The panel met on December 20-21, 2018, at the National Academies of Sciences, Engineering, and Medicine (National Academies) facility in Washington, D.C., to review the In-House Laboratory Independent Research (ILIR) program projects in network sciences conducted in 2018 at the following two U.S. Army Research, Development, and Engineering Centers (RDECs): Aviation and Missile Research, Development, and Engineering Center (AMRDEC) and the Communications–Electronics Research, Development, and Engineering Center (CERDEC). The panel received overview presentations on the ILIR programs at these two RDECs and technical presentations describing the projects. During each presentation the panel engaged in question-and-answer sessions with the presenter, and a general discussion with RDEC staff after the panel had formulated initial impressions and developed additional questions during its closed-session deliberations, conducted after the RDEC staff had concluded their presentations.
The three AMRDEC projects that were presented during the review are Chaos in Linear and Piecewise-Linear Systems, Chaos in Optimal Communication and Radar Waveforms, and Optimal Signal Detection Using Mutual Information. These projects were selected for ILIR funding through a call for proposals and are a part of the research performed by members of AMRDEC’s Nonlinear Dynamics Group. The proposals are subject to a review process that includes anonymous reviewing by outside Army Research Office (ARO) program managers.
Project: Chaos in Linear and Piecewise-Linear Systems
This is a 3-year project that started in October 2016 and will end in September 2019. Its goal is to investigate the fundamental nature of chaotic dynamics using linear and piecewise-linear systems.
Traditionally, chaos is thought to arise from nonlinear dynamics, which makes the analysis of such systems very complex and difficult to pursue, as well as limited, commonly, to statistical and numerical analysis. Building on the work of others who showed that chaos can also arise in linear systems, previous work by the PI and his collaborators in the Nonlinear Dynamics Group has shown solvable piecewise-linear chaos. The significance of this work is that linear systems are amenable to analytical results that can then be used to design systems with desirable properties, or exploit properties of chaotic systems, enabling adoption of chaos-based technologies. The work of this project has provided new insights to classes of chaotic systems. The project has focused on developing robust Markov chaos and Koopman-von Neumann operator theory for low-dimensional chaos that have yielded to storage and playback applications and demonstrated linear chaos in a wave equation.
Chaos cannot arise in finite dimensional linear systems, only in infinite dimensional linear systems. In this project, the PI has shown that a one-dimensional wave equation with a gain can lead to chaos, demonstrating that the system exhibits dense periodic states, topological transitivity, and sensitive dependence to initial conditions—the three conditions required by Devaney to induce chaotic behavior. The work of the PI analyzes this equation that led him to propose new alternative definitions to the original chaos definition of Devaney. These alternative definitions allow, for example, the demonstration of chaotic dependence on initial conditions in systems that are not included by Devaney’s required conditions.
This work has led to several high-quality journal publications, showing its theoretical impact in the chaotic dynamics research of other experts. There is opportunity for several significant practical applications as the PI has pointed out because, for example, the optical gain medium in a laser or optical amplifier can be modeled as a wave equation with gain, given that other applications range from combustion fronts, tsunamis, or infection waves in temporal-spatial epidemics.
Given the theoretical results of this project, the challenge is to continue showing relevance to the Army by addressing real problems encountered by the soldier in the field or showing how to use chaos in proposing solutions that are simple to implement.
Project: Chaos in Optimal Communication and Radar Waveforms
This project is in its second year. The goal is to study optimal communications and matched filters by using chaos in linear systems. Matched filters are commonly used in conventional digital communication and radar systems. The matched filter is the optimal detector, for example, when the signal is deterministic and the communications noise is white. The matched filter is a linear system whose impulse response is matched to the transmitted waveform. To be practical, this waveform needs to be represented by a few known basis functions. In the early 1990s, work by others proposed high-bandwidth digital communications using the symbolic dynamics of a chaotic waveform. In this communication system, information is encoded using arbitrarily small perturbations of a chaotic oscillator. The resulting waveform is consistent with the unperturbed dynamics of a chaotic oscillator leading to an efficient encoding.1 Practical receivers that can detect and decode the information encoded by the symbolic dynamics, however, have been missing. This project is focused on developing practical detectors for these chaos-based communication systems. In particular, the project addresses coherent detection for chaotic waveforms. This has been considered a difficult problem because the lack of a fixed basis to represent the chaotic waveform and its irregular timing have prevented the design of simple-to-implement matched filters. The work of the PI and his collaborators in this project has addressed this problem by providing a novel chaotic oscillator that admits an exact analytic solution that, surprisingly, involves a
