NOMA Digital Semantic Communication for Neuromorphic Multi-task Edge Inference

This paper investigates the use of spiking neural networks (SNNs) for multi-task edge inference under non-orthogonal multiple access (NOMA) transmission. In particular, we deploy SNN-based encoders on distributed sensing devices to extract and transmit digital features from heterogeneous sensing tasks to a common edge server for joint processing. To mitigate co-channel interference, we propose a power control strategy and a task-oriented successive interference cancellation (SIC) scheme tailored for SNN-based task models. Simulation results demonstrate that the proposed scheme achieves 25% and 5% higher inference accuracy for two concurrent edge tasks compared to conventional artificial neural networks (ANNs).