N242-096 TITLE: Context Aware Data Stream Pre-processor for Time-Sensitive Applications

OUSD (R&E) CRITICAL TECHNOLOGY AREA(S): Advanced Computing and Software; Integrated Sensing and Cyber; Trusted AI and Autonomy

OBJECTIVE: Develop a general context aware, self-learning pre-processing solution to systematically resolve a high throughput Radio Frequency (RF) data stream across distributed systems with resource limitations for time-sensitive applications.

DESCRIPTION: Domain specific data sources generate increasing amounts of information that require ultrafast processing for time-sensitive applications. This is particularly true for ultra-wideband signal processing across the RF spectrum where bandwidths are considerably wide (e.g., several GHz). As data streams continue to expand in throughput, however, the volume of inputs for such applications can quickly overwhelm and exceed a system’s storage and processing capacities. Technology advancements are required to distinguish the most significant inputs relative to an application from within high throughput RF data streams, and to efficiently allocate limited resources (e.g., storage, compute, power) for further analysis of the highest value data as part of a larger processing chain. More simply stated, systems supporting data-heavy, time-sensitive applications require a pre-processing capability to determine what data should be stored (both short term and long term), what data needs to be processed immediately, and how to efficiently allocate resources accordingly.

This SBIR topic seeks a general context aware, self-learning pre-processing solution to systematically resolve a high throughput RF data stream across distributed systems with resource limitations for time-sensitive applications. For any given application the pre-processor should be context aware in order to value input data as appropriate, presumably in part by extracting features and matching inputs against elements in a library of prioritized items and/or by detecting anomalous inputs within the data stream, while continually learning and improving its ability to prioritize inputs for processing. Distributed, networked heterogeneous systems supporting the same application should be able to benefit from diffused learning updates of individual nodes. Innovative approaches to determining high value data are also encouraged. Once the data of greatest importance to a time-sensitive application is identified, the pre-processing solution must determine how to allocate system resources and efficiently make the data available for processing subject to any limitations on storage, compute, power, and latency. The pre-processor resulting from this effort should be generalizable and scalable across distributed, heterogeneous systems to maximize the potential applications and broad utility of this solution in the RF domain.

PHASE I: Define and develop a concept framework for a context aware, self-learning pre-processor that distinguishes high value inputs from a voluminous RF data stream at the point of ingest. Conceive and mature a scheme for resource allocation to support ultrafast processing with consideration to constraints on storage, compute, power, and latency. Provide measures of effectiveness, as well as attainable performance characteristics. The framework will need to be generalizable and extensible across a distributed set of heterogeneous hardware systems, with a proposed design for the hardware and software architectures that supports tip and cue of heterogeneous systems to augment processing-related capacities of any individual system as necessary. The design should include a summary of any storage, computing, and power requirements for administering this pre-processor relative to latency requirements. The feasibility of the concept will be established through modeling and simulation. Include, in a Phase II plan, the initial design specifications and capabilities description to build a prototype in Phase II.

PHASE II: Fully develop, verify, and validate a prototype pre-processing solution that demonstrates context awareness, self-learning, and an ability to perform the desired functionality on high throughput RF data streams. Design the prototype to distinguish high value data and then allocate storage and compute resources as part of a larger RF processing chain. Demonstrate the design performance through modeling and physical testing over a range of voluminous RF data streams devised to test processing capacities with and without the pre-processor in place. Use evaluation criteria and results to refine the prototype for an initial, generalizable, scalable design that supports domain specific, time-sensitive applications. Develop a Phase III plan to transition the technology to a system that can be acquired by the Navy.

PHASE III DUAL USE APPLICATIONS: Support Navy system integration of the pre-processor, hardware, and software to include validation testing of a demonstration on RF data streams in a relevant environment, employing any lessons learned from the Phase II evaluation. Incorporate the pre-processor into multiple domain specific, time-sensitive applications for exhibition of generalizability (e.g., signal processing for wireless networks, indications of new spectrum activity, sensing on autonomous vehicles). The pre-processor from this SBIR effort would support data triage with high throughput streams for time-sensitive applications across the automotive industry, infrastructure, energy, health care, and other domains.

