 Imagine having complete understanding of what's inside a building without ever having to step inside it, without ever having to open a window or open a door. That's the kind of thing that scientists and engineers at the U.S. Army's Research Laboratory are working on using research in low-frequency imaging sensors and supercomputers. Hi, I'm Dr. Phil Percanti, the Director of the U.S. Army Combat Capabilities Development Commands, Army Research Laboratory. Welcome to ARL, what we learned today. A podcast where we talk with Army scientists and engineers about the science and technology that will modernize the United States Army. Today we're going to talk with Dr. Ross Adelman, a computer scientist in the lab's sensors and electron devices directorate at our Adelphi Laboratory Center in Maryland. Ross, welcome to the podcast. Thank you. Tell me a little bit about yourself. So I've been here for 13 years. I started as a high school intern and I worked here every summer and winter break through my schooling and after I finished my Ph.D. I joined ARL as a full-time employee. Well, that's really cool. So you've been exposed to the Army for a good long time. Yeah, a very long time. And you've been exposed to science, technology, engineering and mathematics, which is one of the things that the Army Research Laboratory does to promote science by working with high school students in the summer and on breaks to really not only get them interested in science, technology, engineering and math, but also expose them to the United States Army, the problems the Army faces, get some interaction with soldiers. So that one day when you graduate from university with a Ph.D., you'll come back and you'll work here full-time. So you're a success story for that model, Ross. It's really cool. Where'd you go to college? I went to Carnegie Mellon in Pittsburgh for my undergrad and the University of Maryland College Park for my grad school. Oh, where'd you study at Carnegie Mellon? I studied electrical engineering and at Maryland I studied computer science. Now I wonder what you wrote on your college application to get accepted to Carnegie Mellon. Did you talk about the fact that you worked at the U.S. Army Research Laboratory at all? heavily. See that? I was just guessing. But I knew that was the case. That's awesome. That's really awesome because CMU is a great school. We have a wonderful partnership with them. And you said you went to University of Maryland as well. Of course, they're our neighbor. We have a very deep relationship, collaborative relationship with the University of Maryland on many technology areas. So it's great to see that that relationship is paying off as well. Tell me a little bit about what you studied for your Ph.D. So I studied scientific computing and high-performance computing. So using software to solve hard scientific problems. Specifically, I looked at simulating electric and magnetic fields and how they can be exploited to learn about an environment. Did you always think you would wind up working for the Army? I think so, yes. I've been with the same team leader, the same mentor my entire time here. And he's been excellent at helping me understand the problems the Army is facing and helping me at a personal level with schooling and that kind of thing. When I finished my schooling, it was basically an easy decision to come to work there. So Ross, tell me a little bit about what you're working on here at the Army Research Laboratory. So I use computer modeling and simulation running on high-performance computing assets like supercomputers to solve very large scientific problems that are interesting to the Army. Why do you need a supercomputer to solve the things you're working on? So the problems we're looking at are very complex. They involve a lot of different pieces with very complex shapes, material properties and to really accurately model and simulate the effects that these objects have. You need very large and detailed models. And these models take up a lot of space and solving them requires a lot of computational capability to solve. So you focus on the high-performance computing side. What's the physical side of the problem? Is it infrared sensing? Is it RF sensing? What kind of information are you developing models for on these supercomputers? So the team I work on looks at extremely low frequency and very low frequency electric and magnetic fields to understand the environment that soldiers are in. This includes looking into buildings, looking underground, looking into containers, using non-contact electric and magnetic field sensors. Okay, so now I can start to see the reason why you need a supercomputer to model this. So you've got a very dynamic environment like a building which has concrete and has metal infrastructure, perhaps, metal beams. It has wires running through it. It's got people inside the building. It's got walls. It's got furniture. It's got all of these components that make up what would naturally be inside a building. And you're trying to use a supercomputer to model that and then to really see what the electromagnetic footprint is within that building, whether it's people or machines maybe or some other piece of technology without having to look through a window. Yeah, the problems are very complex. And like you mentioned, there's a lot of features in these problems and to really accurately model them and to study their behaviors and then to ultimately exploit them with sensors, we need to model them with very high detail. Yeah, so people are working on what is called a through-wall radar. Those radars are contact radars. You have to actually go up to the building you're trying to see inside of and put the radar right up against the wall. Or as you go out, try to get any space between the radar and the building you try to see in that it becomes a signal problem and you can't get the penetration into the building with the further out you go and range. Your work, though, is in low-frequency RF. So that gives you the penetration depth, right? Yeah, there's always a trade-off depending on what frequencies you're using and a benefit of using very low-frequency electric and magnetic fields is that you lose some resolution but you can penetrate much deeper into the building. Potentially, you could see interior rooms, what people are doing inside those rooms even with thick metal