I/O and Sensor Technology

Wireless Nodes Dynamically Link to Build Intelligent Sensing Networks

The age of ubiquitous sensing and actuation has begun. It is fueled by Moore’s Law and the development of advanced wireless sensor networking platforms.


  • Page 1 of 1
    Bookmark and Share

Sensor networks are a type of wireless system in which a large number of sensors can be deployed. They have been enabled in part by advancements in silicon sensor technology, low-power microcontrollers, chip-based radios, ad-hoc networking protocols, operating systems and languages for embedded systems. The vision for this technology is to improve our ability to monitor, capture and analyze events in our operations and environment so that we may be quicker in identifying opportunities for better performance and more responsive in taking action on those opportunities.

Several key concepts distinguish wireless sensor networks from other types of wireless networks. The first is that of a physically small network node with devices ranging from a millimeter-size silicon chip to the size of a personal digital assistant, which combines sensing, processing and radio communication. The second key concept is the ability to meet the harsh power requirements placed on them. The nodes are typically battery operated, but unlike cell phones or wireless laptops, periodic recharging is not possible in much of a sensor network. Therefore, the nodes are designed to work in the field for years without any maintenance, and their operating lifetime can be extended with energy scavenging techniques.

A third key concept is that each node is programmed with the same peer-to-peer networking protocol that makes a group of individual nodes into a self-configuring mesh network. This enables three benefits in networked sensors. First, there is no dependence on a backbone infrastructure. Such nodes also offer scalability and energy efficiency in communications. Sensor networks often have redundancy in the network topology and therefore the network is robust against individual sensor node or RF link failure.

The true mesh network protocol stacks developed for sensor networks are more scalable in practical situations and may be the only way to achieve reliable quality of service while keeping the power requirements low for battery-operated nodes. The third benefit is that a multi-hop mesh network permits energy efficiency over an equivalent single-hop network (as found in star or hybrid-star networks) for the same distance.

Hardware for Wireless Sensor Network Platforms

The current vision of sensor networks is to have hundreds or thousands of sensors in a single deployment. Initial deployment experience has shown that these systems will require a hierarchy of nodes starting with simple sensor nodes and continuing up to data aggregation, analysis and storage nodes. The higher-level nodes form a distributed infrastructure designed to support the operation of the low-power, low-cost sensor nodes. This tiered architecture is common to virtually all sensor networks and is best illustrated by an example.

Consider the application of a sensor network for an advanced security system. The network will need to function as one cohesive system, but it will be built out of a large number of individual nodes. Within a single security system, a majority of the sensors will be simple sensors such as contact closure, light sensors and motion detectors. The wide distribution of these nodes makes it advantageous for them to be battery operated. The system will likely include a handful of more sophisticated sensors such as cameras, acoustic or chemical detectors located at key places that supplement the simple sensors. All of the sensors will eventually be linked into a building control facility that provides anytime monitoring, even in the presence of partial hardware failure.

The simple sensors that are placed on a window or a door for intrusion detection are good examples of the class of node referred to here as a general sensing device. Their function is simple and specific and requires long-term battery operation. The algorithms and processing required to analyze the raw sensor data are straightforward and require minimal CPU and memory resources. The data rates in and out of the node are also minimal.

In contrast, the acoustic, video and/or chemical sensors require more computational resources to determine if an event has occurred and more communications capacity to report or describe an event. For example, it may be necessary to analyze images in order to track objects moving through the field of view or detect unexpected motion. Once an event is detected the node would then transmit the video stream across the network. While these nodes may benefit from battery operation, wall power will be needed for long-term operation. This characterizes these nodes as high-bandwidth devices (Figure 1).

In addition to including traditional security sensors, wireless sensor networks have a new ability to track mobile assets or personnel by attaching a tiny security tag. These special-purpose sensor nodes are the most synonymous with “smart-dust”—tiny devices operating from very limited energy resources. Today’s security systems have no analog-to-special-purpose sensor nodes. With such a device, an alarm could be triggered if an asset leaves the facility without authorization. In this scenario, it is essential that the device be highly integrated and very low cost. The special-purpose sensor is more powerful than an RFID tag, but is not as general-purpose as a generic sensor node. It is a highly integrated, low-power processor and radio with a minimum set of interfaces. These hardware classes are often referred to as motes (Figure 2).