1 N.J. Corron, J.N. Blakely, and M.T. Stahl, 2010, A matched filter for chaos, Chaos 20(2):023123.
fixed basis function.2 This basis function can be used to design a simple-to-implement matched filter. The resulting system is equivalent to a standard binary phase shift keying (BPSK) digital communication system, enabling near-optimal detection of the binary symbols modulating the symbolic dynamics. The PI’s work then proceeded to analyze the performance of the matched filter detector providing the bit error rate (BER) in white Gaussian additive noise channels. Surprisingly, the work of the PI provides analytical expressions for high and low signal-to-noise ratio scenarios that can be used to evaluate the performance of these chaos-based communication systems.
The PI considered the application of his matched filter and his oscillator in an audio-frequency electronics system, showing its practicality to Army-relevant applications. The chaotic system combines analog and digital systems to implement the hybrid dynamics. The current electronic version developed by the PI operates at low frequencies. Future work will consider the design of an oscillator operating at radio frequencies (RFs) and enabling high-bandwidth chaos communications using symbolic dynamics. The PI’s work has the potential to develop a digital communication system (transmitter and detector) for direct sequence-spread spectrum communications and next generation ultra-wide-band radars.3
AMRDEC’s products include multi-sensor platforms, such as radar systems, and there may be specific opportunities for improving components of these sensor suites by implementing the chaotic linear matched filter.
This work has provided significant theoretical results that have been published in the best journals and conferences. These results open opportunities to develop alternative, simple implementations of optimal detectors in the place of much more complex existing detectors. An interesting challenge is to explore the actual application of these theoretical results to problems of practical significance to the Army. Challenges may exist in moving forward or transitioning this work to applied research; the current work still seems rather fundamental and may not be ready for prototyping. A worthwhile question is how difficult would it be to conduct research and development for implementing the linear chaotic matched filter.
Project: Optimal Signal Detection Using Mutual Information
This project applies an innovative approach to bringing the power of classical information theory to statistical inference problems that address data streams that obey Markovian statistics. In its first year, the project has yielded one conference presentation and a draft paper for Physical Review Letters, a venue in which the PI has been publishing for nearly 20 years. Concrete success has emerged in incorporating information-theoretic ideas into heretofore theoretical hypothesis testing problems. The work is well managed, and external publication review is providing effective oversight.
The immediate goal is to investigate new statistical tests embodying concepts and quantities from information theory to improve signal detection beyond the limits of linear correlation. The hypothesis is that valid significance tests of information-theoretic quantities that account for Markov structure in small data sets are possible. For example, applying classical measures of information to time series data limits the ability to maximize the information gathered by a sensor suite on an unmanned aircraft. Also, estimating the entropy of a data stream following Markovian dynamics is an open problem. Entropy estimation on data streams permits development of techniques to measure the transfer and leakage of information between hardware components containing sensitive information.
This work focuses on conditional mutual information (CMI), which is important because all Shannon inequalities can be expressed as linear combinations of CMIs (e.g., transfer entropy [TE], which
expresses directional causality between two processes or data streams). This project also tries to include the Markov structure of the data to build better tests. To detect a relationship such as mutual information or the direction of causality between two processes, X and Y, the work creates surrogate data that satisfy the null hypothesis and preserve the dependencies between X and Y. Bootstrapping and simulation techniques are used to produce surrogates.
