Comparison of Neuromorphic Coding for Distributed Wireless Spiking Neural Networks Based on Mutual Information and Energy Efficiency

Wireless spiking neural networks (WSNNs) enable energy-efficient communication, particularly beneficial for edge intelligence and learning within both terrestrial systems and Earth-space network configurations (beyond 5G/6G). Recent studies have highlighted that distributed wireless SNNs (DWSNNs) perform well in inference accuracy and energy-efficient operation in edge devices, despite the challenges posed by constrained bandwidth and spike loss probability. This makes the technology appealing for wireless sensor networks (WSNs) in space scenarios, where energy limitations are significant. In this paper, we explore neuromorphic impulse radio (IR) transmission methodologies tailored for DWSNNs, investigating various coding algorithms that implement IR modulations. Our assessment employs information-theoretic measures to evaluate performance in terms of transmission efficiency. Moreover, the different neuromorphic coding techniques will be evaluated by considering the energy consumption of edge devices under the same constraints of limited bandwidth and additive white Gaussian noise (AWGN), in order to highlight possible trade-offs between transmission and edge inference requirements.