May 22, 2019
This discussion covered selected SBIR topics that included some type of Artificial Intelligence or Machine Learning in the description.
More information:
Export Control:
Many of these topics include this statement:
The technology within this topic is restricted under the International Traffic in Arms Regulation (ITAR), 22 CFR Parts 120-130, which controls the export and import of defense-related material and services, including export of sensitive technical data, or the Export Administration Regulation (EAR), 15 CFR Parts 730-774, which controls dual use items. Offerors must disclose any proposed use of foreign nationals (FNs), their country(ies) of origin, the type of visa or work permit possessed, and the statement of work (SOW) tasks intended for accomplishment by the FN(s) in accordance with section 3.5 of the Announcement. Offerors are advised foreign nationals proposed to perform on this topic may be restricted due to the technical data under US Export Control Laws.
AF192-028 EO/IR-Specific State of the Art Machine Learning
OBJECTIVE: Develop EO/IR-specific state of the art machine learning method(s) for improving utility of ISR sensor products to provide enhanced interpretability and extend range performance over visual image assessments.
DESCRIPTION: The Sensors Directorate of the Air Force Research Laboratory (AFRL/RY) and AF Life Cycle Management Center have been partnering on sensor technology research and development for ISR applications covering a range of passive and active EO/IR sensor concepts. Relevant research has the potential to support the DoD in manned and unmanned airframes. For this topic, the research should focus on the capability of performing National Image Interpretability Rating Scale (NIIRS) 5 or better level tasks on NIIRS 4 imagery where the images acquired are degraded due to low signal to noise ratio, atmospheric conditions, etc. These tasks are to be performed on passive single band imagery. The rapid expansion of research in the areas of state of the art machine learning, artificial intelligence, and deep learning open the possibility of improved image interpretability at a given imaging range, as well as the potential for further extending range performance of EO/IR sensor systems. One major challenge is acquiring or accurately modeling datasets for training and learning. Acquired datasets would have to be labeled after the collection to aid with training and learning. Collection of training and learning data will be provided by the offeror, no government facilities, equipment, etc. will be provided. An additional, but related, challenge is that training data may only be collected over a pristine or limited set of conditions. It is important to understand how training datasets and machine learning transfers to other data collection ranges, environmental conditions, and even target variations. This area of research is known as transfer learning. Performance metrics will focus on accomplishing NIIRS 5 or better tasks on NIIRS 4 imagery with 75% accuracy as a threshold and 100% accuracy as an objective.
AF192-049 Data Science Based Approaches for Modeling Optimal Controls and PHM for Power and Thermal System Components.
OBJECTIVE: Develop model order reduction strategies for electrical power and thermal components that retain high accuracy with reduced computational time for real time control and health monitoring applications.
DESCRIPTION: Design and Verification of advanced propulsion and electric power controls requires reduced order models (ROMs) that run in real time. Calculation of power utilization, load factors, parameter estimates, and control mechanisms is a challenge as accurate, predictive algorithms may take an order of magnitude more time to execute (versus clock time) to reach a stable solution. It is desirable to reduce this computational burden to allow real time use of novel algorithms in control systems Propulsion Health Monitoring (PHM), and power and thermal architectures. It has been demonstrated that computational statistics (CS), machine learning (ML), and related artificial intelligence (AI) techniques that access large data sets can learn constrained domains without explicit programming. They can capture a large percentage of the requirements for accuracy of complex component and system loop (feedback) models of waste heat and transient power flow for electric actuation and high electric power usage components such as diode/fiber lasers. Use of key AI techniques such as CS and ML have the potential to reduce model/algorithm execution time by one or more orders of magnitude compared to the state-of-the-art.
AF192-057 Dynamic, Risk-Based, Situational Awareness and Response
OBJECTIVE: Develop a system for integrated threat detection, classification, and situational awareness considering data associated with risk relative to assets, providing capability for fixed and mobile asset security leveraging all available information.
