Semantic Sensor Networks Workshop

A wireless sensor network (WSN) is a wireless network consisting of spatially distributed autonomous devices using sensors to cooperatively monitor physical or environmental conditions, such as temperature, sound, vibration, pressure, motion or pollutants, at different locations.The development of wireless sensor networks was originally motivated by military applications such as battlefield surveillance. However, wireless sensor networks are now used in many civilian application areas, including environment and habitat monitoring, healthcare applications, home automation, and traffic control.

In addition to one or more sensors, each node in a sensor network is typically equipped with a radio transceiver or other wireless communications device, a small microcontroller, and an energy source, usually a battery. The envisaged size of a single sensor node can vary from shoebox-sized nodes down to devices the size of grain of dust, although functioning 'motes' of genuine microscopic dimensions have yet to be created. The cost of sensor nodes is similarly variable, ranging from hundreds of dollars to a few cents, depending on the size of the sensor network and the complexity required of individual sensor nodes.Size and cost constraints on sensor nodes result in corresponding constraints on resources such as energy, memory, computational speed and bandwidth.

A sensor network normally constitutes a Wireless ad-hoc network, meaning that it each sensor supports a multi-hop routing algorithm (several nodes may forward data packets to the base station).

Reactive Sensor Networks(RSN)

To coordinate work in a distributed sensor network, an active network of mobile code is being constructed. The system uses resource bounded optimization to adapt dynamically to a chaotic environment. This 3-year research effort started in July 1999.

Users make demands for information. Placing them at the center of an active network. Sensors are present at multiple locations in the environment. The network forages for information, like ants forage for food.

Body Sensor Networks (BSN)

he last decade has witnessed a rapid surge of interest in new sensing and monitoring devices for healthcare and the use of wearable/wireless devices for clinical applications. One key development in this area is implantable in vivo monitoring and intervention devices. While the problem of long-term stability and biocompatibility is being addressed, several promising prototypes are starting to emerge for managing patients with acute diabetes, for treatment of epilepsy and other debilitating neurological disorders and for monitoring of patients with chronic cardiac diseases. Despite the technological developments of sensing and monitoring devices, issues related to system integration, sensor miniaturization, low-power sensor interface circuitry design, wireless telemetric links and signal processing have still to be investigated. Moreover, issues related to Quality of Service, security, multi-sensory data fusion, and decision support are active research topics. To address general issues related to using wearable/wireless and implantable sensors and to bring together scientists from computing, electronics, bioengineering, medicine and industry, the term BSN – Body Sensor Networks was coined by Prof Guang-Zhong Yang of Imperial College in 2002 and after much preparation, the first International Workshop on BSN was launched in 2004. Some of the key research focuses of the BSN community includes the latest technological developments and clinical applications of:

• Novel bioelectrical, biochemical, biophysical, and mechanical sensors
• Hardware considerations: low power RF transceiver, energy scavenging, battery technology, miniaturisation, system integration, process and cost of manufacturing
• Biocompatibility and materials
• Context awareness and multi-sensor data fusion
• Data inferencing, knowledge discovery, and prediction
• Quality of service, trust and security issues
• Autonomic sensor networks
• Standards and light-weight communication protocols
• Integration with ambient sensing with applications in smart dwellings, and home monitoring
• Wearable and implantable sensor integration and development platforms
• Clinical applications of body-sensor networks

Environmental Sensor Networks

Introduction

Environmental sensor networks have the capability of capturing local and broadly-dispersed information simultaneously; they also have the capacity to respond to sudden change in one l ocation by triggering observations selectively across the network while simultaneously updating the underlying complex system model and/or reconfiguring the network. Data gathered by wireless sensor networks, either fixed or mobile, pose unique challenges for environmental modeling: a complex system is being observed by a dynamical network. Technical challenges in statistics (sampling design to prediction and prediction uncertainty), in mathematics (computational geometry to data fusion to robotics), and in computers science (self-organizing networks to algorithm analysis) combine with the technical challenges of the models themselves and the sciences that underlie them.

