Prediction model
adjustment coefficient
29
Impact of MAC, Modulation and Coding

• Assumptions
• Channel assignment ensures every interferer is
  using a different channel
• There is no interference between channels
• BPSK modulation and no coding

• Tradeoffs
• For fixed C, r increases with PT
• When number of interferers increases, it is
  harder for MAC to zeroes out interference so that
  C can be constant
• Adaptively switching to a better modulation or
  coding scheme can increases r without changing PT

30
Relationship between BER Threshold and EPT
p BER threshold (there is a link between two
nodes only BER is lower than p) pkt packet
error rate of neighbors at coverage boundary M
of bits/packet N max of retransmissions PR
received power at radius r Reliability prob. of
packet loss after N retransmissions
pktN1 Assumptions BPSK modulation, no coding,
BSC, interference zeroed out by MAC BER threshold
10-3 ---gt packet loss rate 0.45
(N3M300)---gt end-to-end loss 4.5 (10
hops) Optimum BER threshold is the same for all
neighbors inside coverage and it doesnt change
with r
31
Effect of N
Optimum BER threshold doesnt change with N
EPT changes little for small p Reliability
changes a lot with N (may not be true for bursty
channel) N1 to 12
32
A Summary of Previous Results

• Average radius changes with number of neighbors
  as predicted by
• For given BER threshold, local topology is
  controlled by adjusting link BER. There are two
  ways to adjust BER
• Increase PT it is harder for channel assignment
  to control collisions and interference
• Use adaptive error control increases number of
  neighbors without increasing number of
  interferers

33
A Summary of Previous Results (2)

• For given link-level reliability, there exists
  an optimum BER threshold, which doesnt change
  with radius, distance or maximum number of
  retransmissions
• These results can be used to do joint
  optimization with other subsystems

34
A Systematic Approach

• Identify the goal for optimization
• Perform functional break down
• Understand the complicated inter-dependency
  between the design of different subsystems
• Quantify design metrics, know the tradeoffs
• Compare different algorithms
• Predict the direction for improvements

Lowest overall power consumption!
35
Future Work

• Define parameterized interface for other
  subsystems in DLL, so a complete architecture for
  DLL will be in place
• Evaluate existing algorithms and specifies the
  direction for improvement
• Support the above using a combination of
  analysis, simulation and empirical approach
• Refine design methodology

36
Milestones

• Low power MAC algorithms
  October 1999
• Data link layer specification
• Research proposal
• Analysis/Simulation
• Qualifying exam October 8, 2001
• Implementation of DLL on test bed
  December, 2001
• DLL Architecture in place October, 2002
• Tradeoffs and design optimizations
  December, 2002
• Functional Pico III node DLL May, 2003
• Graduation May, 2003

37
A Systematic Approach

• Identify the goal for optimization
• Perform functional break down
• Understand the complicated inter-dependency
  between the design of different subsystems
• Quantify design metrics, know the tradeoffs
• Compare different algorithms
• Predict the direction for improvements

Lowest overall power consumption!
38
Importance of My Research

• Creates a systematic way to approach best power
  efficiency
• Provides a guide for algorithm developers
  initiates new algorithms (e.g. adaptive error
  control)
• Specifies design parameters (e.g. optimum BER
  threshold)
• Offers a methodology that help other layers or
  systems to achieve lowest overall power
  consumption

39
Backup Slides
40
Sensor Network Requirements

• Desired performance (e.g. how good the
  environment control is)
• Long operation time
• Easy setup
• Easy maintenance/diagnostics
• Low cost
• Security
• Scalability, size etc.

41
System Operation
Initialization
42
Automatic mapping to implementation
Virtual Component Co-design
43
System Operation (2)
Maintenance
44
System Operation (3)
Data Communication
45
Data communications
46
Metrics Proposed

• Power on time
• Time spent on utilizing TX and RX
• Total number of correctly transmitted packets
  during the lifetime of battery
• Signal energy per successfully transmitted bits

47
What is power?

• Communications
• Value comes from information bits, not from OH
  bits, acks, handshakes to setup
• Values comes from the transfer from source to
  destination, not every hop
• Processing
• Value comes from how you use information
• Encoding/compression/redundancy

48
How to Optimize for Network Life

• Adaptive topology control
• Power efficient operation
• Load balancing

49
Backup Answers

• End-to-end (L hops) packet loss rate

• How to let new neighbors know about new error
  code in adaptive error control?
• Increase PT to reach them the first time. Go
  back to original PT when all neighbors know about
  new error code