2005-2006 Assessment of the Army Research Laboratory (2007)

The Computational and Information Sciences Directorate (CISD) was reviewed by the Panel on Digitization and Communications Science during May 3-5, 2005, and May 23-25, 2006. CISD consists of three research divisions: Computer and Communication Sciences, Battlefield Environment, and High-Performance Computing. It also includes the Corporate Information Office and the Enterprise Management Center, which serve the entire Army Research Laboratory (ARL) through a combination of computing hardware, software, and staff.

CISD performs research for the following purposes: to help design a robust, highly mobile battlefield communications network while ensuring that the information provided to commanders is current, authentic, accurate, and protected; to develop high-fidelity “micro” weather forecasts in near-real time (i.e., to predict weather in 10 minutes or less for the next 0 to 2 hours) in order to support combat intelligence operations and troop engagement decisions; to enhance the decision-making prowess of commanders in the battlefield; and to develop robust physics-based, high-performance computing models and software for concept evaluation, design, and analysis (usually in support of computational science efforts in other ARL directorates).

Tables A.1 and A.2 in Appendix A show the funding profile and the staffing profile for CISD.

Since the last documented review (for the 2003-2004 period), there have been a variety of changes that have affected CISD’s research activities. First is a solidification of management within the directorate, with two of the three divisions receiving permanent, high-quality managers. Second, a new International Technology Alliance (ITA) has been initiated, focused on networking and information sciences. Finally, the major Horizontal Fusion and Warrior’s Edge programs have come and gone, and

the Command and Control in Complex and Urban Terrains (C2CUT) Army Technology Objective (ATO) was incorporated into the Networked Enabled Command and Control ATO led by the Communications-Electronics Research, Development, and Engineering Center as directed by the Assistant Secretary of the Army for Acquisition, Logistics, and Technology. All three have had a distinct impact on CISD’s research portfolio.

The Digitization and Communications Science Panel that reviews CISD yearly has continued to see improvements in the quality of the research being performed by CISD researchers and in the demonstration of the relevance of their work to Army needs. A concern from 2003-2004 about the number of acting management has been solved, with the appointment of a permanent director and permanent division chiefs.

CISD continues to demonstrate technology leadership in several areas. The machine translation of foreign languages, in particular, is such an area. Embedded machine translation systems (computer systems with one or more machine translation engines embedded in them) are developed at ARL to provide the Army with multilingual technologies for text and speech translation, document exploitation, and electronic message exchange. Technologies are adopted and adapted from the outside whenever possible and developed internally when necessary.

Over time, machine translation systems have grown from a laptop with a page-feed scanner, a battery pack, and embedded Forward Area Language Converter (FALCon) software to a range of systems, including a soldier-based system composed of a personal digital assistant (PDA) with a camera and tailored software, to more capable vehicle-based and collateral-space-based systems, each with its own tailored suite of hardware and software. Accomplishments in this area include the following: development of a PDA-based document triage system enabling soldiers in the field to determine the relevance of seized documents; development of a Web-enabled version of Document Exploitation Suite software; development of a Web-accessible testbed, which has been used to create task-based translation accuracy metrics and to determine “best of breed” Arabic components for upgraded software; and CISD’s technical lead of a team that went to Iraq for the initial assessment of a Defense Advanced Research Projects Agency (DARPA) two-way Iraqi-English speech translation prototype.

This language technology group has focused on developing evaluation criteria and test conditions under which commercial technology offerings can be and are being evaluated in an environment that allows identification of the best candidates for Army deployment. Machine translation evaluation is critical to identifying which machine translation engines are good enough for which foreign languages to go on which systems and platforms to support which users performing which tasks. Rather than focusing solely on technology development in an area with a much larger commercial base, the CISD approach ensures that CISD not only knows where the current problems are in the state of the art, but also when progress has been made in solving them and what holes are not being plugged with commercial developments. Prior reviews of CISD by the Army Research Laboratory Technical Assessment Board (ARLTAB) have discussed the potential usefulness of a more focused approach to validation and verification of new technologies—the approach used with machine translation may be worth emulating in other project areas.

