This Is The Most Important Technology On the F-35

The F-35 Joint Strike Fighter, the most expensive weapons program ever, won’t justify its price tag by outmaneuvering other jets (it can’t), flying particularly fast, or even by carrying munitions in a stealthy bomb bay. Instead, the U.S. military is banking on an emerging technology called cognitive electronic warfare to give the jet an almost-living ability to sniff out new hard-to-detect air defenses and invent ways to foil them on the fly.

While the specifics of the jet’s electronic warfare, or EW, package remain opaque, scientists, program watchers and military leaders close to the program say it will be key to the jet’s evolution and its survival against the future’s most advanced airplane-killing technology. In short, cognitive EW is the most important feature on the world’s most sophisticated warplane.

“There are small elements of cognitive EW right now on the F-35, but what we are really looking toward is the future,” Lee Venturino, president and CEO of First Principles, a company that is analyzing the F-35 for the Pentagon, said at a recent Association of Old Crows event in Washington, D.C. “Think of it as a stair-stepper approach. The first step is probably along the ESM [electronic support measures] side. How do I just identify the signals I’ve never seen before?”

To understand what cognitive warfare is, you have to know what it isn’t. EW makes use of the invisible waves of energy that propagate through free space from the movement of electrons, the electromagnetic spectrum. Conventional radar systems generally use fixed waveforms, making them easy to spot, learn about, and develop tactics against. But newer digitally programmable radars can generate never-before-seen waveforms, making them harder to defeat.

A concern that U.S. EW was falling behind the challenges of today's world prompted a 2013 Defense Science Board study that recommended that the military develop agile and adaptive electronic warfare systems that could detect and counter tricky new sensors.

“In the past, what would happen is you’d send out your EA-18,” the military’s top-of-the-line EW aircraft, Deputy Defense Secretary Bob Work said last month in an event at the Center for New American Security. “It would find a new waveform. There was no way for us to do anything about it. The pilot would come back, they would talk about it, they’d replicate it, they’d emulate it. It would go into the ‘ gonculator ,’ goncu-goncu-goncu-gonculatoring, and then you would have something, and then maybe some time down the road, you would have a response.”

That process is far too slow to be effective against digitally programmable radars. “The software [to defeat new waveforms] may take on the order of months or years, but the effectiveness needs to programed within hours or seconds. If it’s an interaction with a radar and a jammer, for example, sometime it’s a microsecond,” said Robert Stein, who co-chaired the Defense Science Board study.

Read “interaction” in that context to mean the critical moment when an adversary, perhaps a single lowly radar operator, detects a U.S. military aircraft on a covert operation. That moment of detection is the sort of world-changing event that happens, literally, in the blink of an eye.

Just before the study came out, the Defense Advanced Research Projects Agency, or DARPA, established the Adaptive Radar Countermeasures program to “enable U.S. airborne EW systems to automatically generate effective countermeasures against new, unknown and adaptive radars in real-time in the field.”

The goal: EW software that can perceive new waveforms and attacks as quickly and as clearly as a living being can hear leaves rustle or see a predator crouching in the distance, then respond creatively to the threat: can I outrun that? Can I fight it? Should I do anything at all? It’s a problem of artificial intelligence: creating a living intelligence in code.

Applying the Brain Algorithm to EW Warfare

There could be no cognitive electromagnetic warfare without cognitive radar, a concept fathered by electronics researcher Simon Haykin in his prescient 2006 paper “Cognitive Radar: A Way Of the Future.”

Cognition is an act we attribute to living things, defined in the Oxford Dictionary as “knowing, perceiving, or conceiving as an act.” Haykin suggests that echo-location, which allows bats with nut-sized brains to detect, identify, and engage targets, is a type of cognition built on deep information processing. “How then does the bat perform all these remarkable tasks? The answer to this fundamental question lies in the fact that soon after birth, the bat uses its innate hard-wired brain to build up rules of behavior through what we usually refer to as experience, hence the remarkable ability of the bat for echo-location.”

Haykin proposed that it’s becoming feasible to build a bat-like computer, thanks to radars and phased-array antennas that allow rapid scanning of waveforms and ever-shrinking, ever-more powerful computers.

