Bluetooth Based Wireless Sensor Networks

Soumya Ranjan Jena (2224100031)

Master in Computer Application ,School of Computer Science
Odisha University of Technology and Research , Bhubaneswar, 751006 ,Odisha
Abstract
Wireless sensor networks of small devices equipped with sensors, microprocessor and
wireless communication interfaces – are a technology that has gained a lot of interest lately.
The broad spectrum of new and interesting applications, ranging from personal health-care to
environmental monitoring and military applications is proposed for such networks. Various
wireless technologies, like simple RF, Bluetooth, UWB or infrared might be used for
communication between sensors. In this paper the main principles, applications and issues of
Bluetooth based wireless sensor networks, as well as an implementation of a simple
Bluetooth based sensor network are described. The main problems experienced during the
implementation and applied solutions are presented.

Keyword
Bluetooth, Wireless Sensor Networks, IoT, Communication, Data Collection, Applications.

1. Introduction
In [1], a vision, called ubiquitous computing, of the world where humans and computers were
seamlessly united is described. The essence of the vision was the creation of environments
saturated with computing and communication in an unobtrusive way. Recently, WWRF
(Wireless World Research Forum) and ISTAG (IST Advisory Group) released their visions
of the future communication networks. Both institutions envisage a vast number of various
intelligent devices, embedded in the environment, sensing, monitoring and actuating the
physical world, communicating with each other and humans. Today, we are witnessing a
rapid proliferation of various wireless devices with considerable computing power and fairly
small size. These devices are still not, to paraphrase Marc Weiser, indistinguishable from the
fabric of everyday life, but are becoming a part of that fabric (Bluetooth headset built as a
part of jeweler for example). Various sensors are already in a broad use today as part of
different devices temperature sensors in home or car heating system, smoke alarms, etc.) or
as standalone devices connected to a network, usually to monitor industrial processes,
equipment or installations. The advancements in MEMS (micro-electrical-mechanical
systems) technology, wireless communications and electric components have enabled
development of small, low-power and low-cost devices, called smart sensor nodes, capable of
performing various sensing tasks, processing data and communicating over wireless
connections. Such devices, when organized into a network, present a powerful platform that
can be used in many interesting applications, like health monitoring, security systems,
detection of chemical agents in air and water etc. During the last few years significant
research efforts were focused on development of required hardware resources and efficient
communication protocols to enable networking and collaboration of smart sensor nodes. In
the next section the main principles of wireless sensor networks are given and research issues
are explained. Section 3 presents Bluetooth issues related to its use in sensor networks. An
implementation of a Bluetooth based sensor network is described in section 4 along with
some of the implementation issues and solutions. Section 5 concludes the paper.

2. Wireless sensor networks

Wireless sensor networks comprise number of small devices equipped with a sensing unit,
microprocessor, and wireless communication interface and power source. In contrast to the
traditional sensor networks that are carefully planned and deployed to the predetermined
positions, wireless sensor networks can be deployed in an ad-hoc manner. Of course, such
deployment requires adequate communication protocols that are able to organize the network
automatically, without the need for human intervention. Beside self-organization capability,
another important feature of wireless sensor networks is collaboration of network nodes
during the task execution. In contrast to the traditional sensor networks where all sensor data
is gathered at a server and then analyzed and fused, data processing and fusion is now
performed by smart nodes themselves. Each node processes raw measurement data in order to
decrease amount of data sent over wireless links and forwards only relevant parts to nodes
responsible for data fusion. Data-centric nature of the network is yet another specific
characteristic of wireless sensor networks. As deployment of smart sensor nodes is not
planned in advance and positions of nodes in the field are not determined, it could happen
that some sensor nodes end in such positions that they either cannot perform required
measurement or the error probability is high. That is why a redundant number of smart nodes
observing the same phenomenon is deployed in the field. These nodes then communicate,
collaborate and share data, thus ensuring better results (each sensor has its own view of the
phenomenon – when these views are combined a better picture of the phenomenon is
obtained). Having this in mind, it is more reasonable for a user to send a data request to all
sensors monitoring the phenomenon than to send it to one specific sensor node. Using a
multicast routing protocol to send messages to all relevant nodes would require unique
addressing scheme in the network. However, due to the sheer number of sensors and user
requirements (user needs information about the phenomenon, does not need information
about the phenomenon from a particular sensor), data-centric approach is used where sensors
are designated using description of data they can provide instead of using unique IDs.
Messages are directed to nodes using routing protocols that can find the route based on the
data description contained in the message. Power efficiency is one of the main requirements
for all protocols and algorithms used in these networks. As power resources of each node are
limited and required lifetime for many scenarios is measured in months and even years, it is
of paramount importance to design system in such a way to ensure power savings whenever
possible.