Selected areas of research in atmospheric acoustics and radio-frequency (RF) propagation in battlefield environments continue to demonstrate a leadership position. Much of the work in acoustical propagation in the Battlefield Environment Division (BED) is firstrate, of world-class quality. This acoustical work is of great value to the Army because transmission loss resulting from propagation through the atmosphere is a key component in predicting acoustic detectability and acoustic signatures and therefore in determining operational acoustic sensor performance. Currently, BED is conducting research to develop and evaluate acoustic propagation models that incorporate the environmental effects of the atmosphere, turbulence, terrain, forests, urban structures, and land-water interfaces on acoustic signatures, and to develop remote sensing techniques using acoustic and seismic waves to provide information on conditions in the battlefield environment, including urban structures.

The combination of in-depth theoretical and wide-ranging experimental work in atmospheric optical propagation performed by researchers at the ARL Intelligent Optics Laboratory may well be the finest in the world in this area. The assembled laboratory facilities for this work are extremely impressive and are all being used for outstanding, Army-relevant research, both basic (e.g., propagation fundamentals) and highly applied (e.g., applications to new communications and imaging systems). Past efforts of this group in adaptive optics for applications in high-energy laser weapons and free-space optical communications have led to significant improvements in the quality of optical transmission through the atmosphere. Its ongoing work continues to push the envelope by focusing on the development of handheld high-bit-rate (gigabit-rate), active, adaptive laser communications systems capable of operating in atmospheric turbulence, and in developing conformal phase-locked fiber-based optical systems for optical communications, imaging, and laser weapon applications. This group’s recognition of the speckle effects in laser illumination applications is insightful, and the proposed solutions under investigation are novel and quite possibly effective.

It is clear that ARL continues to view high-performance computing (HPC) as a critical technology driven by requirements from a variety of applications across multiple directorates; these applications are in areas including armor and armaments, atmospheric modeling, aerodynamics, and computational biology. CISD has a unique role in that it is both a user and developer of codes for HPC facilities. Some projects, such as the Simple Parallel Object-Oriented Computing Environment for the Finite Element Method (SPOOCEFEM) code, represent solid examples of software engineering. SPOOCEFEM was an infrastructure development project, its primary goal being the creation of a scalable framework for unstructured mesh, finite-element simulations. The code was written in C++, and significant effort was employed to ensure that the overheads associated with object-oriented programming were minimized while retaining strong features such as reusability and maintainability. This prize-winning code was intended to be used primarily in composite manufacturing processes, and its performance was good. It had all of the attributes needed for a successful supercomputer-level code, object-oriented modular construction, concerns for scalability, validation, and verification among them. Although this effort has ended, aspects of it may be used in future projects and it demonstrates that CISD can perform effective software engineering work.

Developing leading-edge research expertise can involve both external hires and internal support for the advanced graduate education of CISD personnel. Over the past 2 years the panel has seen multiple examples of high-quality Ph.D. graduate work, for the most part with a solid relationship to Army needs, being done by current CISD employees.

Finally, a new International Technology Alliance has been established to focus on network theory, security across systems of systems, sensor information processing and delivery, and distributed coalition planning and decision making. These are all topics of emerging importance to future Army systems, and they mesh well with many of the other CISD undertakings, such as information fusion.

