Faculty & Staff

Our Mining Engineering students have the opportunity to practice their mining skills in a real mine. Learn more from the Department of Mining and Geological Engineering. go there now »

world-class research | successful students | distinguished alumni

Faculty & Staff
>> home ... here
UA Engineering Faculty & Staff: Dedicated to World-Class Research
By Ed Stiles, College of Engineering
February 11, 2008

Engineers at The University of Arizona are looking to bats as role models for advanced radar systems.

“Bats do exactly what we’re trying to do,” said Nathan Goodman, assistant professor of electrical and computer engineering. “They have cognitive sonar and we’re developing cognitive radar, which is in the electromagnetic spectrum, but the idea is almost exactly the same. We would love to build a system that does what bats can do.”

Goodman has received a three-year grant from the Air Force Office of Scientific Research to begin work on the mathematical framework and implementation issues of cognitive radar.

Although engineers are still a long way from developing radars with the sophisticated capabilities of bat sonar, looking at how bats function is a good way to understand what cognitive radar is all about.

Bats use different chirps as they scan for, identify and approach a target, explained Goodman, who is developing similar strategies for dynamically adapting waveforms in response to echo data returned from a target.

When hunting, bats first use a waveform that is well suited for wide surveillance of their environment. They then use other waveforms for refining their recognition of a possible target, as well as estimating its lcoation and speed.

Then there’s the terminal stage where the waveforms change again. “They don’t care about characterizing the object any longer because they know what it is – a moth, for instance,” Goodman said. “But they now want to know precisely where it’s at and how it’s moving.”

Narrowing the Possibilities

“We typically evaluate and modify the frequency content of our transmitted waveform,” Goodman said. “Different targets have different resonances, and that means waveforms will be reflected back more strongly at certain frequencies than at others.”

The radar could be set up initially to transmit a waveform with a fixed amount of power over a wide frequency range, but that would dilute the power transmitted at any one frequency, making the system less sensitive for tasks such as target identification, he said. The better way is to hypothesize what targets could be present, rate those hypotheses in terms of probability, know what their likely resonances are, and to transmit only at those frequencies.

“Suppose, for instance, that the radar knows a target is present and can assume that it’s one of four possible target types,” Goodman said. “Let’s label the target types as A, B, C and D. These four types each have known characteristics and the radar can customize a waveform that is tuned to those characteristics. But say the system receives echoes that show that A and B are unlikely. Then the system can adapt its next waveform to further tune to the remaining possibilities – C and D. This can result in a faster decision and can save transmit energy.”

Radar That Thinks

Intelligence is a key component of cognitive radar, Goodman explained. Rather than being hard-wired, it can change in response to its environment.

Cognitive radar needs to interpret the signals it receives in order to better understand its environment and modify subsequent transmissions to further refine the search and provide additional information, Goodman said. Cognitive radar’s intelligence is built on a statistical method called Bayesian analysis, he said.

The receiver has prior knowledge of what it’s looking for – perhaps a terrain map, likely targets to be found in that terrain, and a history of movements in that area, for instance. Then it forms a number of hypotheses about what may be out there based on the first transmitted signals.

“There are an infinite number of combinations that could be out there, Goodman explained. “So we have to come up with a strategy for identifying the most likely starting hypotheses and for assigning to those hypotheses a probability of being true. This probabilistic representation of the radar environment quantifies what the cognitive radar system does and does not know about the environment.”

“Bayesian theory also gives us a way of updating the probabilities of our various hypotheses,” he added. “Let’s say there are four alternatives and each one has a 25 percent chance of being true. Then we transmit a signal and get a return signal. If we’ve measured something significant, all these scenarios probably will no longer be 25 percent likely. The Bayesian equations provide a mathematical way of quantifying this change in understanding.”

Using Multiple Radar Systems

In addition to measuring its environment and changing its behavior in response to received data, cognitive radar also can use multiple radar systems to send out multiple waveforms. In this configuration, the systems work together to develop in-depth knowledge more quickly.

“The system decides not only how to shape the radar’s waveform in response to received data, but also whether to employ multiple radars, whether a mobile radar system should move to a better position, or whether to request help from a different sensor type,” Goodman said.

Currently, Goodman and his students are working with computer simulations. “Eventually, extracting data and properly analyzing it in a real-world situation will be a very challenging part of the research,” he said.

Amazingly, bats are able to do all this recognition and analysis with a brain that’s smaller than a golf ball. “I don’t think the bat is doing Bayesian equations in its head,” Goodman said. “But it is constantly updating what it thinks is true about its environment.”

“We want cognitive radar systems to be able to do the same kinds of things in the electromagnetic region and just as fast.”