2.1. Querying and tasking

From the user point of view, querying and tasking are two main services provided by wireless
sensor networks. Queries are used when user requires only the current value of the observed
phenomenon. As wireless sensor networks are data-centric networks, the user does not query
a specific node for the information it might provide, but defines data (type, location,
accuracy, time, etc.) he/she is interested in and requests it from all nodes that can provide the
answer. For example, user can look for “temperature in the north region of the observed area”
or for “location of all sensors where chemical agents are present and their level is above the
threshold”. Tasking is a more complex operation and is used when a phenomenon has to be
observed over a longer period of time. For example, a user can ask a sensor network to detect
a specific type of vehicle in the area and to monitor its movement. In order to execute the
task, different types of sensors have to collaborate: seismic to detect motion, video and audio
to detect type of vehicle etc. Information about the vehicle trajectory is forwarded to the user.
Both queries and tasks are injected to the network by the gateway which also collects replies
and forwards them to users.

2.2. Gateway functionality

Smart sensor nodes scattered in the field collect data and send it to users via gateway using
multiple hop routes (Figure 1).

Figure 1. A Wireless Sensor Network

The main functions of a gateway are the following:
 Communication with sensor network - Short-range wireless communication is used
(Bluetooth, UWB, RF, IR, etc.). Provides functions like discovery of smart sensor
nodes, generic methods for sending and receiving data to and from sensors, routing,
etc.;
 Gateway logic – Controls gateway interfaces and data flow to and from sensor
network. It provides an abstraction level with API that describes the existing sensors
and their characteristics; provides functions for uniform access to sensors regardless
of their type, location or network topology, inject queries and tasks and collect
replies;
 Communication with users – Gateway communicates with users or other sensor
networks over the Internet, wide area networks (GPRS, UMTS), satellite or some
short-range communication technology. It is possible to build a hierarchy of
gateways, i.e. to connect gateways described above to a backbone and then to provide
a higher-level gateway that is used as a bridge to other networks and users.

2.3. Applications
The features described above ensure a wide range of applications for sensor networks. Some
of the possible scenarios are given below:
 Health monitoring - Wireless sensor networks can be used in various ways to improve
or enhance health care services. Monitoring of patients, health diagnostics, drug
administration in hospitals, telemonitoring of human physiological data and tracking
and monitoring doctors and patients inside a hospital are some of the possible
scenarios. Various sensors (blood pressure, heart monitoring, etc.) can be attached to
the patient’s body to collect physiological data that can be either stored locally (on a
PDA or home PC) or forwarded directly to the hospital server or to the doctor in
charge. There are several advantages of such monitoring: it is more comfortable for
patients, doctors can have 24 hours access to patients and can better understand the
patient’s condition and last, but not the least, incurred expenses are much less than
when such tests are done in hospitals. Wearable sensors can also be used to track
patients and doctors in the hospital or to monitor and detect behavior and health
condition of elderly persons and children.
 Environmental monitoring - Fire detection, water pollution monitoring, tracking
movements of birds, animals or insects, detection of chemical and biological agents
 are some of the examples of environmental applications of wireless sensor networks.
For example, numerous smart sensor nodes with temperature sensors on board can be
dropped from an airplane over a remote forest. After successful landing, these devices
will self-organize the network and will monitor temperature profile in the forest. As
soon as the fire starts, that information along with the location of fire is transferred to
the command center that can act before the fire spreads to cover a large area.
 Military and security - The initial push towards wireless sensor network research
came from military agencies. Military applications are various and vary from
monitoring soldiers in the field, to tracking vehicles or enemy movement. Sensors
attached to soldiers, vehicles and equipment can gather information about their
condition and location to help planning activities on the battlefield. In case of nuclear
or biological attacks, sensor fields can gather valuable information about the intensity,
radiation levels or type of chemical agents without exposing people to danger.
Seismic, acoustic and video sensors could be deployed to monitor critical terrain and
approach routes or reconnaissance of enemy terrain and forces can be done.
 Industrial safety - Similar to personal health-care scenarios, wireless sensor networks
can be used for “health-care” of buildings, bridges or highways. In such scenarios
thousands of various sensors are deployed in and around monitored object and
relevant information is gathered and analysed in order to assess condition of the
object after a natural or man-made disaster. Similarly, sensor can be used to monitor
the status of different machines in factories, along with the air pollution or fire
monitoring.
 Other applications - Home automation, smart environments, environmental control in
office spaces, detecting car thefts, vehicle monitoring and tracking, interactive toys
are examples of other possible applications.