Sampling from wireless networks:
Cost of spatio-temporal data in terms of both energy and delay: Each sample has a footprint in power in space and time, some value to one or more process models (e.g., importance of parameters in space and time, sensitivity of estimates to the observation), and some cost (e.g., data transmission). Is it possible to derive frameworks such that the utility of each sample exceeds its cost?

Frameworks for adaptive sampling: game-theoretic, reinforced learning, dynamic experimental design. How can a sampling scheme to respond to highly non-stationary dynamics in energy-constrained sampling networks. Are modes of operation controlled by adaptive state machines enough?

Environmental modeling from sensor networks:
Model complexity and adequacy: In the trade-off of dimensionality and predictive accuracy, what are the diagnostics for excessive vs. insufficient parametrization? Can models be developed so that reduced forms (e.g., deleting submodels, subsets of parameters, or reducing resolution of observations and/or parameter specification) still function simultaneously with near-optimality at several scales?

Prediction Uncertainty: Appropriate modeling of sources of uncertainty due to sampling (e.g., "lost" samples, outliers, bad sensors, measurement noise, and unreliable communication), and integration of data models with process models.

Model adequacy: Is there coherent noise in bio-micrometeorological systems that should drive exploration of new regions of the state space?

Networks, forests and global change:
Process level understanding of how forested ecosystems respond to global change is critical for anticipating consequences of human impacts on landscapes. To be successful an approach will entail integrated models of a complex system and will involve heterogeneous data and analyses that directly address uncertainty and model selection issues.

Specific analyses will focus on increasing the capacity to forecast consequences of global change, using existing data from forest sensor networks as both testbed and primary research objective.

Inference on scaling relationships will be implemented in simulation of whole forests to examine potential consequences of changing climate and atmospheric CO2 for forest diversity and carbon sequestration. These results will have immediate application to the problem of forecasting biosphere responses to atmospheric change.

Reactive Sensor Networks(RSN)

To coordinate work in a distributed sensor network, an active network of mobile code is being constructed. The system uses resource bounded optimization to adapt dynamically to a chaotic environment. This 3-year research effort started in July 1999.

Users make demands for information. Placing them at the center of an active network. Sensors are present at multiple locations in the environment. The network forages for information, like ants forage for food.

Nuclear smuggling sensor networks upgraded

The U.S. Department of Homeland Security is funding a $1.9 million upgrade of a computer program involving detection of nuclear smuggling.

The grant was awarded University of Texas at Austin Professor David Morton and colleagues to expand an existing computer model that guides the placement of sensors to detect nuclear smuggling attempts.

The U.S. Department of Homeland Security provided the funds to improve the design of sensor networks in Russia and other former Soviet Union nations that have insufficient security for their stores of nuclear weapons and radioactive material.

"Russia's got the biggest border of any country on the planet, making it highly unlikely the country could seal its borders," said Morton. "So the real issue becomes: given the limited resources and the fact that radiation detectors can cost upward of $1 million to set up, can we provide a computer tool that locates the detectors optimally?"

The United States has provided more than $100 million to place radiation detectors at Russian sites where smugglers could escape with material for preparing nuclear weapons or dirty bombs. The computer model prioritizes decisions on site selection.

Semantic Sensor Networks Workshop

Semantic technologies are often proposed as important components of complex, cross-jurisdictional, heterogeneous, dynamic information systems. The needs and opportunities arising from the rapidly growing capabilities of networked sensing devices are a challenging case.

Current and future sensing systems involve distributed wired and wireless networks consisting of large numbers of sensors, including active and passive RFID tags. Geographically distributed sensor nodes are capable of forming ad hoc networking topologies that interconnect with backend information management systems and services. Sensor nodes are expected to be dynamically inserted and removed from a network due to deployment of new sensor nodes, failure of deployed sensor nodes, and mobility of tagged objects or sensing platforms.

The goal of a sensor networking system is to improve the situational awareness of business activities across widely distributed deployment environments involving a large number of diverse active and passive sensor nodes. Important applications include natural resource management, product lifecycle management, supply chain management and situation awareness on the battlefield.