A newly emerging area of far-reaching importance to the Army is information fusion, the extraction of actionable intelligence from disparate data sources. In recent years, the Office of the Secretary of Defense’s Horizontal Fusion and Army’s Warrior’s Edge programs have focused, with little real progress, on developing large-scale architectures for such applications. Recent refocusing of the research direction by CISD on “small fusion” problems, such as letting a soldier know what is around the corner of a building in real time, is far more realistic and liable to lead to deployable tools earlier. The challenge will be to know when it is time to integrate different fusion applications (both from ARL and from the broader community) into more generic capabilities, especially those including learning, and in verifying and validating against metrics that are not yet defined. The bake-off methodology employed by ARL in the machine translation area is a model worth considering here. This essentially means determining good evaluation metrics and constructing very stressful scenarios and test cells in which candidate applications can be demonstrated in a common environment, with the best ones and remaining issues identified.

Another area with tremendous implications for Army missions is ad hoc networks, especially wireless and sensors at the network nodes. Ad hoc networks are electronic networks in which the individual nodes attempting to communicate come in and out of contact with one another, can move dynamically (and thus affect which other nodes they may be in contact with), and may encounter environmental constraints (power, bandwidth, real time, security) not present in traditional networks. Examples include ground-dispersed sensor networks distributed around a battlefield for movement detection, air-dispersed networks for local weather prediction, networks of individual soldiers as they interact on the battlefield, and even on an individual soldier as he loads or unloads different gear with Bluetooth connectivity. The problems here are significant, ranging from power management for maximizing the lifetime of an overall sensing capability, to guarantees of real-time delivery, to security for ensuring that misleading data cannot be entered into a network or that adversaries cannot surreptitiously extract vital information from a network. The potential correlation with information fusion is obvious. ARL has developed an automated service discovery capability that, by maintaining network awareness (connection state, bandwidth, and topology), protects domain application from network disruptions due to mobility and terrain so that established connections between producers and consumers can be preserved.

ARL currently has multiple projects underway in this area, ranging from purely practical demonstrations of some capabilities using off-the-shelf networking equipment and protocols, to algorithms for specific network applications. Once again there is also a large external commercial, university, and government-run suite of projects underway in similar areas, and the challenge to ARL will be to move expeditiously toward deployable technologies while avoiding duplication of other work and yet still ensuring that the Army’s unique requirements are being solved.

Of related and increasing importance are network security issues, especially ones that rely on wireless networks. While CISD maintains a significant Network Intrusion project, much of it is focused on active day-to-day defense of current wired networks, using for the most part tools derived from industry and academia. Current resources do not permit much real targeted research into Army and battlefield issues, especially for wireless networks. However, ARL has the capability and facilities to construct very stressful scenarios and test cells in which security issues can be demonstrated in a common environment, the best ones identified, and remaining issues brought to the surface. Thus, some variation of the bake-off model demonstrated by the machine translation group may again be appropriate as additional resources are identified.

High-performance computing is becoming an essential part of a broad swath of ARL projects. Despite the relatively large budget allocated to HPC, however, the bulk is spent on hardware and

hardware-related infrastructure, with only a few million dollars spent on applications and tools. Even here, such in-house tools as SPOOCEFEM are developed and then apparently left mostly unused. A principal concern regarding this area is the possibility that the worthy ambitions of ARL may outstrip the resources available. Even the application of existing simulation tools requires significant investments to understand the applicability and limitations of the algorithms and software, as well as for the extension of such tools to meet specialized ARL needs. Algorithm and code development originating within ARL must be carefully planned, particularly in an era of limited resources: the development of new simulation capabilities involves substantial expenditures over long periods of time.

Petascale computing is an enterprise with large uncertainties and risks: the hardware is an order of magnitude more expensive than current systems and will require a much larger space and power infrastructure; and the vendors will be supplying much less of the software stack needed to use the machines productively. Efforts by ARL to develop in-house system architectural evaluation tools may thus be of limited use; a more productive approach for the long term may be to focus on what ARL mission-critical applications need to scale to the peta range, understand the real requirements of such applications, and work toward those unique algorithms or software that will be needed to implement the applications when such systems become available.

