Resonate-and-Fire Neurons for Radar Interference Detection
Julian Hille, Daniel Auge, Cyprian Grassmann, Alois Knoll · 2022
Radar devices sense the environment and detect range, velocity, and angel of arrival by applying multiple Fourier transformations. However, these calculations are expensive and assume that the data are in memory. Frequency-Modulated-Continuous-Wave sensors are primarily used in the automotive industry but are affected by signal superposition with other sensors. It can introduce ghost targets or increase the noise such that low reflective targets are lost. Inspired by the energy efficiency of Spiking Neural Networks, we show that Resonate-and-Fire neurons are able to encode the temporal radar signal into spikes and use a population of Leaky Integrate-and-Fire neurons to distinguish between the normality and patterns such as interference or saturation. We use simulations to prove the concept and achieve in this preliminary study an average accuracy of 85% by utilizing Back-Propagation Through Time with surrogate gradient learning.