This project provides AMRDEC with the opportunity to improve the information extraction from limited sensor suites aboard unmanned platforms, which leads to increased seeker performance, an important contribution to the effectiveness of the Army. It allows the opportunity to provide continuing improvements to the communication, ranging, and detection capabilities of such platforms. As presently understood, information theory is a fundamental mathematical discipline dealing in asymptotic results on abstractly defined quantities. This funded ILIR project presents an opportunity to contribute to an applied theory of information, one that allows inferences to be made from finite data in addition to statements about theoretical limits. It therefore directly addresses the technological needs of the organization, providing greater fundamental understanding of the underlying principles of its technological needs.
The effort to extend the PI’s exact test for mutual information4 to transfer entropy (TE) using chi-squared formulations of the problem yielded poor performance. The beta distribution is an excellent fit for the TE under the null hypothesis, but no method of deriving the parameters of the distribution beforehand has been found. Nevertheless, this is cutting-edge research that needs to continue to be supported in light of results already obtained.
Project: Compressed Sensing for Massive Multiple Antenna
Massive Multiple Antenna (MIMO) is a candidate technology for 5G cellular communications. This is a technology that achieves high spectral and energy efficiency by using a large number of antennas at the base station to service a much smaller number of mobile users.5 The main issues with this technology are the cost associated with the large number of receiver chains and the associated amount of computations. Dealing with the hardware and computational requirements of massive MIMO is a topic that is currently attracting intense interest among the scientific community.
The project investigates the use of sparse sensing in the spatial dimension, as a way of reducing the number of receive antennas involved at a given time. In particular, the work exploits the sparsity of the channel in the angular domain and investigates the problem of estimating the MIMO channel matrix by sampling only a subset of receive antennas and then estimating the entire channel matrix using sparse signal recovery techniques.
The project presented some initial results on the bit-error-rate performance of a zero-forcing beam-forming receiver that uses the sparse sampling-based estimate of the channel. Also, comparisons were presented to a receiver that uses the least-squares estimate of the channel, obtained based on the sampled antennas only. The comparison showed improved performance of the proposed approach. The research is interesting, and perhaps in its next phase it could focus on optimal sampling schemes in space and time (instead of random sampling).
4 S.D. Pethel and D.W. Hahs, 2014, Exact test of independence using mutual information, Entropy 16:2839–2849.
5 J.A. Franklin and A.B. Cooper, 2018, “Kronecker Compressed Sensing for Massive MIMO,” 52nd Annual Conference on Information Sciences and Systems, March 21–23, Princeton University, Princeton, NJ.
A challenge that arises with sparse sampling (i.e., when using a smaller number of receive antennas) is a signal-to-noise ratio loss. The research needs to quantify the loss, and the proposed receiver needs to be compared to a receiver that uses a channel estimate obtained based on the sampling of all receive antennas. Trade-offs between savings and receiver performance need to be studied. The validity of the channel model needs to be quantified in the context of scenarios that are of interest to the Army. The selection and estimation of parameters that appear in the currently used model (e.g., number of paths) needs to be discussed, along with the time horizon over which the channel parameters remain constant. It is also important to consider how often the channel would need to be estimated in a realistic 5G setting. Spectral leakage resulting from grid mismatch needs to be studied. Connections to the literature on sparse sampling for massive MIMO used in radar target detection also need to be explored.
The project considers a problem that is of relevance to the Army. Massive MIMO could be applied to radar target estimation; however, deploying a separate receiver chain of front-end circuits in a dense circuit board is a significant challenge. Compressive sensing in the spatial direction has the potential to reduce the required number of front-end circuits and the overall associated computational complexity. Therefore, this project presents opportunities for further research.
The project has been performed in collaboration with a Johns Hopkins University faculty member. The work has resulted in one conference publication and one conference submission. The PI will benefit from continuing to read the literature and from seeking input from the scientific community by submitting his work to conferences and journals.
The overall area of massive MIMO is relevant to the goals of the Army, because it will be part of the 5G protocol, soon to be deployed for cellular communications. Deep understanding of the issues involved and awareness of the research being conducted by others on addressing these issues will be essential to the Army. Research experience and knowledge of the literature are the necessary tools for Army researchers to quickly identify and address problems arising in military scenarios.
