Sensor Network Workshop Report

1
Sensor Network Workshop Report

• Yi Pan

2
Report Outline

• Outline and concentration of the workshop
• Related issues with my research
• Relevancies
• Ideas applicable
• Interesting topics in the workshop
• Research issues in sensor network

3
Workshop Outline

• Sensor network application domains
• Civilian applications
• Environmental monitoring
• Pollution, temperature, noise, moisture,
  biological motes sensor
• Industrial monitoring system
• Natural gas pipeline, nuclear plants, chemical
  emission monitoring
• Disaster response facility
• Sensors help to locate/map the disaster scene,
  find living people, etc.
• Seismic event monitoring
• Sensors help to monitor the structural failure of
  a building
• Sensors help to locate the center of an
  earthquake
• Intelligent buildings
• Sensors help to monitor/control human activities
  in a building, ie for security reason
• Others
• Military applications
• Battle field guarding system
• Sensors deployed to help soldiers detect hostile
  invasion
• UAV recon system
• Sensors deployed to help collect information for
  recon UAVs

4
Workshop Outline

• Techniques involved in sensor networks
• Sensor development
• Major issue is the hardware design
• Constraints on sensors
• Power supply not infinite, limited power
• Processing ability low processing power
• Communication bandwidth/distance low-bandwidth,
  short distance communication
• Heterogeneity of different types of sensors
• Sensors differ a lot from each other
• Size can range from meters to millimeters
• Image sensor for plasma collision detection is
  large
• Smartdust aims to produce 1 mm3 size autonomous
  sensors
• Power supply can also range a lot but yet can not
  assume infinite power supply
• Big sensors like Aibo platform is able to carry
  long-lasting battery
• Small sensors can only carry small battery or
  rely on solar cells
• Mobility
• AIBOs can move actively
• some may move passively. Ie. carried by wind,
  fuild flow, etc.
• Many others can not move. Ie. Smartdusts

5
Workshop Outline

• Techniques involved in sensor networks (Contd)
• Network/communication challenges
• Based on the features of sensor hardware, new
  constraints are put in networking and
  communication techniques
• Bandwidth
• To save power, low bandwidth is recommended
• Power
• Nodes need to frequently go into sleep mode to
  save power. ( extreme case 99 of time sleeping)
• Transmission range is also constraint
• Protocol implementation is limited to
  light-weight protocols to avoid heavy computation
• Topology construction
• Sensors may move or go to sleep. As a result,
  network topology changes
• Control the mobility and active mode of nodes to
  construct a suitable topology is needed
• Self-organization is required since manual
  configuration of sensor network is infeasible
• Location/mapping with physical world
• Sensor network is used to connect the virtual
  world monitored in the computer with the real
  physical world
• Location of sensors to the physical place is also
  needed
• Common requirements on network/communication
  protocol
• Based on different types of sensors, the level of
  constraints on network/communication may be quite
  different
• Common requirements will be

6
Workshop Outline

• Techniques involved in sensor networks (Contd)
• Application
• New challenges large volume of data
• Tera bits or Peta bits data will be available
  through extensive deployed sensor network in the
  future
• Image sensors for plasma collision detector
  creates data over 1TB per second
• Astronomers use telescope network over all the
  world expects TB data everyday
• Need to convert data ? information ? knowledge
  quickly
• Backend supercomputing support
• TeraGrid is one example of super computers to
  process the large volume of data
• Hierarchical data processing techniques
• Hierarchical data concentration units (DCU) are
  employed to convert the local data into local
  information and knowledge
• Security, privacy, and safety
• In application of sensor network, to provide
  enough security, privacy and safety for the data
  collected by the sensors is also one critical
  task
• Reports show that large volume of insensitive
  data can reveal something important
• Ie. extensive sampling of temperature changes in
  the bedroom can reveal your privacy

7
Workshop Outline

• Other issues
• Political issues
• Whether people want to live in a sensored world?
• Privacy? Abuse of sensor data can be very
  dangerous
• Ecological issues
• The sensors can be a new source of pollution
• How to regulate?
• Economic issues
• Whether to deploy sensor networks for emergencies
  only or for daily usage?
• A sensor network provides applications in both
  scenarios is preferable since the cost is low
• The performance of such a general sensor network
  may not be acceptable in emergencies
• But a sensor network working in normal situation
  can help to avoid or detect the emergencies
  earlier to reduce the loss