At the root of this cognitive processing is Bayes’s theorem:

P in the theorem means probability. A is the answer and X is a condition that will influence the probability. Thomas Bayes published the theorem in 1764, but it’s only in recent decades that it’s gained real popularity among statisticians, computer scientists, and machine learning experts. Bayesian algorithms don’t necessarily provide the most accurate answer the first time you use it. But as new information and data become available, you run the formula over and over again to get answers in which you can have more and more confidence.

The advent of the Network Age, with its massive amounts of continually streaming data, has made Bayesian analysis more useful than some more traditional types of statistical analysis, especially for helping machines to learn. The human brain, too, learns both imperfectly but continually on the basis of streaming stimuli, as opposed to outputting a single value after crunching a big package of information.

Applied to radar, Haykin imagined a Bayesian algorithm working like this:

“For a given search area, radar returns are collected over a certain period of time. 2) For each range-azimuth resolution cell in the search space, the probability that the cell contains a target is computed. 3) With the evolution of target probability distribution resulting from the recursive computation of step 2 over time, target tracks are detected, and corresponding hard decisions on possible targets are subsequently made.”

Haykin’s paper helped spark the Defense Department’s interest in cognitive EW and machine learning. BAE Systems and Raytheon are among the defense contractors that have emerged as key players. Today, Bayesian statistical methods are at the core of virtually every effort to apply machine learning to EW.

“I would say, generally, Bayesian algorithms are a core to machine learning and we certainly apply them across a wide range of domains that we operate in,” said Josh Niedzwiecki, who directs BAE’s sensor processing and exploitation group.

BAE provides the F-35’s EW package.

Niedzwiecki’s 200-person group includes PhDs from top universities with backgrounds in machine learning, physics, statistical signal processing, and computational neuroscience among other fields, all working to apply machine learning algorithms to radar energy, video image processing, acoustic signal processing, and more. “They understand how the brain works, how we learn,” Niedzwiecki says of the group. Bayesian statistical methods are the foundation of all of that.

But machine learning algorithms can’t learn without data, lots of it. While Facebook can access records from a billion-plus users, getting data from adversaries about the unique waveforms that they’re experimenting with is a more challenging task. The military can’t just ask China to opt-in to an information-sharing agreement.

Generally, the best information is gathered on real-world missions, but this has its limitations. “There are certain tactical scenarios where that becomes very difficult because my mission might preclude me from hanging around for very long. I might be in a platform or in a mission scenario where I have to get in and get out,” says Niedzwiecki. “The way you take advantage of that is to learn over time. So I’m recording this data, I’m building my model, and given the data that I’m seeing and the hypothesis I’m testing during that mission, I’m seeing something about how to change the model to be more accurate next time. I want to take that data and use that for the current mission and the next mission. Those are some of the things that are starting to be thought about.”

For the United States, EW dominance will be a matter not just of designing more exquisite sensors or writing smarter algorithms. It will require the disciplined execution of data collection processes -- something that has to happen military-wide every time a radar operator encounters a new waveform, but doesn’t, the Defense Science Board study found. “In those places where we do have recorders, operators tend to turn them off. Because sometimes they create issues with the equipment with which they’re embedded,” said Stein. But, he continued, “last night, in some conflict, some place, unexpected things happened. What are we going to do about it? We better have the tapes, the digits, that recorded what went on last night. Let’s peel it apart. Let’s see why what happened, happened. We tend not to do that.”

When F-35 pilots have to slip past the programmable radars of the future, their success is going to depend on a lot of data collection that happens off the plane.

The EW Arms Race

For a peek at the future of plane-killing technology that the F-35 may go up against, look at the Nebo-M, Russia’s premiere programmable radar system. The Nebo-M consists of three radars on separate trucks: a VHF that does the wide scanning and higher frequency UHF and X-Band that do the more precise triangulation. The system fuses the data from these three data streams to draw a bead on even stealthy aircraft.

“The radar is designed to automatically detect and track airborne targets such as ballistic missiles, stealth aircraft, or drones, as well as hypersonic targets. In the circular scan mode the complex is able to track up to 200 aerodynamic targets at a distance and at altitudes of up to 600 kilometers. In sector scan mode, Nebo-M can track to 20 ballistic targets at ranges of up to 1,800 kilometers and at an altitude of up to 1,200 kilometers,” Russian-State media outlet RT claimed back in February. The Russian military planners in October to extend radar coverage across the entirety of Russia by 2020, according to RT.