walls. When you say you lose resolution, what that really says is that you may get a blurry image. Correct. You may get a blurry image but you can see more of the scene than other methods may provide. So you're developing a context of what's happening inside the building. You potentially could provide a situational understanding of what's going on inside a building and you probably would need to use another couple of sensors and combine those sensors in a way that makes sense to give a full sight picture, if you will, of all the activity in a building. Absolutely. This would only be one piece of the puzzle. There would be multiple sensor suites operating to give a complete picture of what's in the building so that when a soldier goes in to clear a building or clear a room, before they enter the room, because it's a very dangerous situation, they want to know how many people are in the room, what kind of weapons they might have, are there women and children who you don't want to engage with? What's really hard about this problem? The first is actually designing the numerical software. There's a lot of mathematical complexity in the software, designing the algorithms to scale well on supercomputers and use memory efficiently. The second part that's difficult is actually constructing the models that you want to simulate in the first place, collecting dimensional data of a building that you want to do run tests on, building the different pieces in the model, meshing these models to feed into these simulation software. This all requires a lot of manual effort and a lot of time on the part of these modelers. Something we're looking into is automating these steps so that you can automatically collect dimensional data from different sources, mash them together, automatically fit surfaces to these data, and create these models. To me that sounds like an engineering challenge, more than anything, right? You're collecting information about a building, you can do that today, it's well known how to do, it's just labor intensive. Correct. Where's the science? So in the first part, developing the solvers, there's a lot of theory behind actually writing the code. It's not just writing software code, it's actually organizing the math into a structure that you can then implement. This is to make it run efficiently on a supercomputer. Correct, so it's fundamental computer science problems of developing fast algorithms that are efficient on a supercomputer. Yeah, and we should say a little bit about what's different between a supercomputer and say the laptop on somebody's desk. Yeah, so the supercomputers we're using, for example I'm using Excalibur at the ARL Supercomputing Resource Center. What's the name of a computer? Excalibur is the name of the computer. Because we like having really cool names for our supercomputers and we like having acronyms for everything else. Yeah, we like to name supercomputers after old weapon systems. Or current weapon systems. Another computer we use is Centennial, which actually marks the 100th anniversary of Aberdeen Proving Ground, that's what they named it after. People should know that the first supercomputer of a kind was ENIAC that was developed at the US Army's Aberdeen Proving Ground in the 1940s. Absolutely, they even have an old piece of ENIAC at the Supercomputer Resource Center. It's very, very cool. A lot of history at that building. But to give you an idea of how big these supercomputers are, a single computer that you might have next to your desk has between 8 and 32 CPU cores on them. Central processing units. The brain of any laptop or any desktop computer. Right, and maybe 8 to 32 gigabytes of main memory. A supercomputer like Excalibur has 100,000 of these central processing unit cores and many petabytes. A petabyte is 1,000 terabytes, which is 1,000 gigabytes. So just an extraordinary amount of memory and computing power compared to a single desktop machine. And this is what allows us to solve these ginormous problems that you couldn't even think of solving on a desktop machine. Ginormous is a technical term. I got you. But the science here is that to make these, how many cores did you say? 100,000 cores. 100,000 cores work efficiently is really the key because you have to parallelize every operation, every add, multiply, subtract, divide operation that a computer makes. All has to be done in parallel across these 100,000 cores. And that's the trick, I think, is to get algorithms that run efficiently on these kinds of supercomputers. So are you on that side mainly or are you on the physics side? So I split my time 50-50 between these two problems. So on one hand, I'm doing what you just explained, which is taking algorithms that were developed for desktop machines that just run on one computer and rewriting their algorithms to function on thousands of computers simultaneously to trade information when necessary and to make use of all these cores simultaneously in a really nice way. The other half of my time is then using the software to actually solve the physics problem that we were talking about. Right, now that physics problem, if it requires a supercomputer today to use very low-frequency sensors either actively or passively, in other words, put a low-frequency into a building or try to sense the low frequencies that are coming out of a building. To use that information and to give it to soldiers in a way that's useful will require some art, I would say, to take this switcher working on on a supercomputer and actually get it into a piece of technology that a soldier can use, like a phone or some other very small computer, right? So to me, that seems like a real challenge. Let's talk a little bit about how you make the leap from a supercomputer to technology that a soldier can use either in his vehicle or even handheld. I would say it can be done in two different ways. The first is to use these extremely large models, extremely complex models to understand the underlying physics of the phenomenon that you're exploring. You wouldn't necessarily implement these numerical methods in a piece of hardware, but you'd rather use these algorithms, these numerical methods to understand the problem so that you can design algorithms that you then implement on the hardware that get deployed. The second is to use these very high-performance numerical methods and these very large models to