Fingerprints of Power

Soon after terrorists flew hijacked airliners into World Trade Center and the Pentagon, a steady stream of videos featuring al-Qaida mastermind Osama bin Laden excoriating the West came pouring out of … somewhere.

For nearly a decade, until the operation in Pakistan that ended in his death, intelligence analysts combed over the videos—some shot indoors, others apparently filmed in the mouth of a cave—for visual clues to his whereabouts.

Min Wu, a Maryland electrical engineering professor who specializes in forensic analysis of electronic media signals, suspected that virtual “fingerprints” were embedded in the videos, flowing directly from the electric power grids in the areas at least some of the videos were recorded.

“We were asking, by looking at these videos, can we determine the likely location where he is making his propaganda videos?” she says.

At the time, the answer was an unqualified no. But Wu has since developed a method that’s increasing the possibility of looking at a video and knowing precisely where in the world it was made.

Her work tracking the unique signals put out by electrical power grids earned Wu and her Ph.D. students Adi Hajj-Ahmad and Hui Su the 2015 UMD Invention of the Year Award. She received the award in 2012 as well, for a method to track times and locations of digital recordings, which can help uncover tampering.

The technique for both projects relies on fluctuations in something called the “mains frequency” continuously pumped out by power grids.

“It’s an environmental sound … the power hum,” she says. “The frequency goes slightly up and down, depending on the load placed on the grid.”

In recent years, police and other agencies have begun to record this electronic network frequency (ENF) for forensics purposes. A common law enforcement use: comparing the subtle electrical hum in audio recordings to the ups and downs of a recorded ENF signal to establish that they haven’t been edited or falsified.

A pioneering Romanian researcher in ENF analysis, Catalin Grigoras, demonstrated that the extracted ENF signal, among other uses, shows roughly where a recording was made by analyzing frequencies from different electrical grids—such as the twin grids that serve the eastern and western halves of the United States.

His ENF studies were misunderstood and scoffed at first, says Grigoras, now director of the National Center for Media Forensics at the University of Colorado Denver.

“In 2000 and 2001, I wanted to present it at international conferences, but my proposals were rejected, and people laughed at me again,” he says. But now that the technique is accepted, “some of the persons that laughed and called me ‘the crazy Romanian’ were later fair enough to come to me and say sorry, and to ask for more details about it.”

The technique for which Wu and her students received the 2012 UMD award builds on that foundation, and the 2015 invention pushes the technique further into the visual realm by analyzing light sources within videos, which can be correlated to ENF records to establish times and locations. Besides forensics and security, the technique will enable what Wu called “immersive digital experiences” in virtual reality and other technologies.

The technique takes creative advantage of newer video technology. Unlike older cameras, which had “global” shutters that exposed an entire frame of video concurrently, newer cameras generally use a “rolling shutter” mechanism that exposes the light sensor in tiny strips.

“The benefit is that it gives us many more data points to work with,” she says.

That’s because a global video shutter gives a single reference for light levels within the scene perhaps 30 times a second, but a rolling shutter gives new data every time another strip of pixels is exposed. That means potentially 1,000 times more information on varying light output based on grid fluctuations.

There are limitations, Wu says. For one, such analysis requires the presence of an electrical grid and an indoor or outdoor electrical lighting source—meaning some of the bin Laden cave videos are still beyond reach of the technique. It also relies on a recording of the ENF signal. But Wu is pursuing workarounds.

“We are always working on greater capability,” she says.

Among those is the ability to zero in on locations in much more detail. Simply knowing which power grid a recording came from isn’t necessarily useful when, in the case of the United States, that’s an area of more than a million square miles.

But recordings made at different locations within the grid can have microvariations in frequency, allowing triangulation to locate a recording within 50 miles or so. It’s a degree of precision she hopes to improve on, Wu says.

Grigoras says Wu’s work is helping expand the forensic use of power grid signals.

“Min Wu’s work on ENF is brilliant,” he says. “It made me feel comfortable seeing that I wasn’t the only ‘crazy’ guy dreaming of ENF.”