Topic

Modern military operations are conducted in contested RF spectrum environments, where adversaries’ actions produce a growing number of complex spectral signatures. The operational need for automation of RF signal classification and anomaly detection using ML techniques addresses threat detection, pattern recognition, and predictive analysis within C5ISR systems – which currently require a manual, human-in-the-loop process. With an exponential increased demand for automated signal processing and the limited manpower available with this expertise, this capability will dramatically increase the capacity of SIGINT analysis and processing functions which are critical features for C5ISR systems. Standard machine learning approaches are often insufficient as they require massive, labeled datasets that do not exist for future conflicts and frequently produce "black box" solutions that are difficult for commanders to trust, interpret, or certify. This topic seeks an alternative approach for an ML-based solution that can classify RF signals based on similar patterns of emission and provide the flexibility of integration on COTS platforms to facilitate integration with existing C5ISR systems. The desired ML capability will be modular and scalable to fine-tune classification inference results for varying base modulations and frequency ranges. The proposed solution and approach must demonstrate the following critical attributes: 1. Dominant Performance: The system must generate run-time identification and classification inferences that are demonstrably faster as compared to expert SIGINT analysts in complex RF environments scenarios. 2. Human-Interoperable: Generated results must be transparent and understandable, composed of modular components (i.e., not a monolithic neural network). Commanders must be able to understand the "why" behind the recommendations. 3. Scalable: The approach must be capable of scaling from tactical engagements (e.g., individual flight combat) to operational-level scenarios involving thousands of assets across multiple domains (air, sea, land) and extended time horizons. 4. Computational Efficiency: The solution must operate effectively on modest computational footprints (e.g., single or small-cluster CPU-based workstations), avoiding reliance on cost-prohibitive, large-scale GPU clusters for its core training and inference loops. 5. Improved Accuracy: The approach must describe the methodology and metrics for improved accuracy over time as additional data sets, results, and resources are employed. 6. Delivery & Integration: The approach must describe the Continuous Integration / Continuous Deployment methodology to include automation for rapid development changes, testing, and containerized deployment. This solicitation is for a Direct to Phase II (D2P2) award. Offerors are expected to have already achieved significant technical maturity and be prepared to demonstrate existing capabilities upon request.