8
Relevance

• To my current project, the following scenario has
  a week relevance
• In military battle field guarding systems

3
4
2
1
Multi-hop packet routing(real-time
communications)
9
Relevance

• To maintain the real time communication in the
  previous scenario, mobility support is one major
  issue
• Questions
• In sensor network scenario, I am not sure whether
  the multi-path concept is still valid or not,
  since the relaying points need to be limited to
  save power
• Need to compare network layer real-time
  re-routing mobility support in this case since
  the major issue may be just maintaining the
  connectivity
• Depends on the type of data transmitted in this
  scenario
• Low-bandwidth data/text/audio vs high-bandwidth
  video

10
Relevance

• One paper in the research view of UCSD is related
  to my current project
• End-to-end differentiation of congestion and
  wireless loss
• Purpose
• Modify TFRC to perform better facing wireless
  loss
• Scheme
• Differentiate wireless loss and congestion loss
  by packet inter-arrival time
• The speculated packet inter-arrival time doesnt
  change a lot in wireless losses
• The speculated packet inter-arrival time changes
  a lot in congestion losses
• Treat wireless loss and congestion loss
  differently
• Relevance
• May be one reference rate control
• But it is concentrated on differentiation of
  wireless loss instead of mobility
• One comments on that paper is it doesnt take
  into consideration of wireless MAC
  re-transmission time. In wireless loss, the
  re-transmission can make the speculation scheme
  fail to differentiate congestion loss and
  wireless loss

11
Relevance

• Inspired idea from that paper
• I am thinking about using the averaged
  re-transmission time for each packets as an
  indicator of wireless loss vs congestion loss
• Intuitively, the average re-transmission time of
  each packets is an indicator of congestion level
  in the wireless single link
• A random loss typically will not cause large
  changes in averaged re-transmission time
• A congestion in wireless link will cause large
  changes in re-transmission time
• Receiver can detect this difference and let the
  sender to use this information for rate control
• Applicable in wireless last-hop networks with
  shared media in MAC
• Dedicated wireless link like CDMA will not have
  re-transmission scheme

12
Interesting Topics

• Smart Dust
• The research project currently is mainly
  concentrated on hardware development
• Intel has a demo with dynamic topology
  construction algorithm based on packet loss rate
• Passive, not moving sensors with a UAV as a
  roaming data collector
• Mini-OS on chip is developed

13
Research Issues

• The main research issue inspired by the workshop
  in sensor networks
• Dynamic network topology construction
• The network topology need to be controlled to
  reduce the power consumption
• The constructed network topology still needs to
  satisfy the communication purpose
• Enough coverage
• Efficient routing
• Etc.
• Self-organization is required because no human
  operation can configure large amount of sensors
  in the network
• Clustering, self-repairing mechanisms are needed
• Continuous connectivity need to be maintained
• Hierarchical sensor data collection/application
• Consider the heterogeneity of devices we need in
  the sensor network and the need to convert large
  volume of data into knowledge, hierarchical
  sensor data collection is needed
• Mobile agents working as DCUs can do local data
  collection and may do some local small scale data
  conversion
• Integration of global/regional knowledge need to
  be done by some powerful servers remotely
  accessible by the mobile agents
• Several levels of data concentration may apply in
  this system

14
Inspired Ideas

• Sensor network topology dynamics
• In constructing the network topology, several
  possible topics can be considered
• Case I Apply the power-law in random graph
  generator to construct the sensor network
  topology
• In random graph generator, the nodes degree has
  a power-law relationship with the number of nodes
  with the same degree
• It may be applied to create the network topology
  for a large scale sensor network
• Every node will choose a probability to be active
  based on the number of neighbors it hears. The
  more the neighbors, the less the probability. The
  probability is controlled by the power-law

15
Inspired Ideas

• Sensor network topology dynamics
• Case 2 Load-adaptive network topology
  construction
• In the workshop reports, one observation of the
  network topology dynamics is
• The central nodes in a matrix which often have
  more traffic pass through will have shorter
  lifetime given a limited battery supply
• Idea in choosing the sleep/active mode, we may
  need to activate more nodes in the area where the
  traffic is heavy to share the load
• Methods to achieve the goal (depending on the
  capability of sensors)
• Solution 1 make the activation probability of
  nodes in the heavy loaded area higher
• Solution 2 make nearby actively movable sensors
  concentrating to the heavy traffic
  (Wave-algorithm?)
• Load-adaptive routing in ad hoc
• One idea from my mind
• Consider the 802.11b and other wireless MAC
  protocols using retransmission to solve the MAC
  collision, the load-adaptive routing can be
  considered to use average re-transmission time as
  an congestion indicator for each MAC link. It may
  be more adaptive and accurate than the hop-number.