Project: Evaluation of Synthetic Infrared Environments for Training Deep Neural Nets
This project focuses on using synthetically generated infrared (IR) images to train neural networks. It seeks to improve synthetic IR image generation based on deep neural network (DNN) requirements. It uses an image-quality metric corresponding to the performance of DNN training. The project also seeks to reduce the cost and need for field data collection through the use of modeling and simulation tools; expand current data sets; benefit next-generation combat vehicle, soldier lethality, and long-range precision fire projects; and improve artificial intelligence algorithms in general.
This is an emerging research area, which is highly relevant to the mission of the Army. It also occupies a research niche that uses the availability of IR spectra databases from numerous field collections. Synthetically generated IR images also lead to significantly reduced costs in training DNNs.
The presentation during the review would have benefited from a clearer explanation on the relationship among real IR data, synthetically generated data, and the image-quality metric. The PI would benefit from additional guidance and interactions on the project, although the value of mentoring by his graduate advisor is evident. The PI also needs to enhance the level of interaction with the external research community to obtain the necessary feedback and peer review required to maintain the quality of the research. More PI time (beyond the current 50 percent) would benefit the project.
Project: Multimodal Content Delivery Using Resource-Aware Artificial Intelligence
This project addresses the problem of limited resources at the tactical edge and is being researched in collaboration with a professor at Cornell University. The goal is to enable the network (e.g., drones)
to intelligently choose a subset of information to present that will still convey reasonable information for situation awareness of the human, and enough information for reasonable drone placement, while having low bandwidth and battery requirements. The research also attempts to refine the algorithms by learning in the field.
This project has different opportunities for contributions. Intelligent summarization of information is an important, widely studied topic. Data visualization is critical not just because of battery life, but because it is essential for humans to be able to quickly understand the important aspects of a situation and correctly react. There is much research that has been done on this essential topic, and there are still opportunities for innovation, perhaps in resource-constrained approaches. It is an interesting question whether machine learning in the field can optimize power use versus carefully crafted algorithms developed beforehand.
The given algorithm would dynamically choose a good placement of drones based on information shared by the edge devices. Although such algorithms exist, it would be interesting to see if a good but perhaps less optimal drone placement algorithm can be rendered if the edge devices can choose a subset of information to share, so as to save battery life. Dynamically choosing different light frequencies to find combinations that will give the most information with the least amount of data would, again, help to save battery life. Another approach to saving power is to experiment with different communication modalities (e.g., visible light or ultraviolet light) to minimize power for transmission.
There are challenges for this project. Data visualization has been an important topic for decades, and it is not obvious whether there are opportunities for breakthroughs here. It is also not obvious whether unsupervised machine learning can contribute much to better information for battery use versus intelligent algorithms developed offline in advance. This research is still quite preliminary.
Project: Radiating Structure and Loading Methodologies Versus and Reliant Approaches for Conformal Antenna System Design
This work addresses research questions in antenna design, a basic research topic with both deep roots in physics and substantial implications for the Army. The focus of the analytic work is the study of the electromagnetic fields in the vicinity of the antenna, with the goal of understanding whether or not the use of metamaterials offers a fundamental advantage in the field strength—for example, in the near field. One Army problem driving this research is the design of antenna systems for the Army’s Future Vertical Lift (FVL) effort, and the demands of that domain—conformal, electrically small, and proximate to a reflective surface—significantly challenge the typical analytic framework for antenna design (e.g., the need for a quarter-wavelength antenna length). Metamaterials, which are artificial materials not found in nature that are designed with structural features such as arrays of subwavelength elements and with specific electromagnetic permittivity and permeability, have been proposed as an approach to dealing with the proximate reflector.
This research presents significant opportunities for the Army. The Army as a fighting force depends on wireless communications at the tactical edge, and the PI developed an analytic model for this problem. This novel model represents an opportunity for the Army because the problems addressed by this modeling research (i.e., the need for a small antenna in the very high frequency [VHF] and ultra-high frequency [UHF] bands) would be useful in a wide variety of Army domains. If the mathematical analysis and modeling developed through this ILIR effort are supported by the results from the PI’s experimental plans, the resulting electrically small antenna (ESA) designs can have a transformational impact on vehicular networking for the Army. The PI on this ILIR has published primarily using the U.S. patent system. While the U.S. Patent and Trademark Office has many examiners who are subject matter
experts, the PI did not report interactions with the broader scientific community, such as presentations at technical conferences or submission of the work to a peer-reviewed venue, such as an archival journal.