DESCRIPTION: The USAF is tasked with securing both fixed-site facilities and mobile assets against a multiplicity of potential threats, including conventional weapons (rockets, artillery, mortars (RAM)), and shorter-range weapons (small arms, rocket-propelled grenades (RPGs)). Enemy use of vehicle-borne explosive devices and other improvised weapons are also of great concern. An emerging threat to site security is the rapid proliferation of low- cost UAS (“drones”). Weaponized drones, particularly operating in coordinated swarms, pose an immediate risk. Their low cost and commercial availability have enabled an increasingly deadly role in theater. Other threats include cyber-directed attacks against such infrastructure as the power grid or water supply. In some location, biological weapons, severe weather, wildfires, civil unrest, and many others threats can also pose significant operational hazards to secured facilities………
This topic envisions utilization of emerging technologies in the realm of machine learning. Such systems are capable of improving their predictive accuracy (assessment of “truth”) as they are provided with more and more data. In the past few years, computational hardware required to utilize these tools has advanced to the point where such systems can be deployed in tactical command centers with minimal additional facilities requirements. Some of these systems utilize graphics processors (GPUs), which have been in wide circulation for about a decade. Additionally, new types of processor architectures are being developed specifically for deep learning frameworks. Coinciding with the advances in hardware, software tools are available that make development of applications readily accessible.
DTRA192-001 Artificial Intelligence for the Testing and Analysis of Energetic Fills
OBJECTIVE: DTRA seeks an A.I. enabled operating system to be applied to energetic materials (explosives) testing chambers for a modern, efficient, and rigorous capability to quickly and thoroughly characterize such materials and provide a database with automated analytical capability to show, e.g., historical performance trends, performance comparisons, variations in conditions between data points, estimate scaling performance, estimate data point error, etc.
DESCRIPTION: The envisioned product should be able to communicate with and control a wide variety of instrumentation and diagnostic systems such that environmental, calibration curves, and other metadata information can be automatically logged and correlated with the data from measuring the explosive material performance. Such instrumentation may include both analog and digital components such as thermocouples, pressure transducers, spectrometers, cameras, oscilloscopes, etc. Data from all connected systems should be, or be able to be, automatically correlated with time and space information. The system should be able to monitor connected systems to allow, e.g., self-health checks or report when recalibration may be required. The developed operating system should allow automated or user-defined sequencing and test procedures to be defined and applied. The system should be able to provide automated analysis of tests to provide immediate feedback such as whether a test article fell within an expected performance range or whether all instrumentation captured an event successfully. Additionally, the system should eventually be able to suggest appropriate testing to fill in data gaps in its database using rules such as from Design of Experiments. Modern data visualization techniques, such as virtual or augmented reality, should be supported such that events could be replayed with data layers from non-visual sensors available for overlay (e.g. be able to see a color layer representing pressure or temperature overlaid on a 3D movie of the event). More-over the system should be able to apply intelligent analysis to the database developed in order to predict baseline performance of new compounds, scaling performance of tested compounds, etc. The developed product should not be specific to a single chamber but should instead be able to be applied to various chambers given appropriate setup.
DTRA192-002 Deep Reinforcement Learning Methods and Simulated Learning Environments for Counter-Unmanned Aircraft Systems (C-UAS) Applications
OBJECTIVE: To develop advanced artificial intelligence (AI) for competitive offensive as well as defensive unmanned systems to counter hostile threats that lead to degraded performance. This topic seeks development of (1) computational methods that use a single reinforcement agent to solve complex, multi-task problems and (2) simulated learning environments that can be used to train as well as to evaluate putative solutions.
DESCRIPTION: AI has recently been described by experts within the U.S. Department of Defense Autonomy Community of Interest (COI) as “the next arms race”. The implications of adversarial use of AI and its successively greater incorporation into unmanned and autonomous systems remain both unknown and a considerable source of apprehension. Exponential growth in the C-UAS industry, for example, points to mounting concerns regarding use of drones in civilian and military settings, and particular concern surrounds the potential for highly coordinated and disruptive attacks mediated by groups of small (i.e., “swarming”) unmanned systems. Increased domestic and global investment in such technologies also enhances the probability that swarming systems and other C-UAS technologies could be exploited by the adversary to thwart U.S. military operations for which UAS are preferentially used.
MDA19-009 Self-Coding Cyber Fixes
OBJECTIVE: Automate fixes for cyber vulnerabilities in source code based on results from commercially available scanning software.
DESCRIPTION: This topic seeks innovative modern methods or technologies that alleviate the burden associated with fixing cyber vulnerabilities in already developed source code. This topic does not seek to identify the vulnerabilities as there are many commercial-off-the-shelf tools that perform this function. A listing of such tools is included in the references. The idea is that once scanning software has identified vulnerabilities in source code through some sort of report, innovative algorithms based on machine learning or artificial intelligence methodologies can take those results and automate code fixes. Proper documentation and traceability is critical to this process as developers will need to understand what and how a cyber-vulnerability is being fixed in the source code base. Cyber hardening code is a top priority of the Department of Defense (DoD) and such a tool would save millions of dollars benefiting not only the DoD industry but many commercial entities as well. There are many categories associated with cyber vulnerabilities that will need to be addressed and each category comes with its own unique challenges in how to address them. Major categories that will need to be covered are: Buffer Overflows, Injection Vulnerabilities, Sensitive Data Exposure, Broken Authentication and Security Misconfiguration just to name a few. Different types of source code will also need to be addressed (i.e. C, C++, Java, Python, etc.) as well as a discussion on how different compilers may affect the end result.
