Background

Traditionally, sensor systems are designed using a networking scheme called a star topology (Figure 1), in which measurements from sensors across the network are transmitted directly from the sensor to a central collection point. The data is then processed at this central point and distributed to the end users. This approach works well with networks consisting of a small number of sensors; however, significant resource scaling issues exist using a star topology with large networks. For this reason, Los Alamos National Laboratory (LANL) has been developing techniques for distributed sensor networks with collective computation (DSN-CC).

Figure 1.

DSN-CC represents an innovative approach to in situ sensing. DSN-CCs are networks of sensors capable of intercommunicating and executing complex computations in the field without central control (Figure 2). This method of computation and communication has demonstrated clear advantages over traditional approaches where sensors simply collect and transmit data out of the network.

Figure 2.

Unlike a traditional array, there is no central (and vulnerable) control point to which all data must be passed. In the DSN-CC, information and conclusions exist throughout the network. This avoids several problems encountered by the traditional approach. Classically, raw data is transmitted for long distances, requiring large amounts of power and bandwidth. Central processing incurs delays and leaves the system vulnerable to single-point failures. In contrast, DSN-CCs operate only with short-range transmission and in situ data processing. This approach saves communication bandwidth, provides redundancy, eliminates single-point failures, and delivers conclusions rapidly to users.

Figure 3.
DSN-CC nodes communicate with neighboring nodes in order to exchange measurements and cooperatively solve a sensing problem or to cue a different sensor. A concise conclusion or cue request is rapidly propagated across the network. Users of the network can obtain this information by listening to any portion of the network. This concept of in situ computation has seen little exploration, but allows for capabilities not possible with traditional sensor arrays due to reliability issues, prohibitive cost, scaling of power, computing, and bandwidth usage. Small, low power sensors (Figure 3) with such functionality are now emerging from companies such as Crossbow Technology.

Efforts at LANL are focusing on developing DSN-CC systems in both simulation as well as in hardware using commercial off-the-shelf platforms. To this end, LANL has developed an open source simulation engine for base-lining application-specific networks. Upgrades to the simulator are ongoing as hardware platforms evolve and emerge. Additionally, by focusing on particular applications LANL’s DSN-CC team has been able to move from simulations to hardware systems. Those applications are described in our most recent publications. The ongoing goal of this project continues to focus on the demonstration of in situ collective computation abilities of networks using inexpensive, readily available off-the-shelf technology and hardware.