The goal of the Semantic Sensor Net workshop is to develop an understanding of the ways semantic web technologies, including ontologies, agent architectures and semantic web services can contribute to the growth, application and deployment of large-scale sensor networks. The workshop will provide an inter-disciplinary forum to explore and promote these concepts.

Topics include, but are not limited to:

• Ontologies for sensor and RFID networks
• Semantic web services architectures for sensor networks
• Semantic data integration in large-scale heterogeneous sensor networks
• Semantic middleware for active and passive sensor networks
• Semantic algorithms for data fusion and situation awareness
• Experience in applications of semantic technologies in sensor networks
• Rule-based sensor systems
• Reasoning with incomplete or uncertain information in sensor networks
• Semantic policy management in inter-organisational networks
• Semantic feedback and control
• Scalability in semantic sensor networks
• Sensor network topology management using semantic reasoning
• Emergent semantics in sensor network systems

Wireless Ad Hoc Sensor Networks

A wireless ad hoc sensor network consists of a number of sensors spread across a geographical area. Each sensor has wireless communication capability and some level of intelligence for signal processing and networking of the data. Some examples of wireless ad hoc sensor networks are the following:

1. Military sensor networks to detect and gain as much information as possible about enemy movements, explosions, and other phenomena of interest.
   
2. Sensor networks to detect and characterize Chemical, Biological, Radiological, Nuclear, and Explosive (CBRNE) attacks and material.
   
3. Sensor networks to detect and monitor environmental changes in plains, forests, oceans, etc.
   
4. Wireless traffic sensor networks to monitor vehicle traffic on highways or in congested parts of a city.
   
5. Wireless surveillance sensor networks for providing security in shopping malls, parking garages, and other facilities.
   
6. Wireless parking lot sensor networks to determine which spots are occupied and which are free.

The above list suggests that wireless ad hoc sensor networks offer certain capabilities and enhancements in operational efficiency in civilian applications as well as assist in the national effort to increase alertness to potential terrorist threats.

Two ways to classify wireless ad hoc sensor networks are whether or not the nodes are individually addressable, and whether the data in the network is aggregated. The sensor nodes in a parking lot network should be individually addressable, so that one can determine the locations of all the free spaces. This application shows that it may be necessary to broadcast a message to all the nodes in the network. If one wants to determine the temperature in a corner of a room, then addressability may not be so important. Any node in the given region can respond. The ability of the sensor network to aggregate the data collected can greatly reduce the number of messages that need to be transmitted across the network. This function of data fusion is discussed more below.

The basic goals of a wireless ad hoc sensor network generally depend upon the application, but the following tasks are common to many networks:

1. Determine the value of some parameter at a given location: In an environmental network, one might one to know the temperature, atmospheric pressure, amount of sunlight, and the relative humidity at a number of locations. This example shows that a given sensor node may be connected to different types of sensors, each with a different sampling rate and range of allowed values.
   
2. Detect the occurrence of events of interest and estimate parameters of the detected event or events: In the traffic sensor network, one would like to detect a vehicle moving through an intersection and estimate the speed and direction of the vehicle.
   
3. Classify a detected object: Is a vehicle in a traffic sensor network a car, a mini-van, a light truck, a bus, etc.
   
4. Track an object: In a military sensor network, track an enemy tank as it moves through the geographic area covered by the network.

In these four tasks, an important requirement of the sensor network is that the required data be disseminated to the proper end users. In some cases, there are fairly strict time requirements on this communication. For example, the detection of an intruder in a surveillance network should be immediately communicated to the police so that action can be taken.

Wireless ad hoc sensor network requirements include the following:

1. Large number of (mostly stationary) sensors: Aside from the deployment of sensors on the ocean surface or the use of mobile, unmanned, robotic sensors in military operations, most nodes in a smart sensor network are stationary. Networks of 10,000 or even 100,000 nodes are envisioned, so scalability is a major issue.
   
2. Low energy use: Since in many applications the sensor nodes will be placed in a remote area, service of a node may not be possible. In this case, the lifetime of a node may be determined by the battery life, thereby requiring the minimization of energy expenditure.
   