NGA192-002 Machine Learning for Enhanced Remote Nuclear Detection
OBJECTIVE: Leverage advances in machine learning technologies and deep learning to address specific defense challenges in nuclear security.
DESCRIPTION: Application of Machine Learning (ML) to nuclear security is a recent development.[1,2] In particular, efforts have been directed at using ML in anomaly detection (i.e., energy spectroscopy), as well as in nuclear material classification using spectroscopic features.[3,4] Both of these examples use supervised learning approaches, in which a classifier is derived from training sets to accurately map future input vectors. Reinforced learning approaches such as unsupervised methods may be applicable to distributed sensors scenarios. The main challenge to the realization of this approach is the paucity of representative training data and the poor-quality test cases. Novel methods in material tracking and identification are needed.[5] A hybrid approach using supervised learning for spectral identification of a particular sensors combined with unsupervised anomaly detection of globally distributed networks of sensors could offer a breakthrough approach to the solution of these unique challenges of nuclear security.[6]
NGA192-004 Simulating Coherent Activity Data
OBJECTIVE: Create high fidelity activity simulations of dynamic targets to enable training and measurable testing of large-scale analytics and collection strategies.
DESCRIPTION: In today’s machine learning and artificial intelligence landscape, labeled data is king. In order to augment the highly labor-intensive data labelling process, or the expensive prospect of conducting live ground-truth experiments, there is significant research into the ability to incorporate simulated data without sacrificing robustness [1]. While this research shows promise for static imagery, there is currently a dearth of investment in simulating data with high fidelity dynamic targets and actors.
As ISR sensors and collection strategies incorporate machine learning and artificial intelligence in their aim to capture fleeting targets and anticipate adversary movement, simulated dynamic scenes with accurate target behaviors will be required for training and scoring performance not simply in static target detection and classification, but in activity identification, modeling, and maintaining chain of custody. Characterization of algorithms, analytics, and collection strategies in such simulated data would not only advance performance by providing suitable training examples for machine learning approaches, but also streamline the government’s ability to benchmark, discriminate, adopt, and apply these emerging technologies.
The scale of activity relevant for most ISR systems is human actions and interactions as observed through vehicular movement. In current modeling and simulation, a scenario of actors, times, locations, and movements is developed through subject matter expert interviews and a painstaking manual process or through agent-based simulations of NGA-10generalized traffic flows with software such as Simulation of Urban Mobility (SUMO) [2] and MatSim [3]. While these tools construct network flows in a manner sufficient for city planning or epidemiology studies, the underlying statistics of actors transiting to and from locations are artificial and consequently insufficient for the evaluation of ISR tracking, analytics, or collection strategies.
To achieve the required fidelity, performers would be asked to create large scale scenarios in which the underlying statistics of simulated human mobility can be verified against open source empirical observations such as, but not limited to, aggregated cell phone GPS, maritime AIS, Uber movement, relevant department of transportation sensors, or even higher level data sources such as periodic news reporting of object/actor locations or documented analytic models. The performers would not be responsible for accurate degradation of the movement data as an ISR constellation and algorithm would sample it, but instead on the quality of the underlying behaviors and simulated mobility patterns as measured by a statistical comparison between the empirical distributions of activities and their simulated counterparts.
N192-048 Automatic Track Generation Micro Preprocessor for Dismounted Electronic Warfare
OBJECTIVE: Develop an innovative and operationally suitable solution for Electronic Warfare Systems (EWS) Programs of Record (PORs) data pre-processing at the tactical edge that, enabled by artificial intelligence (AI) and machine learning (ML) algorithms, must be able to process vast amounts of raw data to detect, track and recommend actions on signals of interest in a complex electromagnetic environment.