All of the wireless sensors will communicate over a single-hop or a multi-hop mesh network back to a building control center. As part of the system, the mesh network will have one or more end-points or gateways that contain database software designed to process and store the individual sensor readings. Gateway nodes play a vital role as a distributed infrastructure to facilitate low duty cycle operation of the lower-level nodes where grid power may not be present. The gateway devices may themselves form a peer-to-peer network operating at higher bandwidth and lower latency than other nodes, thus attracting and funneling network traffic.

Software for Wireless Sensor Networks

In addition to requiring new hardware platforms, wireless sensor networks require a new software framework. Traditional network abstractions are usually not suitable for wireless sensor networks. The embedded software must have precise control over the hardware and be able to implement application-specific networking protocols. Furthermore, wireless sensor network operating systems must allow high-level applications direct and efficient control over low-level hardware when necessary.

In TinyOS—an operating system designed for wireless sensor networks—a specialized component model, which is enabled by advanced compiler technology, is used to simultaneously provide efficiency and reliability. These same concepts are now being brought into more traditional operating systems for use in gateway class and high-bandwidth nodes.

One good example of where direct, fine-grained control over the underlying hardware is needed is in power saving. Applications may need to precisely activate an 18-volt power source for a vibration sensor or turn off power to a memory system immediately after the data has been transmitted, or the applications may directly access a radio RSSI to make routing decisions. The traditional layered abstraction for both network and sensor stacks leads to inefficiencies. This requirement creates a significant challenge in balancing precise hardware control with ease of general-purpose programming and efficiency. Recent research suggests a common approach to solving this challenge across the range of platforms that involves three elements:

• A general-purpose component framework that eliminates the traditional layering abstractions

• Exposing hardware functions to applications and/or middleware

• Using virtualization and/or interpreted scripts to write sensor network applications

Mesh Networking

One of the critical requirements for robust, low-power sensor mesh networks is the routing protocol stack. The TinyOS operating system in its mote development covers the whole range of layers of the OSI reference model for networks. Because TinyOS is an open-source operating system, all of its layers can be modified to meet the needs of an application.

A flexible mesh networking protocol in TinyOS called XMesh has been developed to provide the mote software developer with a diverse set of networking features. The XMesh protocol stack is an open-architecture, flexible and powerful embedded wireless networking and control platform built to support TinyOS applications. XMesh combines proven high performance with interoperability through the support of ZigBee v1.0 protocol (Figure 3).

The XMesh stack dynamically forms a reliable mesh network between nodes using proven ad hoc routing techniques. XMesh’s routing techniques are based on research done within the TinyOS community that characterized and tested many ad hoc multi-hop protocols and strategies for performance on Crossbow’s mote platform. The minimum-transmission, minimum-cost technology optimally reduces the total number of radio messages throughout the network, thereby maximizing the bandwidth and battery-life, while reducing network traffic and latency. Even without the use of any of its advanced QoS features, XMesh forms a reliable deterministic network, and the performance is shown to be superior to other techniques including shortest-part, DSDV, AODV and other proprietary routing schemes. Low-power mesh networking is a key feature of XMesh.

Sensor networks today can be used to deploy data collection networks that can operate for years without any maintenance. There are multiple classes of hardware devices in sensor networks available today that continue to evolve in capabilities, cost and size with technological advances. Integral to the performance of the sensor network nodes is the software that supports them. Sensor network applications require precise control over the underlying hardware in order to meet the strict power limitations placed on them.

Current development efforts are centering around two software platforms for use in wireless sensor networks. TinyOS provides the precise, efficient, low-level control demanded by general and special-purpose networking nodes. In contrast, special kernel modifications have been added to Linux to allow it to support gateway and high-bandwidth class device operation. Combined with the hardware platforms, TinyOS and embedded Linux are being shaped into a powerful toolbox for building wireless sensor network applications.

Crossbow Technology
San Jose, CA.
(408) 965-3300.