2.4. Research issues

Research issues are numerous and range from hardware issues to design of efficient
communication protocols and distributed data processing algorithms. All solutions have to
be, as said above, power conscious as well as fault tolerant, scalable, robust, with low
production cost etc. On the hardware side, wireless sensor networks require low-power, low-
cost devices that accommodate powerful processor, a sensing unit, wireless communication
interface and power source in a robust and tiny package. These devices have to work
autonomously, to require no maintenance and to adapt to environment. Obviously these
requirements are not easily met. Only since recently technology (MEMS) has become
available that enables production of very small sensing units with low power consumption.
Several solutions have been proposed so far. Physical layer issues range from power efficient
transceiver design to modulation schemes. MAC layer protocols have to support self-
organization of a distributed network and to ensure fair medium access and collision
avoidance. Different power modes have to be supported to enable nodes to save energy
resources when possible, but without affecting network performance. Changes in network
topology due to node malfunction or mobility have to be taken into account and dealt with
automatically. Several MAC protocols have been proposed. On the network level, routing
protocols are required for dissemination of user queries and tasks. Since data-centric
approach is used, the existing routing protocols for ad-hoc networks cannot be used and new
solutions, capable of routing messages based on data attributes are required. Several routing
protocols for different scenarios have been proposed so far. Another important requirement
for routing protocols in sensor networks is collaboration with data aggregation algorithms.
Data aggregation is required to avoid network implosion (this may happen when many nodes
answer the same query and send replies towards the gateway) and overlap problems in data-
centric routing.
Based on the predefined methods, responsible nodes analyse gathered data and combine it
into a set of meaningful information that is forwarded to users. Data aggregation reduces the
amount of network load while preserving validity and amount of information. For certain
applications it can be important to know the source of information (position) and in such
cases that information has to be forwarded as well. The data aggregation process is shown in
Figure 2

Figure 2. Data aggregation example

On the application level, a framework for attribute-based query definition, task building and
their execution at each node as well as collection of replies is required. Sensor network
management protocol has to support control of individual nodes (start/stop), network
configuration updates, location information data exchange, network clustering, data
aggregation rules etc. Sensor network gateway has to provide tools and functions for
presentation of network topology, services, characteristics etc. to users and to connect the
network to other networks and users.

