SBIR-STTR Award

Existing satellite-based systems leverage infrared technology to track missiles during the boost phase by detecting heat generated by the missile engine. Limitations in rapid geographic positioning combined with a very short boost-phase duration makes intercepts very challenging. During the critical post-boost and subsequent midcourse phases, when the missile becomes much colder, missile tracking signatures can be lost. A Space Sensor Layer comprised of Low Earth Orbit (LEO) satellites equipped with radar or optical-based sensors could detect, track, and distinguish warheads from decoys and debris during midcourse and could complement an infrared-based system. Compatibility issues and information exchange delays, however, could affect performance since LEO satellites are limited in size weight and power. Machine Learning approaches for analytics extraction provide significant performance and power saving advantages over conventional methods. Lucid Circuit, a Los Angeles-based startup, is developing an adaptable Artificial Intelligence microchip called AstrumTM for cognitive aerospace applications. By enabling machine learning in LEO space platforms, many risks and challenges are mitigated. Only the resulting critical analytics are transmitted - making them available to other satellites and strategists on the ground in real-time. LEO satellites will be able to perform distributed cognitive analytics while ensuring intelligent data attestation.

The capabilities investigated in the CoDAS (Cognitive Distributed Analytics in Space) project are expected to enable real-time cognitive local and distributed analytics processing in space platforms. This capability will allow space-based platforms, constellations and analytics verticals to autonomously adapt to evolving missions and self-heal while ensuring deterministic scheduling and functional capabilities. Communication and resilience to jamming effectiveness will be enhanced while improving the sensitivity, efficacy and efficiency of in-orbit machine sensor platforms. The capabilities investigated in the project are expected to cognitively improve communication and resilience to jamming effectiveness while improving the sensitivity and efficiency of LEO, MEO & GEO machine learning-based sensor analytics in standalone and distributed in-orbit processing settings. ---------- This program will aim to address the size, weight, power and cost challenges associated sensor data processing and analysis at the edge that today limit platform capabilities in Low Earth Orbit (LEO) as well as other aerospace and airborne settings. Existing platforms are limited by storage and downlink capabilities, power budget, and delayed availability, data assurance, sensor node resiliency as well as adversarial threats. Aerospace platforms need to reliably perform on-board machine learning and signal processing while operating within a constrained size, weight and power envelope in order to sustain the surge in demand for analytics. Moving more processing power on board, adjacent to sensing payloads, reduces the pressure on available communication bandwidth. This, in turn, supports scaling to larger satellite deployments which ultimately increases the reliability of the aerospace analytics platforms. Aerospace platforms with the flexibility and efficiency to execute state-of-the-art machine learning algorithms in-flight will enable the deployment of autonomous, adaptable and resilient assets that can be re-tasked as mission requirements evolve. This cognitive processing combined with sensor innovations will enable new mission capabilities. Satellites will be able to act as cognitive distributed sensor networks. Sensors across several satellites will be able to concurrently track different properties of a set of targets (such as missiles) and provide the relevant analytics in real-time – this would be a case of distributed analytics based on multi-modal sensor data. In applications involving sensor networks, node communications can become limited by severe capacity, energy or active adversarial constraints. The distributed sensor networks provide an application setting in which distributed optimization tasks (including machine learning) can address the aforementioned challenges through distributed training and prediction.