Challenges for this research include power requirements for the desired radiative effects and the need for experimental validation of the analytic models. Independent of the question of trade-offs in the use of conventional materials and metamaterials, ESAs tend to require greater power budgets. A second challenge is posed by the details of a specific domain of applicability—for example, the aforementioned FVL, where rotor placement and its impact on the fields needs to be accounted for in maturing the models to the point where they are suitable for transition. The PI did note that finding time to pursue the ILIR was challenging due to other demands on his time.
Project: Stochastic Geometry and Fields of Randomly Dispersed Sensors
This project focuses on using randomly distributed sensors for sensing and detection over an operational zone. The main questions investigated concern how many sensors are required to achieve specific mission goals, how effective are sensors in detecting signals of interest, and how impactful is the sensor sensitivity. The research relies on using Monte Carlo simulation to estimate average signal propagation characteristics. The project replaces a single, exact city model with a random ensemble built from specifications for block size, street widths, and building dimensions. The researcher proposed that a random ensemble approach to modeling city RF propagation could provide insight and efficient experimentation. Using four city models, the project generates random permutations of the cities and places sensors on the generated grid. These are used to determine the probability of signal detection by a single sensor and the number of sensors needed to provide a specified average number of detections per transmission. The PIs also computed the sensor density per square kilometer. Conclusions regarding required sensing density were obtained for the respective city models.
The study of stochastic geometry for wireless sensing has been a well-recognized research area for some time. This research is basic in nature and quite relevant to the mission of the Army. The idea of using random ensembles of cities for simulation is interesting and relevant because sensing fields need to be designed not for one city’s instantiation but possibly for many in the same mold. Using Monte Carlo simulations is a valid approach to answer the questions regarding sensing coverage.
Stochastic geometry models for wireless networks have been developed over the past 2 decades, and highly sophisticated analysis has been carried out within these models. Channel measurement models have also been extensively investigated for decades. One challenge for the project is to compare and relate the model, approach, and results within the context of the existing work. It is important to assess the additional value being offered by this approach relative to existing approaches. For instance, how do the results of this project compare to commercial offerings such as REMCOM’s Wireless InSite®? Another challenge for the project is to obtain more exposure in the research community. The PI needs to present the work in widely attended, peer-reviewed conferences to obtain feedback on this work.
CERDEC Crosscutting Findings
For the project Evaluation of Synthetic IR Environments for Training Deep Neural Nets and other projects presented during the review, including Multimodal Content Delivery Using Resource Aware Artificial Intelligence, the PIs working on applying machine learning techniques to practical problems need to interact with one another extensively to share ideas, best practices, and research results. In general, machine learning needs to be regarded as a key technique to be widely studied within the ILIR program.
It would also be very interesting to connect the antenna propagation modeling in the Stochastic Geometry and Fields of Randomly Dispersed Sensors project with the Compressed Sensing for Massive Multiple Antenna (MIMO) and the Radiating Structure and Loading Methodologies versus m and e Reliant Approaches for Conformal Antenna System Design projects. It is important to investigate how the detailed modeling of antennas and communications systems needs to be reflected in higher-layer network models.
There are possible synergies between the research being performed on the Radiating Structure and Loading Methodologies versus m and e Reliant Approaches for Conformal Antenna System Design project and two other CERDEC ILIR projects—Compressed Sensing with Massive Multiple Antenna (due to the interactions between the input and output combining algorithms and the antenna design) and Stochastic Geometry and Fields of Randomly Dispersed Antennas (due to the reflector adjacencies that may occur as a result of random sensor placement). The PI did not note any collaborators in the presentation; however, a related ILIR is cited in the presentation entitled A Mathematical Transformation from Metamaterial to Matching Circuit for a Conformal Antenna, which was presented during the review conducted by the Panel on Review of In-House Laboratory Independent Research in Physics at the Army’s Research, Development, and Engineering Centers, and it would be surprising if these two researchers were not collaborating, at least informally.