3. Bluetooth based sensor networks

Bluetooth is a low-cost, short-range wireless technology with small footprint, small power
consumption, reasonable throughput and hence suitable for various small, battery-driven
devices like mobile phones, PDAs, cameras, laptops etc. Development of the Bluetooth
started several years ago with the intention to replace all sorts of cables used to connect
different devices. In meantime the idea has evolved and Bluetooth is now developing not just
as a point-to-point, but as a network technology as well. Bluetooth has gone through periods
of big hype when it was considered as the best short-range technology as well as through
periods when it was considered a failure. However, the last year could be seen as the turning
point year for Bluetooth. A lot of various Bluetooth devices and accessories appeared on the
market, broad range of users is able to use it and first experiences are generally positive. The
main challenge in front of Bluetooth developers now is to prove interoperability between
different manufacturers’ devices and to provide numerous interesting applications. An
example of such applications is wireless sensor networks. Low-cost (target price below
5EUR), low-power Bluetooth modules seem to meet the requirements of wireless sensor
networks well. Ad-hoc connection establishment capability, reasonable throughput (up to
721kbit/s in uplink and 56kbit/s downlink), usage of frequency-hopping scheme with TDD
(time division duplex) to minimize the impact of interference in the ISM band (2.4 GHz),
existence of different power saving modes along with its availability and standardized
specification are the main advantages of Bluetooth over other wireless technologies.
However, there are several issues, like connection establishment delay and networking
functionality, that have to be solved before Bluetooth can be deployed in large sensor
networks. In the following paragraphs, an initial implementation of a simple Bluetooth based
sensor network is described. The main issues experienced during the implementation and
solutions proposed are presented.
3.1. Sensor network related Bluetooth issues
Bluetooth operates in the 2.4GHz frequency band and uses frequency hopping spread
spectrum technique. There are 79 channels, each 1MHz wide, available for hopping. A
Bluetooth device has to be member of a piconet to be able to communicate with other
devices. A piconet is a collection of up to 8 devices that frequency hop together. each piconet
has one master, usually the device that initiated establishment of the piconet, and up to 7
slave devices. Master’s Bluetooth address is used for definition of the frequency hopping
sequence. Slave devices use the master’s clock to synchronize their clocks to be able to hop
simultaneously. When a device wants to establish a piconet it has to perform Inquiry to
discover other Bluetooth devices (these has to perform inquiry scanning at the same time) in
the range. Inquiry procedure is defined in such a way to ensure that two devices will after
some time visit the same frequency at the same time. When that happens, required
information is exchanged (Bluetooth address and clock of the device that will be master of
the piconet) and devices can use paging procedure to establish connection. Time required for
communication establishment can be rather lengthy, taking on average around 5s (minimum
is 0.00375s and maximum is 12.8s-33.28s). This delay can be the limiting factor for
applications that require instant connection establishment. When more than 7 devices needs
to communicate there are two options. The first one is to put one or more devices into the
park state. Bluetooth defines 3 low power modes: sniff, hold and park. When a device is in
the park mode then it disassociates from the piconet, but still maintains timing
synchronization with it. The master of the piconet periodically broadcasts beacons to invite
the slave to rejoin the piconet or to allow the slave to request to rejoin. Of course, the slave
can rejoin the piconet only if there is less than 7 slaves already in the piconet. If this is not the
case, then master has to park one of the active slaves first. All these actions cause delays and
for some applications it can be unacceptable (process control that requires immediate
response from the command centre for example). The other option is to build a scatternet.
Scatternet consist of several piconets connected by devices participating in multiple piconets.
These devices can be slaves in all piconets or master in one piconet and slave in other
piconets. Using scatternets, higher throughput is available and multi-hop connections
between devices in different piconets are possible. Unfortunately, hardware currently
available on the market still does not support this functionality due to several reasons:
 Specification gives no way for a slave to demand park, hold or sniff mode, but can
only request it from the master so there is no guarantee that the slave will be allowed
to leave one piconet and join the other;
 Each time a device switch between piconets it might lose up to two slots for
communication due to difference in piconet’s clocks;
 Scheduling switches between piconets in such a way to maintain communication links
with devices uninterrupted is very difficult, etc. Several schemes for scatternet
operation are proposed, but so far none of them is implemented. A possible solution,
before scatternet is supported by Bluetooth hardware, could be to perform switching
between piconets on the application level.
Proposed scatternet building mechanisms assume that all nodes in the network are peer
nodes. A mitigating circumstance for sensor networks is that gateway can be used to direct
establishment of the scatternet. Using such centralized approach it is possible to generate
more optimum network topology and solve scheduling, bandwidth allocation and routing
easier. However, there are some additional requests for scatternets in sensor networks that
complicate scatternet building. Various sensor types produce different amount of data (video
sensor and temperature sensor for example). If too many high-output sensors are connected to
the same branch in the scatternet it can cause link congestion or buffer overflow in
intermediate nodes. Hence, parameters like number of sensors, amount of data generated by
sensor per measurement and buffer size have to be taken into account during building
scatternet topology.

4. Sensor Network Implementation

The main goal of our implementation was to build a hardware platform and generic software
solutions that can serve as the basis and a testbed for the research of wireless sensor network
protocols. It supports ad-hoc deployment of sensors, sensor characteristics are automatically
collected and presented in a structured way using XML, there are no limits in terms of sensor
type and number of sensors, generic functions for querying sensors and collecting replies and
the basis for attribute-based routing are provided. Software architecture is designed in such a
way that new protocols can be added easily without affecting current functionality. During
implementation some compromises were made either to reduce development time or because
of lack of appropriate hardware and software. Implemented sensor network consists of
several smart sensor nodes and a gateway. Each smart node can have several sensors and is
equipped with a micro-controller and a Bluetooth radio module. Gateway has two wireless
interfaces: Bluetooth for communication with sensors and GPRS for communication with
users. Gateway and smart nodes are members of one piconet and hence maximum 7 smart
nodes can exist simultaneously in the network.