develop simpler models that still represent the world accurately. So these models are very accurate, but they're extremely computationally intensive. Can we use the results from those models to build simpler methods that are computationally easy that can run on a smartphone or a microcontroller but still provide the accurate results or reasonable results you're looking for in a scenario? Right, so the state of the research is really understanding the problem from a very low frequency perspective. Really doing work to understand whether or not this capability is even viable. Absolutely. And we'll add an additional capability for our soldiers to be able to see inside buildings. Right. Without having to open up a hole prematurely or something else, right, to be able to really get accurate situational understanding before any mission were to ever occur. We have to understand the physics of that problem before we ever take it to something that can be hand-held. So the timelines are longer, which is really what ARL is about. We are, as a corporate research laboratory for the Army, heavily focused on the longer term, I like to say 2035 and beyond. So it sounds like this technology is really well placed for that. What do you think about that? Do you think we can get it sooner? I think the knowledge we're producing with these large models can feed products that we're developing right now and that have a more immediate uses. But like you just said, there's potential for things much farther into the future. I think there are physical phenomenon that we want to explore now but don't have the computational capability to explore right now either because we don't have the solvers built for them or there's not even enough computational power even available today. So in that sense, there's still a lot of knowledge to obtain in the next 10 to 15 years that we don't have right now. That's really the fundamental piece of the science behind the research is to develop solvers that can be used on supercomputers that are coming because we haven't stopped these things so they have 100,000 cores in them every several years. The number of cores just keeps continuing as the microelectronics improves. And of course, we're adding to these central processing units. Now we're adding graphical processing units to them which are used for deep learning and other kinds of machine learning and artificial intelligence techniques. So that whole phenomenon, you have to wait, so to speak, to be able to take advantage of that kind of processing to solve your problem. Yes, but we can keep an eye towards the future as these developments are being made. We know that graphics processing units are improving other types of exotic hardware like FPGAs are being used. So programmable gate arrays. Yeah, other exotic hardware like that are being added to these computers and so we can keep track of what these supercomputers are going to have in the future so that we can start thinking about what kind of algorithms we need to design now to take advantage of them. Yeah, so the disruption really will occur in capability when the science that we're doing with regard to had an image in a whole environment using very low-frequency electromagnetic fields. How to have that fundamental scientific understanding write algorithms that can then be put onto computers or handheld devices and can be used in the field. That will be where the disruption and capability, warfighting capability is. But there's still a bit of disruption that needs to occur on the science piece for us to really understand how to accurately take advantage of these electromagnetic fields that you've described. Still excited about your work? Absolutely, yeah. What else gets you excited? Transition is very difficult. It's one of the hardest things I've ever seen in my life. That's something I wasn't expecting coming to Aero was how difficult it is to transition an idea to a usable product. Our team has had some successes in that regard and it's very exciting to see a PM or a PEO or another customer interested in our technology and that's very cool to see. We just used three acronyms that many of the people who might be listening to this won't have a clue but I will tell you that those are the acquisition that's the acquisition side of the Army. So it's very difficult to take an idea out of a laboratory and get it into what's known as a program of record and the Army is actually spending money to buy kit and that kit eventually goes to our war fighters and is a long process that has to occur before any idea is morphed into something tangible that can be put in the hands of a soldier. But the Army Futures Command has been set up to address these problems and I think that what you'll see in the future at least over your career will be the ability to accelerate ideas from the laboratory into war fighting concepts which would then become something of a requirement very quickly and then those requirements for new capability will quickly be resourced and funded and then fielded by the Army. All of that now is really under the auspices of Army Futures Command and General Murray is a strong advocate he's the commanding general of Futures Command a strong advocate for ARL's position in the longer term to ensure that beyond modernization priorities of today which are slated to be fielded in 2028 that ARL continues to look at the longer term needs for our war fighters. That's why it's so important for us to think about 2035 and how we can bring technologies like what you've described which still have a lot of open research questions around them but why we can start to begin to think about how to take advantage of what you're working on and ultimately get it into the hands of soldiers and get them to take advantage of that perhaps longer term and that's our goal as the Army's Corporate Research Lab. So Ross, thanks very much for speaking with me today. I really enjoyed it and I'm so glad that you're here working for the U.S. Army at the Army Research Laboratory and thanks to all of you for joining us for ARL what we learned today. In upcoming episodes we'll continue to discussion about the underpinning research that will build the Army of the Future. Please consider liking and subscribing. Science is a journey of discovery and we're glad you're along for the ride. For the Army Research Laboratory, I'm Dr. Phil Percot.