3. Network self-organization: Given the large number of nodes and their potential placement in hostile locations, it is essential that the network be able to self-organize; manual configuration is not feasible. Moreover, nodes may fail (either from lack of energy or from physical destruction), and new nodes may join the network. Therefore, the network must be able to periodically reconfigure itself so that it can continue to function. Individual nodes may become disconnected from the rest of the network, but a high degree of connectivity must be maintained.
   
4. Collaborative signal processing: Yet another factor that distinguishes these networks from MANETs is that the end goal is detection/estimation of some events of interest, and not just communications. To improve the detection/estimation performance, it is often quite useful to fuse data from multiple sensors. This data fusion requires the transmission of data and control messages, and so it may put constraints on the network architecture.
   
5. Querying ability: A user may want to query an individual node or a group of nodes for information collected in the region. Depending on the amount of data fusion performed, it may not be feasible to transmit a large amount of the data across the network. Instead, various local sink nodes will collect the data from a given area and create summary messages. A query may be directed to the sink node nearest to the desired location.

Sensor types and system architecture:
With the coming availability of low cost, short range radios along with advances in wireless networking, it is expected that wireless ad hoc sensor networks will become commonly deployed. In these networks, each node may be equipped with a variety of sensors, such as acoustic, seismic, infrared, still/motion videocamera, etc. These nodes may be organized in clusters such that a locally occurring event can be detected by most of, if not all, the nodes in a cluster. Each node may have sufficient processing power to make a decision, and it will be able to broadcast this decision to the other nodes in the cluster. One node may act as the cluster master, and it may also contain a longer range radio using a protocol such as IEEE 802.11 or Bluetooth.

Largest Tiny Network Yet

Large-Scale Demonstration of Self-Organizing Wireless Sensor Networks

On Aug. 27, 2001 researchers from the University of California, Berkeley and the Intel Berkeley Research Lab demonstrated a self-organzing wireless sensor network consisting of over 800 tiny low-power sensor nodes. This demonstration highlights work at Berkeley that is funded, in part, by the Defense Advanced Research Project Agency (DARPA) Network Embedded Softrware Technology program and is a leading component of the CITRIS research agenda, as well as the collaboration with Intel. The demonstration was live, involving most of the audience attending the kickoff keynote of the Intel Developers Forum given by Dr. David Tennenhouse, Intel VP and Director of Research.

A sensor network application

The basic concept of a self-organized sensor network was first demonstrated at a moderate scale on stage by Professor David Culler (University Director of the Intel Berkeley Research Lab) and several students from UCB and UCLA. A 1x1.5 inch wireless sensor node was carried on stage and activated by each of several students. As the nodes were turned on, their icons appeared on a display showing the state of the network. An example screenshot appears below. The lines indicate which nodes can 'hear' each other and communicate via their radios. The lines highlighted as green are the links chosen by the network to form an ad hoc, multihop routing structure to transmit sensor data to the display. This structure grew as the nodes were activated and continued to adapt to changing conditions. The background showed the network's 'perception' of light intensity on the stage. Initially, it was white at full illumination. Taking the stage lights down brought it to dark. When selective spots were brought up, the corresponding regions when light. As the students walked away, the network took itself apart.

Live Ad Hoc Sensor Network showing Light Intensity

Tiny Nodes

A large-scale demonstration of the networking capability utilized tiny nodes, just the size of a quarter, that were hidden under 800 chairs in the lower section of the presentation hall. The core is a 4 mhz low power microcontroller (ATMEGA 163) providing 16 KB of flash instruction memmory, 512 bytes of SRAM, ADCs, and primitive peripheral interfaces. A 256 KB EEPROM serves as secondary storage. Sensors, actuators, and a radio network serve as the I/O subsystem. The network utilizes a low-power radio (RF Monolithics T1000) operating at 10 kbps. The node had fours senses: light, temperature, battery level, and radio signal strength. It can actuate two LEDS, control the signal strength of the radio, transmit and receive signals. By adjusting the signal strength, the radio cell size can be varied from a couple of feet to tens of meters, depending on physical environment. A second microcontroller is provided to all the core microcontroller to be reprogrammed over the network. The entire system consumes about 5 mA when active. The radio and the microcontroller consume about as much power as a single LED. In the passive mode, they consume only a few microamps while still checking for radio or sensor stimuli that should cause them to 'wake up'.

