There's a trend that should be obvious, but is often overlooked at higher echelons of the defense industry: More and more of system functionality is now implemented as software running on single board computers, replacing legacy systems based on hard-wired electronic assemblies. The drive for ever more compute density has become the mantra for many of today's advanced military programs. In this, our first Target Report of the year, we analyze five compute-intensive military applications and explore how today's crop of embedded computer form factors and technologies is serving their needs.

As we researched the various programs, it became obvious that there are, of course, many more than five compute-intensive military applications to choose from-there are so many that fit that definition. Ultimately, it was boiled down to the following five somewhat general categories:

-  Multi-Sensor Image Processing and Exploitation

-  Aircraft-based Cluster Computing

-  Persistent Surveillance on UAVs

-  Secure Network Communications

-  C4ISR

The vast amount of sensor data streaming into ISR sensors are demanding ever more processing to exploit the data and make it useful for warfighters. The sidebar "Multi-Sensor Image Exploitation Using Smart Processing" in this article explores that topic.

Some military reconnaissance aircraft have up to seventy computer systems dedicated to different aspects of the surveillance mission. The bylined sidebar "Cluster Computing Suits Up for Aircraft Duty" by Trenton Systems on p.XX in this article, discusses the architectures suited to those needs. Persistent surveillance requires highly compute-intensive imaging capabilities as a UAV remains in an area to detect, locate, characterize, identify, track, target and re-target threats. The bylined sidebar "GPU-based Processing Solves Persistent Surveillance Challenges" by Mercury Computer Systems on p.XX in this article explores the issues invloved.

The combined pressures of high bandwidth, security and data processing make today's military comms systems among the most compute-hungry around. The problems and solutions are explored in the article "Secure Network Communications Pushes Compute Limits" by Kontron on p.XX of this section. And, last but not least, a variety of deployed high-performance systems can be classified as C4ISR (Command, Control, Compute, Communications, Intelligence, Surveillance, Reconnaissance). The article "Next-Gen Processor Technology Redefines Compute Density" by Extreme Engineering Solutions on p.XX of this section, talks about how C4ISR systems are typically very compute intensive with high communication bandwidth requirements.  They are deployed in harsh environments, and they have Size, Weight, and Power (SWaP) constraints.

High Compute-Density Needs across the Industry

Beyond those five application areas, the defense industry if rife with programs that push the limits of computing needs. Many of these are driven by the general desire for autonomous operations, along with real-time, or closer to real-time decision making. Automatic Target Recognition, for example, is in some ways a fairly established technology. But systems that can acquire and track targets completely autonomously-doing so when the target and shooter are moving-call for a whole different level of computing. Signal Intelligence (SIGINT) is another area where added computing muscle serves an endless appetite for security-critical data sifting.

In general, military UAVs and their payloads are by definition compute intensive, and will only become more so. Tasked to capture and download secure, encrypted surveillance data, today's advanced surveillance UAVs require a lot of communications overhead. At present, U.S. military recon UAVs relay almost all UAV captured data to the ground to process it for interpretation and decision making. The goal is to make use of onboard processing muscle to enable UAVs to instead relay the results of their data processing to the ground for decision making.

The benefit is reduced reliance on data link rates in certain applications, particularly imagery collection. In today's UAVs, image formation is done in the air and then sent down. For payloads of the future, the trend is toward fusing data and sending down just the things that are different than the established data base-or some other way of compressing and fusing the information. All of this helps overcome the defining constraint for these systems: the limitations of data link bandwidth.

Waveform-Driven Applications

Meanwhile, waveform-intensive applications like radar and SIGINT seem to have no end to their appetite for signal processing power. Faster DSPs coupled with a broader range of IP cores and development tools for FPGAs are joining forces to form new DSP system architectures. Using those building blocks, board-level subsystems must quickly acquire and process massive amounts of data in real time. As FPGAs evolve to ever greater sophistication, complete systems can now be integrated into one or more FPGAs. That in turn means that the rack and backplane-based systems based on FPGAs offer the compute muscle of yesterday's supercomputers. Modern radar systems are operating over an ever increasing frequency range. Analog conversion technology-both A/D and D/A converters-are also feeding the radar needs of the military.

System developers can now build radar receiver systems with a higher instantaneous bandwidth thanks to the converters, and can handle the corresponding increase in compute power required to process the received data streams using FPGAs. The ASIC-based radar design approaches of the past can achieve the performance needed, but that path lacks the flexibility inherent in designs based on FPGA technology. A wealth of FPGA board-level products are available aimed specifically at this area.