4.1. Smart sensor node implementation:

The smart sensor node comprises 3 functional blocks: sensing, data processing and
communication (Figure 3).
One or more sensors can be attached to the micro controller. Temperature, heart monitor and
smart fabric sensors were used. Intel 8051 micro controller is responsble for the smart sensor
node logic. An application is developed that gathers data from sensors and controls Bluetooth
module and communication with the gateway. It also stores sensors’ profiles and data. This
controller was chosen because of our previous experience with it, which should have reduced
development time. Bluetooth functionality is implemented using an off-the-shelf Bluetooth
device. This device provides set of AT-like commands for control of Bluetooth connections
over an RS232 interface. It was selected because it enables quick prototyping, without the
need for development of a Bluetooth stack suitable for the micro controller. However, there
are several drawbacks of this solution:
 AT commands provide the basic control over the Bluetooth device, but do not support
the full control of the Bluetooth protocol and settings. Two main problems are that
SDP (service discovery protocol) is not supported and it is not possible to assign
major and minor device class, and service class of the device. These deficiencies have
stipulated some of the applied solutions, like algorithm for smart sensor node
discovery;

Figure 3. Smart sensor node implementation

  The size of the smart sensor node is not negligible (5cmx5cmx5cm), i.e. it is much
bigger than required for many applications;
 Only serial port profile is supported.
However, using this off-the-shelf solution we were able to rapidly develop the prototype and
to develop and test higher-level protocols.

4.2. Sensor Network Gateway Implementation

Gateway is implemented on a Pentium laptop with Linux operating system. As complete
software functionality is developed in Java it can easily be transferred to other operating
systems. A Bluetooth PCMCIA card is used for the Bluetooth interface. GPRS connection is
built either over a GPRS PCMCIA card or using infrared connection with GPRS enabled
mobile phone. We have also tried to connect GPRS phone using Bluetooth, but due to lack of
Java PPP stack we were not able to control connection properly, so the first two solutions
were used. A Java Bluetooth stack with standard Java Bluetooth API was used. There are
many Bluetooth stacks available on the market. They are developed for different operating
systems, support different HCI (host controller interface) interfaces (serial port, USB, etc.)
and provide different APIs. This makes it very difficult to make a proper choice of the stack
and later to port developed application to different platforms. Java community undertook the
first effort towards Bluetooth stack API standardization and it has recently resulted in the first
standardized specification - the
JSR-82 JABWT (Java API for Bluetooth Wireless Technology). Gateway plays the role of
the piconet’s master in the sensor network. It controls establishment of the network, gathers
information about the existing smart sensor nodes and sensors attached to them and provide
access to them. Software architecture of the gateway is shown in Figure 4.

Figure 4. The architecture of the implemented gateway A set of core services is developed to
take care of common procedures and services required by all layers. Logging, scheduling,
event subscription and services required for automatic application starting and restarting are
supported. Sensor network communication interface handles communication with sensors
and can also control connections to mobile phone for GPRS connection or to local users that
are using Bluetooth to access the network. Depending on the available hardware resources
more than one sensor network can be attached to the gateway (using Bluetooth or any other
communication interface). The sensor network abstraction layer and its API are independent
of the underlying communication technology and provide information about and access to all
available sensors in the network. Public network interface provide access to GPRS services of
a public network. JXTA middleware is used as a framework for establishment of a P2P
network of gateway over a wide area network. Application layer designates either local
applications that reside on the gateway itself or remote applications that access the network
over the GPRS network.