DESCRIPTION: Marine Corps Systems Command (MCSC) provides dismounted EWS for geo-locating, direction finding and countering threats on the ground and in the air. Currently these systems collect vast amounts of raw and unfiltered data that describe signals from electromagnetic sources in the form of individual pulse descriptor words (PDW) – potentially billions per minute. The raw data is then transmitted back to the tactical operations center (TOC) where it is downloaded, processed and analyzed to identify objects and track targets of interest. The sheer amount of raw data being transmitted over limited bandwidth and post-processed at the TOC is not conducive to real-time signal of interest tracking and hinders the Marines’ ability to react to potential threats. The advent of advanced AI and ML techniques, such as Long Short-Term Memory (LSTM) networks, and the availability of open- source software tools (e.g., TensorFlow) and off-the-shelf processing capabilities (e.g., NVIDIA) provides opportunity to more efficiently and effectively process electromagnetic signal data by enabling preprocessing and filtering at the antennae sensor. The ability to detect composite tracks in real time at the tactical edge will reduce the amount of data necessarily transmitted and post-processed at the TOC, resulting in more efficient signal analysis and ultimately improved effectiveness of EWS capabilities….
N192-065 Artificially Intelligent Object with Virtual Presentation of Engineering and Logistics Data
OBJECTIVE: Develop a web-enabled object and application that encapsulates three major areas of Technical Data (TD) into an all-in-one TD Virtual Reality (VR) structure able to quickly exhibit different views within a viewer, based on role and responsibility; incorporated with Artificial Intelligence (AI) to capture and make predictive maintenance analysis, detect and address anomalies, and provide a complete traceability of maintenance and part history. An AI auto update of all related TD, as the design is improved, with the ability to identify errors within Concurrent Engineering Logistics Layered Structure (CELLS) is desired. Develop and demonstrate a knowledge hub for capturing and distributing the maintenance predictions and TD updates. Communication must be both visual and verbal.
DESCRIPTION: Navy TD are currently stored in several database management systems, both in digital and paper formats, primarily in government systems: JEDMICS (Joint Engineering Data Management Information and Control System) and TMAPS (Technical Manual Application System). Numerous proprietary PLM (Product Lifecycle NAVY – 54Management) systems are also utilized by engineers and contain Computer Aided Design (CAD) systems. Three areas, all separate products, need to be incorporated into CELLS: Engineering models (a digital representation of the engineering design with sufficient metadata to manufacture the end item, system, component, and or part) [Refs 6, 7, 8, 9]; Interactive Electronic Technical Manual Systems (IETMS) (the maintenance instructions and supply details associated with the end item, system, component, and or part) [Refs 2, 3, 4, 5] presently NSIV (NAVAIR Standard IETMS Viewer); and NATOPs (Naval Air Training and Operating Procedures Standardization) (the operator instructions for Navy aviation pilots [Ref 1]). Currently the Navy pays multiple times for the same data in multiple formats. This proposed system would eliminate that and empower our logisticians to be able to view the needed data in real time, lowering cost and increasing efficiency…
N192-128 Innovative Artificial Intelligence Features to Reduce Signal Dropout due to Clipping, Channel Fading, and Multi-path Interference
OBJECTIVE: Develop and demonstrate an Artificial Intelligence (AI) methodology or deep learning Digital Signal Processing (DSP) soft/firm-ware structure for signal recognition and reception that improves the data rate sustainable in the presence of clipping and strong fading, especially in cases where the fading has a periodic temporal structure.
DESCRIPTION: Movement of either endpoint of a communications link or changes in the multi-path scattering by the environment can force many mobile systems to cope with signals with strongly time dependent amplitude (“deep fading”) on time scales of microseconds to seconds. Wideband systems are often built without analog clutter- automatic gain control and hence often experience clipping and/or small signal inadequacy. They are also especially bothered by multi-path fading since different carrier frequencies are impacted differently by the same changes in the reflector environment. Signal dropout within data links is thus common. Antenna diversity is often used to allow the stronger amplitude signal to be chosen at any given time. But this patch, at a minimum, doubles the hardware costs and has DSP back end complexity issues if the copies are not of perfectly identical quality. Additionally, it does nothing to solve the clipping issue. The need is for a methodology to cope in the back end with signals for which the correctness of the received data (e.g., the bit error rate) fluctuates in time. In many of these settings with longer dropped data intervals, the signal amplitude recovers quasi-periodically; reception can restart but a new link establishment protocol is often required to be run, lowering the time available for actual data before the next fade happens and lowering data throughput. Layered signal reception schemes appear to be needed. One might first process each time segment of signal of adequate amplitude to have at least a marginally acceptable bit error rate and estimate that segment of data to produce both value and accuracy/confidence estimates as part of a probabilistic interpretation. Once some number of intervals have been interpreted, attempts can be made to stitch together the successive intervals, for example, by using machine learning/AI techniques to improve the understanding of each segment by virtue of having the data available from the other time intervals. Methods could include working from both ends of two time segments in order to build up an image of signals by concatenating more and more “on” intervals. Consulting multiple disjointed temporal segments of the same underlying signal will allow reuse of the already collected data and refine our knowledge of the modulation and optimize error correction, while benefiting from a continuous time base and allowing adaptive equalization. This sort of real-time training that improves the continuity of receptions ought to reduce the volume of redundant data transmission required. The AI methodologies developed should be demonstrated using some form of commercial off-the-shelf (COTS) processor working in real time on a high-speed (e.g., > 20 GSps) digital data stream that represent a wide (e.g., >4 GHz) instantaneous bandwidth and in a manner consistent with the principles of open system architectures. Approaches that can work in dense signal environments having substantial spectral overlap between multiple simultaneous signals of substantially different magnitude are especially desirable. Performance should be measured against the case of stationary Rx and TX nodes and a stable communications link between them.