A handful of network sensor 'dots'

Lots of 'dots' - getting ready for the big demo

TinyOS

The key innovation is a novel operating system, including network stack, designed especially resource constrained environments where data and control have to be moved rapidly beteen various sensors, actuators, and the network. TinyOS is a component-based, event-driven operating system framework that starts at a few hundred bytes for the scheduler and grows to complete, network applications in a few kilobytes. The demonstration application consisted of nine software components. At the lowest level are abstraction components for the network, the sensors and the LEDs. Building upon this is a very efficient network stack which reaches from modulating the physical link in software up to an "Active Message" application programming interface. One application component provides multihop routing, where nodes can communicate to other nodes that are several radio hops away by having intermediate nodes route and retransmit their packets. Other application components sample sensor data and process it in various ways. Still others manage the basic health of the node or network.

A Self-Organizing Network

For this demonstration, one of the application components provided network discovery and multihop broadcast. A command packet can be transmitted from any node that needs to become the root of a logical network. In the demonstration, this was the node connected to the laptop on stage. Nodes receiving that packet will selectively retransmit the command, allowing the request to 'ripple' out over many levels. The network supported several such commands, including network discovery where each node records the identity of a 'parent' closer to the root. As this request propagates out, a routing tree is grown that spans the network. Many other actions can be requested of the network, such as illuminating LEDs, accessing data, changing modes, or going to sleep. Waking up the network from the very low power sleep mode is especially subtle, because it must use only a minute amount of energy and yet avoid
false-positives, which would awaken the entire network. The entire application in the demo occupied about 8 KB.

The intended plan for the demo was to begin with all the nodes asleep. A wakeup request would be propagated through the network, causing each node to light yellow during the boot phase and then red when ready. A network discovery operation would show yellow briefly as the discovery wavefront propagated across the network and then turn off the red LED. Because the discovery takes only a fraction of a second, we would then issue a sequence of commands to illuminate level 1, then level 2, etc. at five second intervals. Test runs in a near empty auditorium discovered a network of 800 nodes that was four levels deep. In the live demonstration, a few folks had found their nodes, taken them apart, and played with them - causing the network to wake up, so everything was awake when the audience pulled them from under the chairs. The discovery went by so quickly no one saw it and the lights were off. We had to tell the nodes to turn their lights back on and rediscovery the network, which turned out to be eight levels deep. Actuating the crowd level-by-level showed the complex structure of such a large, self-organized network.

Network Questions??

Many issues impact network performance besides the hardware topology. To identify them, you have to ask the right questions.

Full duplex vs. half duplex vs. simplex—Can the nodes talk and listen simultaneously?

Devices that can talk and listen at the same time (e.g., telephones) are full-duplex devices. Citizen Band (CB) and other radio formats are normally half duplex—meaning they can talk and listen, but not simultaneously. In this case, some indication is usually required to let the other party know it’s okay to talk (e.g., by saying “over”). RS-232 is a half-duplex data bus, and Ethernet is a full-duplex connection. Many modern devices can simulate full-duplex performance by switching between transmit and receive fast enough (in milliseconds) that humans can’t perceive the delay. Most cell phones are implemented with this fast switching strategy. Simplex systems communicate in one direction only. Half-duplex systems operate like simplex systems and then reverse the roles of transmitter and receiver.

Analog vs. digital—In what form does the signal enter the hardware medium?

In analog systems, the modulation technique is continuously variable (e.g., voice). Digital systems use an A/D converter to digitize the signal and send a data packet that uses 1s and 0s to represent the analog value. Digital transmissions offer such advantages as reduced fading, reduced noise, and increased throughput. Analog systems potentially have better resolution, though, because no digitization error is involved. In the old analog telephone systems, the voice modulated the electrical resistance and thus the voltage across the carbon granules packed in the mouthpiece. Digital systems replaced the old analog phone circuits many years ago, except for the phone line from your phone to the first switching station. The 4–20 mA system is an analog channel, and RS-232 and Ethernet are digital.