4.3. Discovery of smart sensor nodes

Smart sensor node discovery is the first procedure that is executed upon the gateway
initialization. Its goal is to discover all sensor nodes in the area and to build a list of sensor’s
characteristics and network topology. Afterwards, it is executed periodically to facilitate
addition of new or removal of the existing sensors. The following algorithm is proposed.
When the gateway is initialized, it performs Bluetooth Inquiry procedure. When a new
Bluetooth device is discovered, its major and minor device class are checked (these
parameters are obtained along with the Bluetooth address and other parameters). These
parameters are set by each smart node to define type of the device (Major class = “smart
sensor node”) and type(s) of attached sensors (Minor class = “temperature”). Service class
field can be used to give some additional description of offered services (sensor accuracy,
manufacturer, etc.). If discovered device is not smart node it is discarded. Otherwise, SDP is
invoked and service database of the discovered smart node is searched for sensor services. As
currently there is no specific sensor profile, then database is searched for the serial port
profile connection parameters. Once connection string is obtained from the device, Bluetooth
link is established and data exchange with smart node can start. Due to the deficiencies of the
Bluetooth device we used in smart sensor nodes, it was not possible to implement proposed
algorithm and some shortcuts were used. As SDP is not supported by these devices,
connection string is built by concatenation of the Bluetooth address of the discovered device
and server channel number used by the smart node’s Bluetooth module (always set to 1).
After building the connection string, an attempt is made to establish the connection. If
establishment succeeds, the XMLSense protocol (the protocol we developed for data
exchange with sensors) is invoked and communication with micro controller application
attempted. Obviously this solution causes establishment of unnecessary links (discovered
device is not smart sensor node or is a smart sensor node, but does not have required sensors)
which wastes energy and causes delays.

4.4. Communication with smart sensor nodes

On top of the Bluetooth link, a simple, but very flexible data exchange protocol, the
XMLSense (XML based protocol for data exchange with sensors), has been developed for
sending and receiving data to and from sensors. Using two types of messages gateway can
request either list of sensors attached to the particular smart sensor node or sensor data. Reply
messages have very flexible structure and can relay information about any number or type of
sensor. Sensor profiles are defined by sensor vendors, according to an XML scheme, and are
sent as XML strings in reply messages. As all relevant information about a sensor is
contained in its profile (sensor type, measuring unit, accuracy, manufacturer, calibration date,
etc.), gateway can automatically build knowledge of sensor network and its characteristics,
i.e. sensors can be deployed in an ad-hoc fashion. Bluetooth links are maintained as long as
the gateway and smart sensor node are in the range. This approach has its advantages and
disadvantages. Good side is that as soon as an event happens on the sensor side, the sensor
can send information to gateway about it. Drawbacks are that power resources are wasted on
maintenance of communication link and it is not possible to have more than 7 smart nodes in
one piconet. The additional implementation problem, when JAVA APIs for Bluetooth is
used, is that the power saving modes related functions are not supported.

4.5. Abstraction layer

Gateway’s abstraction layer uses sensor profiles to create list of objects that represent each
sensor in the network. Each object provides methods that enable sending and receiving data
to and from sensor. Specifics of actual data transmission are hidden from users. Applications
can access sensor objects using queries that describe data they are interested in. Comparing
data description in the query and profiles associated with each sensor, gateway determines
which sensor can have an answer and sends data requests to them using methods provided by
each object. Collected replies are formatted into a structured XML document and forwarded
to users. Currently this is a rather simple functionality because all smart nodes have direct
communication links to the gateway and no routing is performed. If larger network and
scatternet topology were used, then similar functionality would have had to be provided by
each master in the scatternet and appropriate attribute-based routing solutions would be
required to disseminate queries. Data aggregation rules would have to be defined and could
be easily combined with the XML structured description of the answer.

5. Conclusions
Wireless sensor networks are an interesting research area with many possible applications.
They are based on collaborative effort of many small devices capable of communicating and
processing data. There are still many open issues ranging from the choice of physical and
MAC layer to design of routing and application level protocols. Bluetooth is a possible
choice for data communication in sensor networks. Good throughput, low-power, low-cost,
standardized specification and hardware availability are Bluetooth advantages, while slow
connection establishment and lack of scatternet support are some of the deficiencies. An
initial implementation of a Bluetooth based sensor network platform is presented.
Implemented functionality and various problems experienced during the implementation are
described. Implemented platform presents a good environment for further research and
development of sensor network protocols and algorithms.

6. References
[1] Mark Weiser, “The Computer for the Twenty-First
Century”, Scientific American, 1991.
[2] Wireless World Research Forum, “Book of Visions”,
http://www.wireless-world-research.org
[3] Scenarios for Ambient Intelligence, EU IST Advisory
Group
[4] http://www.janet.ucla.edu/WINS/lwim-innovative.htm