Military AI innovation, SOSA hot topics at Embedded Tech Trends

In photo: Embedded Tech Trends 2020 covered embedded computing trends such as AI, military open architecture initiatives, and more. Shown is VITA Executive Director Jerry Gipper opening the event. Photo credit: Jena Warren/The Simon Group.

The COTS Confidential Roundtable gathers experts from the defense electronics industry – from major prime contractors to defense component suppliers. Each Roundtable will explore topics important to the military embedded electronics market. This issue, we discuss how embedded computing suppliers are leveraging artificial intelligence (AI) for military applications, the impact of the Sensor Open Systems Architecture (SOSA) Consortium and other open architecture initiatives, and the outlook for the future of embedded technologies in the defense and aerospace markets with sponsors of the Embedded Tech Trends (ETT) conference, held during late January in Atlanta, Georgia.

This time, our panelists are Rodger Hosking, Vice President and Co-founder, Pentek; David Jedynak, Chief Technology Officer, Curtiss-Wright Defense Solutions; John Bratton, Director of Product Marketing, Mercury Systems; and Doug Patterson, Vice President, Global Marketing, Aitech Systems.

MIL-EMBEDDED: Artificial intelligence, or AI, was the topic of many presentations at ETT this year. How is AI making an impact in military electronics? Which applications are benefitting from the technology now? Electronic warfare? Radar? Other?

HOSKING: Virtually all military electronics are benefiting from AI, and the technology is moving quickly. The first applications are “expert systems” that deliver relatively quick decisions and actions for a relatively narrow task. These include classification and identification of all objects in the operating theater of war, and even determining the most effective countermeasures or attack strategies. Another very important application is extracting critical intelligence from the glut of electromagnetic spectrum signals and internet traffic of all types, and then detecting patterns or relationships among those signals for further action. With continued expansion of deployed unmanned military vehicles, autonomous AI systems can help boost their survival rate and mission effectiveness. AI is not just developing smart engines, neural networks, or algorithms. It requires engineering of hundreds of specialized systems, each evolving over time to exploit a broader scope of sensor inputs, and more complex decision-making elements and principles to deliver increasingly accurate, targeted results.

JEDYNAK: Machine learning, deep learning, and AI are revolutionizing applications that help warfighters identify threats and objects from afar, detect unseen dangers, and locate and resolve equipment issues before failures occur. AI is key to the DoD’s Third Offset Strategy [which seeks to outmaneuver advantages made by top adversaries primarily through technology]. The specific examples almost don’t matter – we can take it as a given that it provides a significant leap forward. Think of the automobile versus horse and buggy: While so much remains extremely familiar, and on a spec sheet may not look all that different (i.e., four wheels, room for passengers, leather seats, room for luggage, luxury styling, etc.), the performance/capability difference is immense. In electronic warfare (EW), for example, AI enables machines to identify objects and take appropriate actions in a faster and more accurate way than humans can on their own. In signals intelligence (SIGINT) applications, machine learning and deep learning can be used to automate signal classification, which otherwise typically requires extensive expertise and is prone to human error.

Another key area where machine learning capabilities can be applied to increase the safety of warfighters and equipment is health and usage monitoring systems (HUMS). It enables computers to be trained to intelligently predict when equipment failures are most likely to occur and allows issues to be addressed before a mission is affected. For example, if a HUMS application detects that extra power is being applied to a wheel on a Humvee, it could mean the pressure in that tire has dropped and the tire could collapse. The same technology can also be used to inspect the physical integrity of air and ground vehicles before and after field operations. Ultimately, AI is enabling faster, more accurate identification of threats to today’s military platforms and helping deliver a competitive edge by accelerating and strengthening functions traditionally performed by humans.

BRATTON: From an embedded standpoint, our customers are seeking greater processing power and density to enable their defense systems to execute smarter missions, have increased autonomy, and make sense of and fuse together ever-wider streams of sensor data. Many of these applications are requiring processing hardware with headroom so quickly evolving machine learning, cognitive decision-making and AI processing capability can be deployed without an immediate tech refresh (AI capability doubles every three to six months). Today, AI applications usually reside within the data center. We are seeing a trend towards creating the same processing architecture closer to the data source, often on mobile, remote platforms. The ability to scale, make secure, miniaturize, and deploy the processing power and scalability of the composable data center and its extensive software ecosystem is being packaged as high-performance embedded edge computing (HPEEC) solutions that are enabling next-generation defense systems to be deployed faster and with all the capabilities found in the most contemporary commercially developed technology.