Baseband vs. broadband—Should you use a carrier to increase the number of channels that can be put on the network medium?

If the signal containing information is placed directly on the physical medium, the channel is called baseband. If the signal is placed on a carrier (modulation), the channel is broadband. Because many carriers can be placed on the same medium at different frequencies, a given hardware channel can carry many logical channels. Cable TV is a broadband network, with each TV channel on a different carrier frequency. The TV channel designates the carrier frequency. Optical fibers now carry many more channels than once thought possible by modulating information onto the fiber at different frequencies of the light (colors). Broadband systems are usually more complicated, so the tradeoff must be made with care. Standard Ethernet is a baseband bus, but wireless (radio) buses are usually broadband.

Master-slave vs. peer-to-peer vs. broadcast—How should the nodes interact with each other and with the host?

In master-slave protocols, one node gives the commands, and another node or collection of nodes executes them. The host is usually the master, and the sensors and actuators are usually slaves. This protocol allows tight traffic control because no node is allowed to speak unless requested by the master, and no communication is allowed between slaves except through the master.

In a peer-to-peer network, all nodes are created equal. A node can be a master one moment and then be reconfigured at another time. Peer-to-peer configurations offer the greatest flexibility, but they’re the most difficult to control. Any node can communicate directly with any other node.

Broadcast networks are much like master-slave configurations, but the master can send commands to more than one slave at a time. Many industrial protocols (e.g., IEEE-1451) are based on master-slave (with broadcast) protocols. Wireless systems can be implemented in any of these protocols.

Circuit switched vs. packet switched—How long should a node own a communications channel?

Again, the old analog phone system is used as an example. These systems were circuit-switched networks. You dialed the number, and a circuit was established between the sender and the receiver. The circuit stayed connected as long as the phone call continued. When the parties hung up, the circuit was released and available for another connection.

Packet-switched networks route digital packets of information as they travel along different paths throughout the network. Each packet contains routing information so that the receiver can reassemble the packets into a complete message when they arrive. Complexity is high, but the potential for flexibility and improved channel use is also high. The Internet and World Wide Web are based on packet-switched networks.

Wireless Sensor Network Topologies

As wireless sensors come onto the market and become a practical option, sensor networks take on a whole new dimension. The question arises: How can traditional network topologies be adapted to the new communications medium?

The development of network technologies has prompted sensor folks to consider alternatives that reduce costs and complexity and improve reliability. Early sensor networks used simple twisted shielded–pair (TSP) implementations for each sensor. Later, the industry adopted multidrop buses (e.g., Ethernet). Now we’re starting to see true web-based networks (e.g., the World Wide Web) implemented on the factory floor.

Figure 1. In point-to-point network topologies, each sensor node requires a separate twisted shielded–pair wire connection. The cost is high, configuration management is difficult, and nearly all the information processing is done by the host.

As wireless sensors become real commodities on the market, new options or new arguments for old options are causing professionals to consider network strategies once ruled out. Let’s look at the three classic network topologies (point-to-point, multidrop, and web), assess their strengths and weaknesses, and look at how the rules have changed now that wireless systems are coming online.

In addition, to build functional sensor networks, you’ll probable have to integrate hardware and software from multiple vendors (see the sidebar “Network Questions,”). So along with everything else, you have to come to terms with standards and protocols—those that exist, those that are emerging, and those needed to ensure interoperability on the factory floor.

Point-to-Point Networks
Theoretically, these systems are the most reliable because there is only one single point of failure in the topology—the host itself (see Figure 1). You can improve the system by adding redundant hosts, but wiring two hosts can be a problem. The 4–20 mA standard allows multiple readout circuits if the standard loads are used at each readout. Problems can arise if readout devices load the circuit beyond its capability, but most designers are familiar with the limitations and are sufficiently careful.

Figure 2. In a multidrop network, each sensor node puts its information onto a common medium. This requires careful attention to protocols in hardware and software. The single-wire connection represents a potential single-point failure. But some vendors supply redundant connections to mitigate this potential problem.

