End-to-End Radar and Communication Modulation Recognition with Neuromorphic Computing

The article presents EMRFormer, a novel end-to-end spiking neural network (SNN) architecture designed for automatic modulation recognition (AMR) on neuromorphic hardware. EMRFormer combines an adaptive spike encoder, Integer Leaky Integrate-and-Fire neurons, and integrates spike-separable Convolutional Neural Networks with Spike-Driven Transformers, achieving state-of-the-art accuracy while reducing theoretical energy consumption by over 90%. This model is validated on various datasets and shows significant power efficiency, achieving up to a 5x reduction in power usage compared to traditional GPUs, making it a compelling solution for AMR in resource-constrained environments.