OUSD (R&E) CRITICAL TECHNOLOGY AREA(S): Advanced Computing and Software; Integrated Sensing and Cyber; Trusted AI and Autonomy 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 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. OBJECTIVE: Design and develop a parallel system of adaptive machine learning (ML) cue generators that ensures multi-signal and multi-function electromagnetic spectrum monitoring (ESM) systems with the parallel system will properly respond to signal classes with different probabilities of occurrence and importance. DESCRIPTION: In designing future ESM systems, it is recognized that adaptation to the current signals environment is critical to achieve high functionality. Patterns of life recognition systems are being developed that can keep track of the array of signal types observed and their characteristics including, but potentially not limited to, their pulse descriptor word entries and signal inter-relationships/clusters. An important question for ML systems is whether to allow adaptation to the actual electromagnetic (EM) environment by allowing automated retraining of the algorithms, and if yes, how and how often. It is unclear how long the ability to recognize rare but important signals would last if such retraining is allowed to occur. Today most systems use a single cue generator to locate all the signals defined by some criterion present in the current EM environment. What appears to be needed in the future is a system of N such cue generators, all fed by the same wideband data stream and delivering their conclusions to the same prioritizer/scheduler. That unit would then decide how the system's finite local digital signal processing (DSP) resources will be used to best reduce the current data to actionable information. This SBIR topic is designed to begin prototyping such a system of cue generators operating in parallel, first in Phase I by developing the data movement system required, and then in Phase II demonstrating its functionality in a simple setting and begin the integration of a pattern of life system. Phase I proposals need to include evidence that the proposer already has access to an ML implemented cue generator and an understanding of the complexities inherent in building a scaled up to N=4 or more system using only currently available commercial off-the-shelf (COTS) processor cards of CPU, FPGA, or GPU character and 1 server. Systems requiring use of a single class of COTS components or a proprietary ASIC are less desirable but can be considered if a strong case is made for their functional benefit. The proposals should describe a potential architecture for the system, including how to get the signal data in and out without losing accurate track of the time of arrival and dealing with the fact different cue generators may take different times to complete their analysis of the same Vita 49 packeted signal data stream. An experimental lab demonstration for the N=2 case during the Phase I Base is highly desirable as it would inform future Phases proposal. PHASE I: Flesh out the architecture as described in the proposal. Execute a demonstration, at least by simulation, of an N=2 system using 2 copies of the identified ML cue generator trained to recognize different classes of signals. Proposals should describe this demonstration in detail. At most a minority of the proposed Phase I tasking should go toward improving the function of said cue generator. Generate a proposed Phase II plan, emphasizing issues to be addressed in realizing the ultimately large N system case; how the work would evolve; and what to do in the case of a severely limited SWaP system. The Phase I Option, if exercised, should select the hardware required to implement the proposed Phase II plan and begin to work integration issues. PHASE II: Perform an experimental demonstration of an N=2 system fed by a government off-the-shelf (GOTS) or COTS digitizer and complex environment signal generator or a digitally delivered predefined set of digitized signal environments that include both the trained to signals and others that are used as background. Include verification that the cue generators performance does not decay as the number of background signals increases and that abrupt shifts in the signal content does not stall operation. Identify and implement delivery of what information the patterns of life units need to supply to this cue generator unit, e.g., for retraining purposes, as opposed to supplied to the cue prioritizer. Work to include feeding the results from a GOTS pattern of life generator into the prioritizer and integration of the prioritizer with the cueing system in the Phase II Option if it is exercised. PHASE III DUAL USE APPLICATIONS: Expected government use is in systems that are at least reconfigurable for multiple functions. The most likely phase III is hence to do and demonstrate this integration of the parallelized cueing subsystem into an already multi-functional system. Economics is expected to increase the fraction of systems which are built that way in the future. A commercial application is most likely in the tele-com domain in systems to suppress pirate applications operating on commercial infrastructure by links with signals at amplitudes below the legitimate traffic or above the noise floor but with widely different waveforms. They must be detected first if the bad behavior is to be suppressed. Here the N=2, simplified prioritizer case might be sufficient where the first cue generator is for the expected traffic. The second is then for the pirate signals and the latter are routed to a specialized identifier and logging of incidents tracker. REFERENCES: Digital signal processing. Wikipedia. https://en.wikipedia.org/wiki/Digital_signal_processing Complete Guide to Understanding Signal Processing. electronicsforu.com, January 9, 2023. https://www.electronicsforu.com/technology-trends/learn-electronics/signal-processing KEYWORDS: Signals of interest; machine learning; adaptive digital signal processing; resource management; software defined radios; situational awareness
