Researchers have developed EMRFormer, a novel spiking neural network (SNN) architecture designed for end-to-end automatic modulation recognition (AMR) on resource-constrained neuromorphic hardware. This architecture integrates a spike-driven transformer with adaptive spike encoding and Integer Leaky Integrate-and-Fire neurons to improve information retention and representational capacity. EMRFormer also incorporates spike-separable Convolutional Neural Networks to extract multi-scale temporal features from raw IQ waveforms, achieving state-of-the-art accuracy while reducing theoretical energy consumption by over 90% compared to traditional deep learning methods. Evaluations on a KA200 neuromorphic chip demonstrated up to a five-fold reduction in power usage compared to running on a 3090 GPU or an Orin NX, showcasing its potential for AMR on edge devices. AI
IMPACT Enables more energy-efficient AI applications on resource-constrained edge devices.
RANK_REASON The cluster contains an arXiv preprint detailing a new neural network architecture and its performance evaluation. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