PATTERSON: AI is impacting nearly every area of the military, including some that weren’t even discussed at Embedded Tech Trends (ETT). GPGPU technology is critical to these advancements. Truly, AI is now today only scratching the surface of the potential applications, defense included. The analogy of the “tip of the iceberg” is perfectly applicable here – less than 10% of AI applications are visible; the rest has yet to surface. As AI technology continues to advance and line geometries shrink, even more raw horsepower will be at the applications developers’ fingertips.

One major area is cybersecurity. Adaptive heuristics (and deep learning) have been developed and are being refined to monitor cloud traffic, looking for key words and phrases that can then be placed in context to help thwart and mitigate cyber hacks that threaten military platforms and reduce the level of harm our troops are exposed to, mitigating collateral damage and ultimately saving lives.

Another key area is surveillance and reconnaissance. This is where AI implemented in GPGPUs really shines, as it’s the perfect mix of parallel processing on literally hundreds of cores, tied to video capture and image processing and display (if/when needed), all in the highest-definition video standards.

Another application area is neural networks, where tens to hundreds of cores can be networked together to potentially hundreds of similar subsystems, each with hundreds of cores all addressing one or multiple strategic or tactical situations in parallel. Today, through the technical hardware and software tool innovations brought to the market by companies like NVIDIA, Intel, and others – and adapting these for true defense and rugged applications – teraflops of processing power can be applied at 15 to 30 per node instead of kilowatts per node. This has been a dream of systems designers and engineers for decades, and is now within reach. Imagine the raw compute power of systems containing 500 to 1,000 parallel cores, each with from 1 to 30 TOPS [Tera Operations per Second], all networked and freely communicating to other nodes in the system. AI and GPGPUs specifically are essentially hugely parallel DSP engines all interconnected by crossbar matrices to memory and I/O resources with Gb/sec pipes. The applications are truly endless, limited only by the imagination.

MIL-EMBEDDED: Immediately following ETT, the Army, Navy, and Air Force held the Tri-Services Open Architecture Interoperability Demonstration at the Georgia Tech Research Institute showcasing the advantages of SOSA and other open architecture efforts. Why does this effort seem to be so different in terms of momentum than past initiatives? Military participation? Passion among the industry players? Economics?

HOSKING: All of the above, for sure! Over the years, we all have seen standards and initiatives come and go. Solidly founded upon sound objectives, the Sensor Open Systems Architecture (SOSA) initiative goes well beyond a hardware or software specification. All three services demonstrate commitment to aggregate their own standards into a single standard to benefit from scale, availability, and life cycle support of products. Vendors finally see a way to protect their costly IP development efforts, by competing on innovation and technology instead of simply hardware costs. Primes see more sources of new technology products to enhance systems performance. DoD is already issuing request for proposals for systems, with vendor selection based heavily on open standard architecture content.

JEDYNAK: The main difference is cultural – there’s a very different culture in DoD now, one that is much more focused on doing and deploying fast rather than “silo”-ed thinking – and again, that cultural change is also driven by the Third Offset Strategy. The new culture is driving the fast transition of technology to the field, and helping to overcome what some – only half-jokingly – call the Chinese military’s greatest asset, the DoD’s famously cumbersome acquisition process. Last month, for example, Secretary of Defense Mark Esper, speaking about U.S. competition with China, said, “Our success is contingent upon a cohesive approach across public and private sectors. For the department, this means overhauling our policies and reshaping the culture within the department; between the department and industry; and among our allies and partners around the world."

Military participation in the open architecture and interoperability effort has certainly impacted the momentum. Developers of defense and aerospace solutions have been leveraging open standards to improve interoperability for a number of years now; however, 2019’s Tri-Service “Memorandum for Service Acquisition Executives and Program Executive Officers” drove home the point that these initiatives are no longer optional – they are vital and they are mandatory.

Support from industry players that have long endured the challenges of limited interoperability is likely playing a critical role as well. A true open standards approach offers systems integrators increased flexibility to choose the solution that makes the most sense for their needs, regardless of vendor. It reduces the costs and complexity of upgrading systems and minimizes the SWaP [size, weight, and power] ramifications of adding new functionality to a platform. What’s more, it creates a more fair and competitive marketplace for COTS components – a benefit for vendors and customers alike.

