Low C-SWAP Neuromorphic Radar Classification

Neuromorphic hardware, with its dramatically reduced power consumption and low latency compared to conventions digital computation, offers two possibilities for dramatically improving the performance of radar systems on small, mobile platforms. The first opportunity is to use neuromorphic processing to carry out the signal processing of radar signals, which is usually performed by a combination of DSP and GPU systems.  The second is to take over the DLNN (Deep Learning Neural Network) functions of a radar system, such as classification of targets.  In this proposal, we develop a workflow to tackle both of these challenges.In this work we aim to select key computations, algorithms or neural networks in complex RF and radar signal processing, and recast them for implementation on neuromorphic hardware. Initial steps involve scanning literature for state-of-the-art processing algorithms, followed by the identification of a subset based on importance, latency, and benefit considerations. These selected algorithms are then mapped onto designated BrainChip neuromorphic hardware, leveraging existing DNN-SNN methods with custom inputs where necessary.Furthermore, the study entails the development of a robust benchmarking methodology to evaluate the performance of BrainChip’s Akida neuromorphic hardware in RF and radar signal processing. This methodology integrates existing BrainChip methods, including the system identification process, and incorporates elements of the Disruptive Computing Benchmark methodology pioneered by RTX/Raytheon.Real-time response measurement of the neuromorphic hardware to RF and radar signals will be carried out using hardware in the loop (HWIL) setups. Efforts are made to develop and enhance existing methods for hardware response benchmarking, utilizing both recorded and real data from RTX/Raytheon and BrainChip sources.Performance validation of the neuromorphic hardware in terms of power consumption and timing latency is crucial. In-house methods for measuring neuromorphic power and latency performance will be utilized and enhanced as needed.Confirmation of deterministic outputs from BrainChip neuromorphic hardware is conducted through repeated tests under varied conditions of signal and noise. These tests are compared with conventional processing via modeling and simulation (M&S) environments to ensure repeatability and accuracy.Finally, potential commercial and dual-use neuromorphic applications utilizing BrainChip hardware are identified, with a focus on the defense and aerospace industry, and car manufacturers for ADAS (Advanced Driver-Assist Systems) radar and car-to-car communications. This study serves as a comprehensive framework for the evaluation, validation, and application of neuromorphic hardware and spiking neural networks in RF and radar signal processing, paving the way for future advancements in the aerospace and automotive industries.