SBIR-STTR Award

The inability of dismounted radar-based Electronic Warfare System (EWS) systems to process and analyze track data in realtime poses a potential barrier to increased effectiveness during some Marine Corps missions. The Computationally Efficient Deep Learning-Powered EWS Radar Data Preprocessor (CELER) is a system designed to process millions of EWS radar data points per second using Recurrent Neural Networks. Using Xilinxs Zynq UltraScale+ SoC to process the data in realtime allows the system to be carried with backpack EWS systems like SNCs Modi II, with the system measuring less than 12 x 6 x 4 and weighing less than 5 lbs., including the battery. Using an FPGA allows CELER to be efficient in terms of computing power per Watt, particularly over GPUs. The system will learn to model noise typically found in EWS radar data, filtering out noise and classifying signals of interest in the radar data. We will use a physics-based simulation of vehicles and EM clutter to provide realistic training data for our neural networks. CELER is ideal for battery powered systems, including small unmanned air and ground vehicles, where using the latest machine learning-based inference methods would provide more advanced capabilities, but those systems are power constrained.

Benefit:
The CELER Preprocessor allows processing in realtime, and in the field, data that would typically be sent to another location for post-processing. The SNC Modi II EWS system discussed in this proposal, as well as other similar systems fielded by the Marines and other service branches, pose an ideal opportunity for the CELER Preprocessor. These systems collect valuable, actionable data in the field; however, the lack of size, weight, and power (SWaP)-compatible processing systems limits the potential of these systems. The CELER Preprocessor will change that by moving realtime deep learning-based inference to the tactical edge using small, power efficient FPGAS. This represents an opportunity worth potentially hundreds of millions of dollars of the DoD were to retrofit existing, deployed systems to make them AI-enabled.

Keywords:
TensorFlow, TensorFlow, Keras, Electronic Warfare Systems, EWS, FPGA, RNN, Recurrent Neural Networks, GRU

The inability of dismounted radar-based Electronic Warfare System (EWS) systems to process and analyze track data in realtime poses a potential barrier to increased effectiveness during some Marine Corps missions. Tercero Technologies proposes to continue Phase II development of the Computationally Efficient Deep Learning-Powered EWS Radar Data Preprocessor (CELER), a system designed to process millions of EWS radar data points per second using Recurrent Neural Networks. Using Xilinxs Zynq UltraScale+ SoC to process the data in realtime allows the system to be carried with backpack EWS systems like the SNC Modi II, with the system measuring less than 12 in. x 6 in. x 4 in. and weighing less than 5 lbs., including the battery. Using an FPGA allows CELER to be efficient in terms of computing power per Watt, particularly over GPUs. The system will learn to model noise typically found in EWS radar data, filtering out noise and classifying signals of interest in the radar data. We will use a physics-based simulation of vehicles and EM clutter to provide realistic training data for our neural networks. CELER is ideal for battery powered systems, including small unmanned air and ground vehicles, where using the latest machine learning-based inference methods would provide more advanced capabilities, but those systems are power constrained.

Benefit:
The CELER Preprocessor allows processing in realtime, and in the field, data that would typically be sent to another location for post-processing. The SNC Modi II EWS system discussed in this proposal, as well as other similar systems fielded by the Marines and other service branches, pose an ideal opportunity for the CELER Preprocessor. These systems collect valuable, actionable data in the field; however, the lack of size, weight, and power (SWaP)-compatible processing systems limits the potential of these systems. The CELER Preprocessor will change that by moving realtime deep learning-based inference to the tactical edge using small, power efficient FPGAS. This represents an opportunity worth potentially hundreds of millions of dollars of the DoD were to retrofit existing, deployed systems to make them AI-enabled.