Censoring Sensors
Written by Karen E. Thuermer
Sensor networking is an emerging technology that promises unprecedented ability to monitor and manipulate the physical world. It works via a spatially distributed network of small and inexpensive wireless sensor nodes that have the ability to self-organize into a well-connected network. The nodes can sense (and potentially actuate) the physical environment in a variety of modalities, including acoustic, seismic, thermal and infrared.
Blum directs Lehigh’s Signal Processing and Communication Research Lab. His work has caught the attention of the government, which recently awarded him with grants from four agencies to find ways to save energy in networks of remote sensing devices. Grants from the National Science Foundation and the Army Research Office will support work in sensor networks.
Grants from the Office of Naval Research and the Air Force Office of Scientific Research will support his energy-saving techniques in a new type of radar system called MIMO radar. MIMO (multiple-input and multiple-output) communications systems improve performance by using multiple antennae to transmit and receive radio signals.
MIMO principles have been applied only recently to radar, Blum noted. They have shown promise in overcoming the shortcomings of traditional radar systems, however. The older systems determine the distance and location of a target by sending signals from one point out into space and measuring the time required for the signal to bounce off the target and return.
Several factors can reduce such a system’s reliability, Blum says. The angle of the target might shift, weakening the returning signal. The outgoing signal might hit several objects and diminish, or it might miss the object being tracked. “With regular radar, you often have to ‘see’ the target from just the right direction in order to get an accurate picture of the target,” Blum said. “This is due to the fact that the magnitude of the return signal can change drastically because of a small change in the angle of the target.”
MIMO overcomes these obstacles by transmitting radio signals from several locations. “If you see from multiple viewpoints, you are going to get a much better picture,” Blum said. “One or two illuminations might come back weak, but not all of them. And the antennae in a MIMO radar system do not need to be very far apart in order to get a usable return signal.”
MULTIPLE USES
Already sensor networks are increasingly playing a bigger role in everyday life. For example, sensors can be used to detect and process the signals caused by the stresses from passing vehicles on a bridge, and then transmit that data to a central decision point. The data helps engineers determine more quickly and accurately the likelihood that a bridge will fail, such as in the case of the collapsed bridge in Minnesota two years ago. Sensors can also be used to detect the presence of radioactive or other toxic waste in the environment, and can determine whether or not a cell is malignant. They are used in military surveillance, air traffic control, home automation and other applications.
While the practically unlimited range of applications of sensor networks is evident, scientists contend that the current understanding of sensor network design and management is far from complete. For example, from a communication or information routing perspective, wireless sensor networks are similar to ad hoc wireless communication networks that have been studied more extensively.
However, information flow in a sensor network is fundamentally related to the distributed data collected by the sensor nodes. Consequently, the basic operations of sensing, processing (computation) and communication are interdependent and must, in general, be jointly optimized for efficient use of limited node resources, such as available energy, computational power, and communication bandwidth. Thus, sensor networks augment the channel-centric view of communication networks with a source-centric view and present a rich array of research challenges in distributed information sensing, processing, communication and routing that must be met to realize their promise.
The Army requires high resolution, high sensitivity, but also affordable, multi-hyperspectral and polarization-sensitive, active and passive, IR sensors for target acquisition, recognition and identification in the digital battlefield. The sensing of vehicles, personnel, chemicals, landmines and biological agents is critical for battlespace deconfliction, and new or improved sensing methods to increase battlespace situational awareness are needed. Sensing technologies currently include acoustic, seismic, radar (RF to millimeter wave) and passive electromagnetic, and hyperspectral.
“Today the military is interested in autonomous vehicles that carry their own power and sensors,” said Blum. “In order to allow these autonomous vehicles to operate for the desired mission lengths, it is important to use this power as sparingly as possible.”
The Institute of Electrical and Electronics Engineers (IEEE), the world’s leading professional association for the advancement of technology, reiterates the great interest surrounding joint signal processing and communication design of such networks.
Blum, along with co-author Brain M. Sadler of the Army Research Laboratory, addressed this interest and technological advances in this approach in the July 2008 issue of IEEE Transactions on Signal Processing. In that article, titled “Energy Efficient Signal Detection in Sensor Networks Using Ordered Transmissions,” the authors comment that new approaches were considered where transmissions can be ordered and altered when sufficient evidence is accumulated.
“We demonstrate that these approaches require, on average, fewer sensor transmissions and that the savings can be significant in cases of interest,” they wrote. “We describe a highly efficient approach, which saves transmissions over either the optimum unconstrained energy approach or censoring while achieving the same error probability as these approaches.”
The article points out that the average number of transmissions saved (ANTS) is lower, bounded by a quantity proportional to the number of sensors employed, provided a well-behaved distance measure between the sensor distributions is sufficiently large. For such cases, the ANTS over the optimum unconstrained energy approach is shown to be larger than half the number of sensors employed. Numerical results for a meanshift hypothesis testing problem with Gaussian noise show significant savings even for smaller values of the distance measure.
“Researchers have made a convincing case for a mode of operation called censoring, where sensors transmit only highly informative observations,” Blum wrote. “These sensors transmit only for very large or small likelihood ratios. We refer to this approach as per-sensor censoring.”
GREEN TECHNOLOGY
While most sensor networks transit data wirelessly and are powered by batteries that use large amounts of energy, research being conducted by Blum is introducing significant advancements that allow sensor networks to cut energy consumption without loss in performance. “Our analytical proofs have shown that we can cut energy consumption by more than half with no loss of performance,“ he said.
This makes the technology energy efficient and significantly “green.” By using technology called “ordering,” Blum has been able to show how significant energy can be saved by only having those autonomous vehicles with the most informative data transit this data to the decision point, often called a fusion center. “The majority of the dissipated energy is typically devoted to data transmissions, since the fusion center is typically far from each autonomous vehicle,” he said. “Thus, these savings are very significant.”
By using this approach, Blum and his researchers have been able to prove that the same target detection can be performed as if all the vehicles employed transmit their data.
“We have also proven that over half the vehicles employed will not need to transmit, on average, even while the same performance is maintained as if all the vehicles employed transmit their data,” Blum said. The key to making this work is in ordering the transmissions. “Each autonomous vehicle computes the likelihood of a target being present at a particular location based on its observed data,” he said. “Then it sets a timer which is inversely proportional to the magnitude of the log of this likelihood. After the timer counts down to zero, the autonomous vehicle will transmit its data.”
Thus, highly informative data will be transmitted first while autonomous vehicles with less informative data will wait to transmit. According to Blum, the fusion center will accumulate the sum of the loglikelihoods from those autonomous vehicles that have transmitted their data. When the transmitted data is overwhelming for either target present or absent, the fusion center will halt all transmissions.
Saving the energy is extremely important in these military operations so that batteries are not depleted too quickly. “We are also investigating using these ideas to increase the life of sensor networks used to monitor the health of large structures, like buildings and bridges,” he added. ♦