A challenge for the atmospheric signal propagation work, for which ARL has world-class facilities, is to ensure that appropriate experimental design techniques are used in this process. In cases where there is high potential for revisiting prior experiments, it is important to look for anomalies found in the field, and formal methods for saving, querying, and retrieving such experimental data should be put in place. The potential of a quick-response “virtual laboratory” for answering unforeseen questions quickly is significant and may be worth significant consideration, especially as petascale computing systems are considered.

Polarimetric imaging is an area in which BED has the resources necessary to perform cutting-edge experimental work. The addition of one or two scientists whose primary interests are in the fundamentals of atmospheric turbulence would help to provide the theoretical horsepower necessary to move the science forward. This would facilitate further progress in understanding the fundamentals of atmospheric signal propagation, particularly in battlefield environments when conditions are such that observations cannot be uniquely interpreted, and it could also facilitate collaborative studies with staff who work with computational fluid dynamics and high-resolution, mesoscale atmospheric models.

The issue of validation and verification is one of continuing concern to the Board. Validation refers to the methodology used to ensure that a mathematical model of a phenomenon does in fact mirror the real world correctly, especially in those areas of interest to a particular program; verification is the process of examining whether the implementation of these models, both in terms of computer codes and initializing conditions, is done in a manner that does in fact deliver results consistent with the models. The continuing challenge to ARL is to ensure that projects whose results rely heavily on such models really do have the appropriate amount of attention given them.

In response to prior criticism and recommendations, the Board was pleased to see the emphasis being placed on the validation and verification of a number of BED analysis and forecast models (e.g., battlefield environment, meteorological, and acoustical and other signal propagation). The Board does suggest that more thought now be given to experimental design rather than to ad hoc case studies.

The issue of making systems engineering expertise part of many research problems is becoming of increasing importance to ARL, especially as more and more work is done “for customers,” and as more commercial off-the-shelf technology is used in prototyping and demonstration efforts.

The work at CISD is generally very well targeted to Army needs. For example, several very successful deployments to Iraq of selected technologies have resulted from the Computer and Communication Sciences Division’s approach to automated language translation technology. The Battlefield Environment Division has established its Army and national science niche in defining and predicting the characteristics of meteorological phenomena that are critically important to fixing the properties of the atmosphere and Earth on time and space scales relevant to rural and urban battlefield situations. The emerging focus on “small fusion” holds the potential for a significant advance in soldier survivability and effectiveness in an urban battlefield. The continued use of HPC assets to simulate complex phenomena in advance can continue to accelerate the transition of new technologies to battlefield-deployable status.

Because the BED is uniquely sited and qualified to serve as a national micrometeorological research facility, the Board was pleased to see that Division scientists are significantly expanding their involvement and growing leadership in national and international, multiagency, and multinational experiments. Better understanding of the stable, usually nocturnal, atmospheric boundary layer is a specific area of critical importance to the Army and to the sciences generally. Therefore, focusing BED expertise on this problem and aerosol characterization with respect to soldier health is well advised and offers the potential for national leadership and recognition.

Likewise, work such as the natural language translation that involves a neutral and well-defined mechanism for comparing competitive developments from commercial firms provides all such entrants a clear picture of the state of the art in the basic technology, metrics for evaluating that state, and technologies that offer clear-cut advantages. The emerging International Technology Alliance on Networking offers the potential for a similar broad industrial involvement.

All of the crosscutting issues discussed earlier in this document are of direct relevance to CISD. Just-in-time information fusion is essential to tomorrow’s connected battlefield, but general architectures are not currently available, and focused efforts are needed in smaller steps. Autonomous control of vehicles is not possible without both such information fusion and advanced algorithms designed for real-time environments. HPC is a key pillar of ARL research, and the limited CISD resources available for real research need to be targeted for maximum long-term impact. Network security, especially wireless and ad hoc networks, will impact all directorates, but CISD has the expertise to understand the underlying issues and limitations of current protocols. Validation and verification continue to be of concern across virtually all CISD projects.