N192-129 Early Detection of Information Campaigns by Adversarial State and Non-State Actors
OBJECTIVE: This SBIR topic will focus on attempts to detect hybrid, “cyborg” information actors, backing, aiding, and amplifying human networks distributing propaganda and highly charged messages. The current state of botnet detection merely identifies automated features such as identical content, identical targets, coordination of message dispersal, and similar measurable enhanced capabilities; “smart” botnets that target individuals (such as super spreaders and super friends) and topic groups are becoming more widespread and are capable of greater impact. Sentiment models alone, and bot detection methods alone, are insufficient to detect and defend against these smart botnets that coordinate and amplify and normalize messages of hate, anger, and violence that are typical of cyber warfare.
DESCRIPTION: Online agitation has resulted in riots, attacks on tourists, ethnic violence, gender violence, instigation of cyber-attacks, murder, and terrorism (see references for a small list of examples). This agitation is aided and abetted by swarms of coordinated “bots”, “fake” accounts, and online loudspeakers of various types from single influential individuals to platforms like Twitter, Whatsapp, blogs, and YouTube that are subject to algorithmic manipulations, often combined with social engineering. Volatile content is combined with other types of messaging to exploit crises and create conditions of panic, uncertainty, and hate. Military missions are increasingly under attack by propaganda, distortion campaigns, and influence operations crafted by state and non-state actors to undermine social trust and diminish the military’s ability to control its own messages. Further, online agitation creates very real dangers in situations of crisis such as disasters and police actions where the military must deploy to secure the safety of civilians. State-backed adversaries have invested in artificial intelligence (AI) and data mining technologies to craft sophisticated “botnet armies” and other stochastic manipulations, the better to support human propagandists and online agitators. These need to be identified and assessed for vulnerabilities and impact; guidance for counter-measures would be the next needed step….
N192-132 Accelerating Knowledge Acquisition for Close Combat Warriors
OBJECTIVE: To develop an adaptive training system that leverages advances in artificial intelligence and decisions sciences, and incorporates commercially available educational technologies that align with military systems (e.g., Moodle), to accelerate the acquisition of knowledge and increase learning gains with a specific focus on close combat-related tasks.
DESCRIPTION: Rote or mass learning is critical for developing foundational knowledge to support higher order decision making. However, current military education technologies and methodologies are focused on industrial age vs. information age methods of learning. A convergence of key enablers exists to pivot away from the mass industrial age of training and education towards a tailored education and training approach by exploiting the availability of ubiquitous computing, advances in machine learning, and science of learning. Furthermore, opportunities exist that are ideal candidates for use of technologies and approaches (e.g., students awaiting the start of a training course)…
A19-116 Integrated 2-color thermal polarimetric sensor and deep neural network system for artificial intelligence and machine learning (AI&ML) based automatic target detection and identification
OBJECTIVE: Develop a 2-color mid and long-wave infrared (MidIR and LWIR) thermal polarimetric camera system with incorporated artificial intelligence and machine learning (AI&ML) capability for enhanced target detection and identification.