BRATTON: The Tri-Service demonstration illustrated how the SOSA approach is influencing the development and deployment of low-risk, high-performance defense computing systems. SOSA builds in multiple key differentiators that deliver the capabilities required to maintain a technological separation between our defense systems and those of our competitors. SOSA has:

· Tri-Service support and increasing alignment from industry and prime contractors for scalability and affordability

· Security that is built-in and not bolted on and uses a single (12 V) power distribution rail

· Leverages a commercial business model enabling all stakeholders to achieve what they require for success

· Common OpenVPX profiles and console ports for greater interoperability

· Ubiquitous common system manager and off-the-shelf software compatibility

· Compatibility with VICTORY [Vehicular Integration for C4ISR/EW Interoperability], MORA [Modular OpenRF Architecture], CMOSS [C4ISR/EW Modular Open Suite of Standards], and other major embedded modular open system architectures for program velocity

The Tri-Service compatibility demonstration showed how these commercial and technical attributes are driving the SOSA-aligned ecosystem, efficiently putting the best commercially developed technology into the hands of our service members faster.

PATTERSON: The key is the attraction of the Tri-Service adoption, which today is truly reaching across the aisles and breaking down the old, once stovepiped, separation of the military services, moving towards some form of commonality and potential unity. It’s now gained passion and momentum in the industry itself and is being thoroughly reinforced in the press as becoming the next motherhood and apple pie idiom, dare I say, achieving nirvana. Whether or not it will actually reach nirvana is another thing; it could be also be a groupthink mentality, all nodding and chanting that SOSA is great. So, at the moment, the jury is still out, but hope springs eternal in the current market. In terms of military participation, it is being mandated by conformance (compliance) to the standards and, once published by the program offices, there will literally be no other option-- your products either conform or not.

MIL-EMBEDDED: What will be the next big thing for military embedded technology five or ten years down the road? Predict the future.

HOSKING: Maintaining military superiority will require special attention to cybersecurity, space-based weapons, hypersonic weapons, surface fleet protection, and autonomous systems. Funding for all has certainly gained traction in the last several years. Essential capabilities in imaging, recognition, detection, classification, and identification will continue to be refined through advances in sensors, AI, and machine learning. The greatest leverage against government and military adversaries is our ability to deter aggression though overwhelming advantages in these critical capabilities.

JEDYNAK: On the hardware front, future roadmaps will be smaller and more ubiquitous, more like Lego bricks than cellphones. A good goal concept, recently briefed by the U.S. Army, is the Iron Man/J.A.R.V.I.S. model, to provide seamless integration of intelligent systems with warfighters via always-on/resilient networks. This approach will enable situational awareness across multiple battle domains (physical, electromagnetic, cyber) and leverage AI to appropriately triage, decimate, route, fuse, and present actionable data to the warfighter in real time. It will also provide the warfighter with the ability to easily control their unmanned assets in a natural manner (e.g., natural language commands, gestures, biofeedback). And once again, all of this is part of the Third Offset Strategy emphasis on “man-machine teaming.” From the hardware perspective, it means an emphasis on mixed-signal intelligent systems – physical sensors plus RF sensors plus network communications plus AI, all cyber hardened with easy scalability and extensibility from the smallest of systems (little drones) to big HPEC (armored supercomputers), and everything in between.

AI will continue to automate and improve defense applications, evolving the intelligent and connected battlefield. The increase in unmanned platforms will have a positive impact in protecting human lives from danger. Even better, the proliferation of autonomous vehicles and unmanned air taxis in the commercial space will likely accelerate technology advancements in unmanned defense platforms.

BRATTON: The rate of technological advancement has never been higher. With initiatives like SOSA, the conditions are set where defense system development and deployment can also progress at the speed of technology itself. Consider the commercial digital convergence that created converged media, information systems, smartphones, and autonomous vehicles. This commercial digital convergence is seen all facets of our lives, has a proven roadmap, and is enabling new technology breakthroughs in processing domains everywhere. This capability will drive the military digital transformation enabling platforms to shrink and become more capable and adaptable for mission autonomy. Early adoption is underway within unmanned land, sea, and air platforms, including unmanned aerial vehicles. The latter is where the military digital convergence is accelerating the fastest, as extreme-SWaP performance, proven system integrity, and processing power, among other requirements, are the prerequisite to success. If trends continue, then the commercial domain will deliver autonomous smart ground and air taxis; within the defense and aerospace domains, equally capable smart platforms will be deployed that maintain a capability gap between our systems and those of other great powers and competitors.

PATTERSON: The future, it seems, is limited only now by our imaginations. As newer semiconductor technologies continue to advance and lithographies continue to shrink, power and performance mount while memory capacities continue to grow. Add to the mix the advances in AI software and advanced programming tools and languages, and the future is, indeed, limitless. I do, however, hope we keep the “Terminator” movies in mind and stay far away from a “Skynet” that becomes self-aware. Isaac Asimov’s three rules of robotics are as valid today and into the future as when he penned “I, Robot” in 1950:

· A robot may not injure a human being or, through inaction, allow a human being to come to harm.

· A robot must obey orders given it by human beings except where such orders would conflict with the First Law.

· A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.