The radio spectra are increasingly crowded and confused, and smart devices are making it worse. Already jammed full of telecom signals, the airwaves have become increasingly congested with the popularity of the Internet of Things.
Naturally, this has battlefield implications.
“Is this signal enemy or friendly or some unaffiliated third party? Is my radio being interfered with by a malicious actor, or is there some uninformed third party who didn’t know the spectrum plan?” said Paul Tilghman, a program manager in DARPA’s microsystems technology office.
“We need to understand what is going on in the spectrum and what function that spectrum is being used for in this context.”
That office has an initiative underway to apply to the techniques of artificial intelligence and machine learning to bear on the issue. The agency held a preliminary spectrum collaboration challenge event in December and competing teams are slated to begin testing in March for an event later this year. The event has the potential for $3.5 million in prizes.
The spectrum challenge
In the past, hardware-based radios have featured fixed functionality and operated on allotted spectra. That meant military leaders could tell at a glance what was happening over the airwaves. “Simply by looking up the frequency, that would give you all the information you needed,” Tilghman said.
A new generation of software-defined radios has muddled the picture. “We went from a world with real clear boundaries defined by hardware to a place where people can buy a software-defined radio and cook up a whole new waveform that no one has ever seen before. That makes it challenging to understand what’s happening the spectrum,” Tilghman said.
At the same time, the rise of autonomous devices and IoT connectivity has made the radio spectra increasingly crowded, adding a further level of obfuscation for those trying to interpret electromagnetic activity.
All this has an impact on ISR operations, which may depend on the ability to clearly interpret radio activity.
“The growing demand for bandwidth has sparked increased discussions in the microwave remote sensing community of how to respond to this crowded spectrum environment and how to deal with the consequent issues of radio frequency interference,” researchers Michael Spencer and Fawwaz Ulaby write in the IEEE journal Geoscience and Remote Sensing.
Thus far the radio community has applied manual fixes to deal with the problem. “We have operators with specific systems trying to look at and interpret the spectrum,” Tilghman said. “Because we have a small number of highly specialized systems and a small number of operators, we’re looking at the world through a soda straw.”
Now DARPA is looking at artificial intelligence as a possible remedy.
The AI fix
As with so many emerging AI applications, the object here is to get the machine to perform – quickly, repeatedly and accurately – a task that would be too time-consuming for a human to effectively manage.
In a general sense, DARPA would like to see solutions in which AI-driven software could comb through the thicket of signal, winnowing out those blips that don’t matter and highlighting any activity that merit a deeper look.
“We can think of machine learning as a force multiplier,” Tilghman said. “If I can go through and pick out lots of signals that are not important, I can take those off the operator’s plate so they can focus on the signals that are most important.”
The machines will have to be smart enough to recognize familiar or predictable sources of radio signal, and also savvy enough to notice signal that is unusual or unexpected. That ability to spot anomalous behavior will be key to the ISR mission, as intel analysts seek to tease out the subtle threads of enemy activity from within the dense weave of legitimate signal.
Interestingly, NASA is pursuing a similar line with its investigation into “cognitive radios,” a project aimed at leveraging AI to maintain radio contact with distant spacecraft. “Modern space communications systems use complex software to support science and exploration missions,” said principal investigator Janette C. Briones in a NASA press release. “By applying artificial intelligence and machine learning, satellites control these systems seamlessly, making real-time decisions without awaiting instruction.”
While DARPA began looking for AI fix to the spectrum problem, the state of the art in industry is still “quite nascent,” but momentum is building, Tilghman said. “We have gone from engaging with the community to getting real thoughts about how this problem set is unique.”
The telecom industry and others who rely on spectrum availability are at least as eager as their military counterparts to find better spectrum management tools. In this case the military may be able to leverage the commercial world’s own interest.
“We all have a need to understand what is happening in our spectrum. I might have tactical reasons for wanting to understand how an enemy is impinging on my spectra, whereas a commercial carrier might want to know whether someone is impinging on their spectra, just because they have paid a lot of money for that spectra,” he said. “There is a real commercial interest to this as well.”