The AMRDEC project Chaos in Optimal Communication and Radar Waveforms may benefit from interactions with CERDEC researchers. CERDEC builds tactical communications systems that attempt to serve the needs of the modern, data-rich battlespace. Meeting those needs is constrained by achievable bandwidth capabilities of existing communications systems. The feasibility of a direct-sequence spread-spectrum communications system needs to be jointly investigated by collaborations between AMRDEC and CERDEC.
There may also be synergism between the AMRDEC project, Optimal Signal Detection Using Mutual Information, and the CERDEC project, Evaluation of Synthetic IR Environments for Training Deep Neural Nets. The PIs may wish to consider how information-theoretic tools can improve DNN training.
On a broader scope, the ILIR program needs to be strongly supported for a number of important reasons. First, the ILIR program in fundamental research is an important means by which top-quality technical personnel can be attracted to and retained by Army RDECs. Second, the ILIR program serves as a crucial link between the RDECs and the larger research community. In order to capitalize on new ideas and trends, the RDECs need in-house technical expertise that is deeply engaged with the technological world. Third, the RDEC ILIR centers assume a unique research role in that they are more directly connected with the pressing technical challenges faced by the Army. This enables them to provide the seed corn that can lead to important applied research and beyond.
Basic research has a highly beneficial impact on the future of the Army because it allows Army personnel to think creatively and to aim for transformative enhancements in the Army’s capabilities. To ensure the quality of the research program, an external peer-review process needs to be used as much as possible for basic research projects.
To recruit and retain top-quality technical talent, AMRDEC and CERDEC need to aim for a balanced mix of junior and senior personnel. Project durations of 2-3 years are preferable to shorter durations to allow time for exploration, recovering from possible wrong turns characteristic of basic research, and writing up the results for publication in an appropriate venue. Also, mentoring and quarterly status
reviews by RDEC managers would be beneficial, as would mentoring processes in which senior ILIR PIs mentor junior PIs. Junior personnel need to receive appropriate mentorship and guidance from senior personnel in the selection, planning, and execution of ILIR projects.
Interaction with the scientific community in the ILIR researcher’s discipline clarifies the claims of novelty and enhances the impact of the research. In particular, publication in appropriate rigorously peer-reviewed venues germane to the discipline (e.g., Physics Review Letters) needs to be encouraged. Where possible, travel support for scientific meeting attendance and presentations needs to be made available. All the reviewed RDEC ILIR programs discussed in this chapter maintain and extend connectivity with outside academic institutions and industry. This ensures a steady level of intellectual exchange, cross-pollination, and timely feedback, which is necessary to support a high-quality research program. Interaction with senior scientists and subject matter experts elsewhere within the Department of Defense (e.g., ARO and the Defense Advanced Research Projects Agency [DARPA]) would also be beneficial.
In line with this suggestion of outreach to the scientific community, best practices need to be followed in giving technical project presentations. For instance, presenters need to start by describing what problem is being addressed and the goals for the research. They need to discuss existing approaches, give a conceptual overview of the approach taken, and talk about paths explored that did not lead to desired results and those that show promise. They also need to discuss planned milestones and metrics for measuring success. Furthermore, the presenters need to give enough conceptual information so that scientists outside the specific specialty can easily follow the presentation.
The ILIR-supported investigators whose research was reviewed showed awareness of the significant impacts for their RDECs derived from their work. If this ILIR is to prove fruitful, the level of involvement by its researchers needs to be appropriately balanced with the researchers’ other responsibilities.
Recommendation: AMRDEC and CERDEC should provide greater mentorship and oversight to junior staff members.
Recommendation: AMRDEC and CERDEC should use the peer-review process as much as possible for ILIR projects.
Recommendation: To advance greater scientific understandings that will enhance their ILIR projects, AMRDEC and CERDEC should promote interactions of their ILIR researchers with the broader scientific community and greater collaboration with other RDECs and Army laboratories.