Sensor Networks, New Face of Computing

Wireless sensor networks have rapidly emerged as a technology that can bring great benefits to all aspects of our life and society, ranging from home to environment and from healthcare to national security and defence. This has been possible due to a confluence of recent advances in wireless communication, VLSI design and micro-fabrications, and low-cost low-power embedded microprocessor systems. 

A sensor network typically consists of a large number of small, inexpensive sensor nodes, each endowed with sensing, computing, and communication capabilities.

These nodes talk to each other and/or to one or more servers where the data may be integrated, fused, analyzed and made available to other systems and human

operators. 

Sensor networks, thus, extend the Internet into the physical world. The sensors collect information about the world-it may be physical parameters such as temperature, humidity, water level and pressure, atmospheric moisture content, and soil motion and vibrations; or chemical sensors that assess the content of soil, water, and air; or it may be an imaging or video sensor; or it might just sense the presence and identity of the object. There is no inherent restriction on what property may be sensed-anything that is appropriate for the particular application at hand is valid. The power of sensor networks comes from the fact that a large number of sensors are coupled to a vast source of real-world knowledge. Ultimately, this technology has the potential to seamlessly merge the physical world and information world to bring the power of global knowledge for managing local events and precise local data to global analysis.

The potential benefits of sensor networks for our life and society are, thus, immense. In a country like India, which is prone to natural disasters such as floods, landslides, cyclonic storms and earthquakes, this technology offers a way to continuously monitor the environment and give us early warnings of impending catastrophic events. For instance, Microsoft Research India is collaborating with Prof U B Desai at IIT Bombay who is working on a project aimed at landslide detection. His team's goal is to monitor the earth in landslide prone areas to look for tiny vibrations and movements in the rock and soil over a large area. The data from a distributed area can be fused, integrated and analyzed together with known geographical and seismic properties of that area to predict impending landslides. Similarly, Prof B N Jain of IIT Delhi is exploring the use of sensor networks in the context of other natural events such as floods.

However, research in this field is still at a stage of infancy; to successfully realize these vast benefits, a core set of technical problems has to be solved. These include development of small, low cost sensors, appropriate communication systems and networking protocols, a programming environment for developing applications on the sensor networks, database and visualization capabilities that will exploit the data and provide the user access to the information in a relevant and succinct fashion. 

The development of small, inexpensive sensors, with reasonable computing and communication capabilities has been a major driver of this research area. As in the case of other electronics and VLSI technology, there have been continuous improvements in the performance and power of sensor nodes, while the cost and size of the units have reduced.

One of the most vibrant research areas today relates to communication in wireless sensor networks. Many traditional communication protocols used in wireless and cellular networking do not automatically apply to sensor networks for several reasons. For instance, sensor nodes have limited power and energy capacity which limit the distance over which each node may be able to transmit and receive information. Traditional methods used in cellular and other networks that use a base station and dedicated resource assignment strategies will not be suitable because there is usually no single central controlling node. A natural approach may be to consider ad-hoc mesh networking methods. However, sensors are not often robust or rugged and in many cases may be lost or destroyed by natural forces. 



This has to be factored into the communication protocols. Various protocols are now being developed for communication and routing of information in sensor networks, of which many are adaptations of existing wireless and cellular protocols that handle the specific energy, power, noise and reliability issues associated with sensor networks.

Images of the very compact Berkeley Motes, small sensor devices, developed by Berkeley University

Another important aspect of sensor network research relates to programming and developing sensor network applications. One of the more widely used environments is TinyOS, an event-based operating environment originally developed by University of California, Berkeley. It is designed to support the concurrency -intensive operations typical of sensor nodes. Programming a network of sensors is inherently challenging, because unlike existing platforms, systems such as sensor networks are decentralized, embedded in the physical world, and interact with people. In addition the programming environments have to account for node uncertainty, power and energy constraints and even for inter node collaboration. Research on this topic is still in its early stages. While there are some tools aimed at programming individual nodes, the development of large-scale applications that run across a network of such nodes require new programming paradigms, tools and platforms. A few groups are exploring the issue of how to program sensor networks. For example, the networked-embedded computing group at Microsoft Research, Redmond USA (led by Dr Feng Zhao) is developing new architectures, models and tools for organizing and programming these systems.

Ultimately, to realize the power of sensor networks, we need the fusion of integration of the sensed data together with contextual knowledge. The existing database and storage technologies require considerable adaptation to handle sensor network scenarios where data is inherently distributed and changing dynamically in an unpredictable and ad-hoc fashion. The information needed will extend beyond the information available at single nodes or even a set of nodes as data from other geographically distributed sensors will have to be integrated and fused. Queries will be at a much higher level than the data and will have to be parsed into queries about a diverse variety of sensors. The data analysis, fusion, and integration algorithms and the database will have to be designed together and the entire system has to function robustly, given unreliable nodes and ad-hoc connectivity. This poses new research challenges that extend beyond conventional database design.

Despite the challenges, the field of sensor networks is one of the most exciting emerging research areas. These networks have vast potential to revolutionize the way we monitor and live in our environments. However, the discipline is still young, the technologies are still fresh, and it may be some time before we can realize the full benefits of this new and emerging computing paradigm.

Dr P Anandan, MD, Microsoft Research India