Disruptive Technology: GPGPUs

While the attraction of FPGAs in high-end military systems is today as solid as ever, a disruptive technology could have the potential to unseat them. This disruptive technology is the emerging idea of using the latest crop of high-performance graphics processors to handle general-purpose processing tasks. Driven by architectural advancements in recent years, the scope of applications to which GPUs can be applied has grown dramatically. Feeding this notion of GPUs as general-purpose processing engines, NVIDIA developed a parallel computing architecture called CUDA (an acronym for Compute Unified Device Architecture) that addresses a key weakness of FPGA parallel processing systems: the complexity of programming them. CUDA is the computing engine in NVIDIA graphics processing units (GPUs) that is accessible to software developers through industry standard programming languages.

GPGPU technology can, for example, deliver more capable detection systems, increase the autonomy of unmanned vehicles and provide a wide-ranging improvement in survivability across a broad spread of applications. According to GE Fanuc, a major defense prime contractor has a radar application to the CUDA environment and achieved a 15x improvement in performance. In another case, according to GE Fanuc, the productivity of the CUDA environment is illustrated by the brief time-just over two weeks-it took another prime contractor to migrate an application to the CUDA environment.

This focus of computer-centric military applications seems to be in synch with the direction upcoming DoD budgets will shift to. Even as the political landscape changes, and forces within the government drive the budget down, the embedded computer component of the overall DoD budget is going to increase dramatically-a trend that's been occurring consistently now for several years.

Curtiss-Wright Controls Embedded Computing
Leesburg, VA.
(703) 779-7800.
[www.cwcembedded.com].

Extreme Engineering Solutions
Middleton, WI.
(608) 833-1155.
[www.xes-inc.com].

GE Fanuc Intelligent Platforms
Charlottesville, VA.
(800) 368-2738.
[www.gefanuc.com].

Kontron America
Poway, CA.
(858) 677-0877.
[www.us.kontron.com].

Mercury Computer Systems
Chelmsford, MA.
(978) 256-0052.
[www.mc.com].

Trenton Systems
Gainesville, GA.
(770) 287-3100.
[www.TrentonTechnology.com].

SIDEBAR:

Multi-Sensor Image Exploitation Using Smart Processing

TOM ROBERTS, PRODUCT MARKETING MANAGER, MERCURY COMPUTER SYSTEMS

New generations of ISR sensors are surveying ever wider areas with constantly increasing levels of sophistication. Electro-optic infrared (EO/IR), synthetic aperture radar (SAR), hyper-spectral imaging (HSI), laser radar (LADAR), SIGINT and COMINT systems scan vast spaces and collect an enormous amount of raw data. Mounted on long-endurance, unmanned platforms, an array of these sensor technologies can deliver the persistent surveillance necessary to find and fix an elusive, insurgent enemy.

Experience shows, however, that the volume of multi-sensor data collected in just one mission overwhelms the human imagination and crushes existing tasking, processing, exploitation and dissemination (TPED) architectures (Figure 1). How can we extract truly valuable information from all that data and get the information to the people who really need it, when they need it?

Today's image exploitation systems are already overloaded at two levels by the current volume of data. On the first level, deployed tactical data links cannot transmit the full data streams generated by multiple sensors. On the second level, the volume of data that is successfully transmitted to a ground station exceeds the current TPED architecture. An estimated 90 percent of the received data is not even examined-it essentially just "falls on the floor" and remains unused.

The situation requires a new approach, a novel TPED architecture that calls upon next-generation computing technology to multiply the effectiveness of human analysts and reduce the delivery time for actionable intelligence. This "Smart Processing" architecture performs the initial stages of processing and exploitation directly on the tactical platforms, rather than on the ground stations. Both analysts and the remote devices of forward-deployed personnel direct the processing and exploitation in an on-demand fashion. Computing, communications and networking are located as close as possible to the critical tactical users.

A Smart Processing architecture first directs the multi-sensor data streams to an on-platform real-time computing system. This embedded computer uses advanced signal and image processing algorithms to make a first pass through the incoming data, prioritizing the data for downstream analysis, and tasking the collection of additional data to more rapidly find and fix targets. For example, an aided target recognition (AiTR) algorithm can examine large numbers of SAR images and/or motion imagery and provide an alert to potential threats.

Smart Processing also includes real-time cross-cueing, using multi-intelligence sensors to detect, track and engage threats with a higher degree of precision, supporting multiple simultaneous users and delivering mission-specific, tailored sensing to warfighters on the tactical edge. These are just a few examples of Smart Processing's potential capabilities.

Unleashing these capabilities demands a huge leap in deployed processing power, based on embedded, real-time computers that are small, powerful and rugged. Those are just the baseline requirements. Successful deployments also require that these computer systems be inherently networkable, and able to be configured and re-configured dynamically into flexible, mission-focused networks.