REFERENCES:

1. Geetha, J.; Jayalakshmi, D.S.; Ganiga, R.R.; Kottur, S.Z. and Surabhi, T. "Improvised Distributed Data Streaming Scheduler in Storm." International Conference on Communication, Computing and Electronics Systems. Lecture Notes in Electrical Engineering, vol 733. Springer, Singapore, 2021. https://doi.org/10.1007/978-981-33-4909-4_42
2. Gokarn, Ila; Sabbella, Hemanth; Hu, Yigong; Abdelzaher, Tarek and Misra, Archan. "MOSAIC: Spatially-Multiplexed Edge AI Optimization over Multiple Concurrent Video Sensing Streams." Proceedings of the 14th Conference on ACM Multimedia Systems (MMSys '23). Association for Computing Machinery, New York, NY, USA, 2023, pp. 278–288. https://doi.org/10.1145/3587819.3590986
3. Moso, Juliet Chebet; Cormier, Stepháne; de Runz Cyril; Fouchal, Hacène and Wandeto, John Mwangi. "Streaming-Based Anomaly Detection in ITS Messages." Applied Sciences, 13(12):7313, 2023. https://doi.org/10.3390/app13127313
4. Ngo, Duc-Minh; Tran-Thanh, Binh; Dang, Truong; Tran, Tuan; Thinh, Tran Ngoc and Pham-Quoc, Cuong. "High-Throughput Machine Learning Approaches for Network Attacks Detection on FPGA." Context-Aware Systems and Applications, and Nature of Computation and Communication. ICCASA ICTCC 2019 2019. Lecture Notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering, vol 298. Springer, Cham. https://doi.org/10.1007/978-3-030-34365-1_5
5. Nikolopoulos, Vasileios; Nikolaidou, Mara; Voreakou, Maria and Anagnostopoulos, Dimosthenis. "Context Diffusion in Fog Colonies: Exploring Autonomous Fog Node Operation Using ECTORAS." IoT 2022, 3, pp. 91-108. https://doi.org/10.3390/iot3010005
6. Seeliger, Robert; Müller, Christoph and Arbanowski, Stefan. "Green streaming through utilization of AI-based content aware encoding." 2022 IEEE International Conference on Internet of Things and Intelligence Systems (IoTaIS), BALI, Indonesia, 2022, pp. 43-49. https://ieeexplore.ieee.org/document/9975919
7. Silviya Nancy, J.; Udhayakumar, S.; Pavithra, J.; Preethi, R. and Revathy, G. "Context Aware Self Learning Voice Assistant for Smart Navigation with Contextual LSTM." Advanced Informatics for Computing Research. ICAICR 2019. Communications in Computer and Information Science, vol 1075. Springer, Singapore, 2019. https://doi.org/10.1007/978-981-15-0108-1_41
8. Wang, Runze; Moazzen, Iman and Zhu, Wei ing. "A Computation-Efficient Neural Network for VAD using Multi-Channel Feature." 2022 30th European Signal Processing Conference (EUSIPCO), Belgrade, Serbia, 2022, pp. 170-174. doi: 10.23919/EUSIPCO55093.2022.9909914
9. Liu, Ying; Lita, Lucian V.; Niculescu, R. Stefan; Bai, Kun; Mitra, Prasenjit and Giles, C. Lee. "Real-time data pre-processing technique for efficient feature extraction in large scale datasets." Proceedings of the 17th ACM conference on Information and knowledge management (CIKM '08). Association for Computing Machinery, New York, NY, USA, 2008, pp. 981–990. https://doi.org/10.1145/1458082.1458211

KEYWORDS: Pre-processor; Context aware; Self-learning; Radio Frequency (RF); Data stream; Distributed signal processing; Resource allocation

---