Some networks provide frequency-modulated (FM) signals on the wires to carry multiple sensor readings on separate FM channels. Some standards (e.g., the HART bus) support multiplexing of digital signals on the existing analog wiring in older plants. These architectures blur the distinction between point-to-point and multidrop networks.

Early wireless networks were simple radio-frequency (RF) implementations of this topology. These networks used RF modems to convert the RS-232 signal to a radio signal and back again. Fluke (Everett, Wash ington) developed a digital voltmeter that could be configured to accept a voltage signal and transmit the signal over a dedicated radio frequency channel. The reliability of these implementations was sometimes suspect because most were designed with simple FM coding. Interference and multipath propagation effects caused significant degradation in factory environments, so many networks proved to be unreliable unless designers were particularly careful. The Federal Communications Commission licensed companies and devices to operate at the allocated frequencies.

Complete wireless local area networks (LANs) were implemented using this technique.These were successful in the office environment but didn’t fare as well in factories. Many designers implemented remote data acquisition systems with this topology by using a data concentrator in the field to feed the data to a radio transmitter for transmission to the hosts, where the signals were demultiplexed into the original sensor signals.

Multidrop Networks

Figure 3. In a web topology, all nodes are potentially connected to all other nodes. Connectivity among a large collection of sensors gets complex because all nodes must have a connection to all other nodes. Some connections can be eliminated by using repeaters and routers to make virtual connections. The World Wide Web is a good example of this topology.

Multidrop buses began to appear in the late 70s and early 80s. One of these, Modbus from Modicon (Schneider Auto mation, North Andover, Massa chusetts), led the way into the industrial sphere, followed by several proprietary and open buses (e.g., the Manufacturing Auto mation Protocol, QBus, and VME Bus).

The emergence of intelligent sensors and microcomputers capable of operating in industrial environments irrevocably changed the sensor network landscape. Multidrop networks (buses) reduced the number of wires required to connect field devices to the host, but they also introduced another single point of failure—the cable. Several suppliers of industrial-grade products offered redundant cabling designs, but these came with an increase in complexity (see Figure 2).

Once the industry began the migration to multidrop buses, problems associated with digitization began to emerge. With the previous point-to-point systems, digitization occurred in the host, where a single clock could be used to time stamp when the analog signals from multiple sensors were acquired. With the distributed intelligence required to implement a multidrop network, synchronization of clocks became a critical issue in some applications. This remains an important design parameter for any distributed digital system.

Figure 4. An architecture consisting of a decoder for each channel and a direct-sequence spread-spectrum receiver can perform simultaneous sampling because the same baseband signal goes to each decoder. But the decoders represent a significant cost, power, and size limitation.

The introduction of Ethernet in the mid-80s was a landmark in standardization, if not technological innovation. A group of large companies agreed that the future of computer networking required an open interconnect standard that would allow multiple-vendor systems to work together with minimal difficulty.

Researchers looked closely at the carrier sense multiple access with collision detection (CSMA/CD) protocol when they investigated the behavior of networks under stress. But they considered most industrial applications too time critical for such a nondeterministic protocol. Now, fifteen years later, most factories have converted their shop floor networks to Ethernet because it is the best compromise between cost and performance. Many companies now offer solutions that use Ethernet to implement suitable robust industrial networks.

Wireless systems use the same types of protocols to implement multidrop topologies, simulating hard-wired connections with RF links. The IEEE-802.11 standard was the first wireless standard that promised to bring the interoperability of Ethernet connectivity to wireless networks. Many of these, however, are not compatible at the over-the-air level.

Web Networks

Figure 5. Simultaneous sampling is more difficult with this receiver architecture. The selected channel codes can be stored and stepped through so that each channel’s data gets to the data system bus.

The promise of the web topology (i.e., when all nodes are connected all the time) had to wait until vendors developed a way to interconnect nodes without the required wiring connections. A network of any appreciable size becomes infeasible if all wires must be connected specifically for the network (see Figure 3). Early star topologies were successful as long as the star wasn’t too large. The World Wide Web illustrates what is possible, though, if you can use wiring that is already in place. The telephone network provides the available connectivity in most parts of the country, although at less than suitable speeds in many locations.

