Lizhi Charlie Zhong

1
An Ultra-low Power and Distributed Access
Protocol for Broadband Wireless Networks

• Lizhi Charlie Zhong
• May 9, 2001

2
Sensor Networks

• Tens or hundreds of sensors
• Each measures the temperature, light intensity,
  humidity, noise level, airflow speed etc. in its
  locality
• Self organized into a network
• Cooperative processing

3
Applications for Sensor Networks

• Microclimate in public place
• Smart home, toys society
• Battlefield
• Better net surfing experience

• Scientific research
• Home nurse care, inpatient services
• Interactive museums
• Games

Sensor anything that has certain knowledge of
its local environment
4
Sensor Network Requirements

• Low power and/or self powered
• Low cost, small size
• Wireless ad-hoc
• Robust,no single point of failure
• Can be setup on the fly
• Media access control (MAC)
• Optimized for power
• Distributed
• Simple and robust
• No synchronization required

5
Pico Radio Project

• Power consumption 100?w
• Cost less than 50 cents
• Size about 1 cm3
• Strategies
• Vertical optimization
• Power is not only optimized for a particular
  layer, but for all layers
• Horizontal optimization
• Power is not only optimized for a particular
  node, but for the entire network

6
Simple MAC

• Carrier sense multiple access (CSMA)
• Knowledge about network is not needed
• Synchronization not required
• No dedicated control channel
• No overhead in handshakes
• Channel assignment is not needed
• Quickly adapt to network changes

• Multi-channel
• Reduce collision
• Increase throughput
• Reduce delay
• No exposed terminal problem, alleviate hidden
  terminal problem
• Sleep mode
• Wakeup radio full duty cycle at 1?W
• With destination ID and channel modulated

Traffic density
Less than one user per channel on average
7
Multi-channel CSMA

• Simple procedure
• Randomly pick a channel
• If it is idle, use it
• If it is busy, go to step 1) and repeat on the
  remaining channels until all channels have been
  checked
• If all channels are busy, set a random timer for
  each of them
• Use the channel whose timer expires first and
  clear all other timers
• Channels can be code or sub carrier

8
Advantages of Our Access Protocol

• Power is optimized in every aspect
• Supports sleep mode
• Fully distributed
• Doesnt require synchronization, global or local
• Simple and robust
• No dedicated control channel
• Very low delay
• No need to coordinate broadcast and scheduled
  unicasts
• Applicable to any broadband wireless network

9
Ways to Deal With Redundancy

• Temporal and spatial correlation between sensor
  data are very high
• Difference between samples has smaller range, so
  it can be represented by fewer bits
• Joint decoding using previous samples
• Distributed source coding
• Source compression
• Trade redundancy for reliability

10
Our Solution Exploiting Redundancy

• Obtain the highly correlated data
• Destination itself monitor
• Select based on geographic range and time
• Get the data from neighbors
• Slowly increase the network span
• Estimate the original data from above
• Eliminate acknowledgment end-to-end error
  detection or periodic checking triggers
  estimation
• Save at least 304 bits for every data packet!

11
Simple Data Estimation

• Last sample of the same sensor
• Moving average of the same sensors data
• The best linear estimate using only the same
  sensors data
• The best linear estimate using the data from the
  same sensor and its neighbors
• Use different estimator based on data types

12
Pico Node
13
Simulation Setup
14
Light Intensity Vs. Time
Sampled every second, W100 P10
15
Temperature Vs. Time
Sampled every second, W100 P10
16
Comparison of Four Different Estimators
Light intensity data sampled at every 10
seconds,W250 P2
17
Instantaneous Error Rate of Different Estimators
Light intensity data sampled at every 10 seconds,
W250 P2
18
Mean Square Error Vs. Estimator Order
Light intensity data sampled at every 10 seconds,
W250
19
Mean Square Error Vs. Window Size
Light intensity data sampled at every 10 seconds,
P2
20
Power Saving Techniques

• Keep the design simple
• Turn the radio off when it is not doing anything
• Trade bandwidth efficiency for power efficiency
• Exploit redundancy