DESCRIPTION: During the past decade two different technological areas have advanced significantly, i.e., thermal polarimetric camera systems and AI&ML capabilities for data analysis and exploitation. Currently, DoD spend many tens of millions of dollars per year developing and testing thermal sensor systems designed for 24⁄7day/night surveillance capabilities for a wide variety of tactical scenarios, e.g., detection of buried landmines and IEDs, identification of camouflaged/hidden targets, and night-time facial recognition.[1-4] The advances in AI&ML are driven by new algorithms, notably deep neural networks (DNN), and the maturation of graphical processing unit (GPU) technology optimized for intensive matrix computations. The latest AI&ML algorithms can be trained relatively quickly on low cost GPUs to perform inference on GPUs in real-time. In particular, deep convolutional neural networks (CNN) have demonstrated their potential for accurate object detection and classification. [5-8]
In order to exploit these advances in polarimetric imaging and AI&ML, we propose the development of an “integrated” multimodal thermal imaging and data exploitation system designed to provide “real-time” scene understanding and situational awareness. Such a system would greatly reduce the time and cost required to bring soldier specific image based solutions to the battlefield.
A19-125 Advanced Machine Learning Target Recognition in Munitions
OBJECTIVE: Apply Machine Learning/Artificial Intelligence to target recognition algorithms in gun launched munitions.
DESCRIPTION: The Army requires advancement in Autonomous Target Recognition (ATR) algorithms for seekers in gun launched applications. Currently, seekers are capable of target detection in low clutter environments. To field a fully effective weapon that is also safe for use in conditions where there is high fratricide or collateral damage concern, the ability to discriminate between target types and between friend and foe rapidly (within minutes) and under extremely dynamic conditions is required. This topic will apply advanced and innovative machine learning and/or artificial intelligence to current and future target sensor packages that will be used in artillery, tank and mortar munitions among others. This includes but is not limited to new algorithmic approaches and/or sensor fusion approaches to improve ATR capability at extended slant ranges (3-7km), and while searching large Field of Views (FOV) (up to 3000m radius). The ability to conduct ATR in relatively inexpensive (<$10K unit at 1000 units/year) seeker architectures is critical. The munitions will experience high shock (up to 45,000 g’s) throughout a range of temperature extremes (- 25 to +145 degrees F operating range). The algorithms shall be capable of operating on emerging commercial GPU products suitable to 155mm artillery SWaP-C constraints.
Detailed requirements will be provided after contract award.
A19-126 Advanced Machine Learning for Non-Destructive Testing
OBJECTIVE: Apply Machine Learning/Artificial Intelligence to aid in interpretation of radiography inspection results during non-destructive testing.
DESCRIPTION: The Army relies on radiography inspection (e.g. x-ray and neutron) for non-destructive testing of munitions during production and special investigations. Interpreting the visual results of the inspections is a ARMY-49challenge and requires highly trained individuals (Level III Radiographers) to determine what, if any, problems actually exist. This topic will apply advanced and innovative machine learning and/or artificial intelligence to current and future non-destructive radiography inspection methods that use electronic imaging to identify defects and aid the operator in proper and timely interpretation of the results. As this technology is meant to be incorporated in a production line, the expectation is that it will support three-dimensional inspection and interpretation of defects at a production rate of up to 1 unit per minute, and items up to 6.5 inches in diameter. Defects include cavities, porosity, piping, voids, gaps, low density, annular rings, cracks and inclusions ranging from 0.002” to 0.020”. The technology must reside on a standard computer system linked to the inspection equipment and receive the electronic images from the radiography system. Specific interface requirements will be provided after contract award. This topic will also develop and deliver the output screens that provide the proper data and information that a Level II radiographer is trained to understand.
A19-129 Advanced Signal Detection and Characterization Utilizing Artificial Intelligence (IL)/Machine Learning (ML)
OBJECTIVE: Design and build an electronic signal detection and characterization unit that utilizes artificial intelligence (AI) and machine learning (ML), that can perform continuous monitoring of the electromagnetic spectrum (EMS), and that can provide signal characteristics to an interface.
DESCRIPTION: Recent advances in the computing world has allowed for algorithmic advances in the detection and characterization of the electromagnetic spectrum (EMS). Specifically, the incorporation of such things as neural networks and training processes has elevated artificial intelligence (AI) and machine learning (ML) as key innovation areas for detecting, characterizing and cataloging highly complex signal types in the EMS. The current effort would mature these AI/ML concepts to develop a signal detection and characterization system for electronic signals. The unit would be able to detect and characterization various signal types and modulations. It would also provide performance and monitoring tools to provide real-time feedback to operators. The incorporation of data analytics for validation and visualization would be included in the unit. The system would follow a Modular, Open Systems Approach (MOSA) to allow integration into a variety of Army systems. The MOSA approach would also provide extensible ML and Deep Learning (DL) functions to expand upon key features and signal types. The system would contain only Commercial, Off-The-Shelf (COTS) products.