The advantages of web connectivity for sensor networks become clear as the level of intelligence in each sensor increases. Cooperating sensors can form a temporary configuration that provides sufficient capacity to replace the host. Self-hosting networks then become self-configuring and finally, years from now, perhaps even self-aware. But several problems remain and are the topic of significant research, such as size and power consumption reduction, throughput and performance during transmissions, and algorithms for allocating priorities and authority.

In a wireless web network, individual nodes have the potential of being constantly connected (physically) with many other nodes in the network. How the network is configured at any instant becomes a matter of how the software configures it. In a code division multiple access (CDMA) network, the radios can receive all channels at once. Figures 4 and 5 illustrate the two simplest alternatives for implementing a CDMA-based data receiver.

The architecture suggested in Figure 4 requires a separate decoder for each channel. This requires hardware to be dedicated to channels that may not be currently important but could be required later. Figure 5 eliminates the need for dedicated hardware but introduces the problem of simultaneous sampling. The decoder-per-channel implementation samples the data stream looking for a particular channel code embedded in the chip stream. The single decoder will decode a new data stream for each channel unless the data stream is stored and decoded over and over with different candidate codes for each channel. Both implementations represent a compromise and should be implemented carefully, depending on the application.

Network routing is a serious concern in web architectures. Because all nodes can’t reach all other nodes in a single hop, a repeating mechanism is required. The assigned input and output channels dictate to each node which signals are meant for its own use and which should be passed on to the next node. The routing is one of the things that makes web architectures more complicated to implement than the others.

In sensor or mobile phone networks, nodes can come and go frequently. How the network responds to the reconfiguration has a severe impact on performance and reliability. Mobile ad hoc networking is a hot topic in the research community because reconfiguring on the fly makes all networks better. Without this technology, sensor networks will be severely limited in harsh environments, where connections can change quickly as the RF environment changes.

So What?
Network topologies usually work best when they map closely to the topology of the application. If the application looks hierarchical, then a hierarchical (point-to-point or multidrop) topology might be most suitable. But if the application looks like a collection of peers interacting and cooperating, then a web architecture might work best.

The potential for web connectivity in the sensor world seems most tempting. The dynamic nature provides the opportunity for cooperating sensors to form smart clusters that can work together to solve a problem, then reorganize to solve the next one. As the hardware and software technologies mature, you’ll see more and more web implementations showing up on factory floors.

sensor network

A sensor network is a group of specialized transducers with a communications infrastructure intended to monitor and record conditions at diverse locations. Commonly monitored parameters are temperature, humidity, pressure, wind direction and speed, illumination intensity, vibration intensity, sound intensity, power-line voltage, chemical concentrations, pollutant levels and vital body functions.

A sensor network consists of multiple detection stations called sensor nodes, each of which is small, lightweight and portable. Every sensor node is equipped with a transducer, microcomputer, transceiver and power source. The transducer generates electrical signals based on sensed physical effects and phenomena. The microcomputer processes and stores the sensor output. The transceiver, which can be hard-wired or wireless, receives commands from a central computer and transmits data to that computer. The power for each sensor node is derived from the electric utility or from a battery.

Potential applications of sensor networks include:

• Industrial automation
• Automated and smart homes
• Video surveillance
• Traffic monitoring
• Medical device monitoring
• Monitoring of weather conditions
• Air traffic control
• Robot control.

Next Century Challenges: Scalable Coordination in Sensor Networks

Networked sensors those that coordinate amongst them- selves to achieve a larger sensing task will revolutionize information gathering and processing both in urban environments and in inhospitable terrain. The sheer numbers of these sensors and the expected dynamics in these environments present unique challenges in the design of unattended autonomous sensor networks. These challenges lead us to hypothesize that sensor network coordination applications may need to be structured divergently from traditional network applications. In particular, we believe that localized algorithms (in which simple local node behavior achieves a desired global objective) may be necessary for sensor net-work coordination. In this paper, we describe localized algorithms, and then discuss directed delusion, a simple communication model for describing localized algorithms.