16
Inspired Ideas

• Sensor network topology dynamics
• In any of the above schemes, the
  self-organization should be considered
• Clustering may be one main answer to that
• Hierarchy/flat clusters? Not sure
• Mobility problems associated with the clustering
• Rotate the active mode among a group of nearby
  sensors with the same features may be to
  handoff the current routing/connectivity softly
  to allow the continuous communication
• If a group of mobile agents are used as DCUs for
  a group of sensors, connectivity from sensors to
  the DCUs may also need to be handoff softly

17
Inspired Ideas

• Hierarchical sensor data collection/application
• May well be applicable in disastrous and homeland
  security scenario
• Massively deployed sensors give simple data on
  spot and have limited transmission range
• Sensors have direct contact with first-hand
  reporters and survivors
• Emergency response team members or robots (AIBO
  platforms) carrying mobile devices with a little
  bit more powerful CPUs and communication
  capabilities are deployed into the scene
• Local data collection and dynamic network
  topology construction is applied to construct a
  integrated network
• Local data collection/conversion may be needed
• Different types of sensors may need different
  data collection/conversion agents
• The data collection/conversion agents can be
  bio-net entities migrating on the platform
  carried by all mobile devices
• One mobile devices may not be able to run all
  data agents the scene requires and thus may not
  have the required data agents in a particular
  scene
• The data agents located to a mobile devices may
  not be needed when the mobile device enters a new
  scene
• Composite services? May be needed
• More extensive computation and data analysis is
  shipped to servers connecting to the mobile
  agents to provide useful knowledge
• Quick response and guidance can be made based on
  the online knowledge acquired through the
  integrated sensor network

18
Other Ideas

• A cross-layer scheme for multi-hop wireless
  communication
• I consider wireless MAC re-transmission as one
  important indicator for the link status
• Heavy loaded/bad wireless links will have higher
  re-transmission in link level and thus produce
  un-desired features for end-to-end communication
• End-to-end transmission control faces new
  challenges considering that issue
• Large variation in round trip time
• Hop numbers no longer provides sufficient
  estimation on end-to-end delay
• One more issue channel capture/TCP unfairness in
  multi-hop wireless networks

19
Other Ideas

• Cross-layer schemes may be solutions
• Load-sensitive ad hoc routing
• If using the multiplication of link
  retransmission time with the number of links as a
  measurement for the delay on a path, we may be
  able to provide a route with less variation in
  round trip time
• Dynamic network topology construction
• With the dynamic network topology construction,
  we may be able to alleviate the load on the path
  by providing multiple paths in heavy loaded area
• Thus, without changing the routing algorithm, we
  can provide less variation in end-to-end delay

20
Other Ideas

• Cross-layer schemes may be solutions (Contd)
• Transport adaptive acknowledgement
• Considering the channel capture problem of TCP in
  multi-hop wireless networks, highly synchronized
  acknowledgement messages using the same wireless
  channel is the main problem
• Professor Mario Gerlas group has investigated
  MAC modification to alleviate the problem
• Can we use the retransmission time as a load
  indicator and modify the receivers acknowledge
  scheme to alleviate the channel capture problem?
• For example, by using an ACK window control
  adaptive to the single MAC links average
  retransmission time, we may be able to provide a
  fair competition of backward ACK traffic between
  two TCP connections
• This modification doesnt need to change the
  sender side. Tradeoff is a less sensitiveness of
  the senders congestion control because less ACKs
  are received.
• When the receiver has the wired connection, it
  simply sets the ACK window fully open and behaves
  normally as a TCP receiver end. So the scheme
  should be applicable in wired/wireless hybrid
  environments
• By discussion with Professor Magdas student, I
  noticed the current TFRC implementation requires
  high computation cost in reality.
• Observation shows cost for the rate control of
  TFRC is high and the accurate rate control
  feedback is only needed to control the flow
  dynamics when congestion happens
• Adaptive acknowledgement scheme may also help to
  alleviate the TFRC heavy computation cost problem
  by making the number of feedback control
  calculation adaptive to the congestion status