Sensing‑aided beam prediction leverages auxiliary modalities such as vision and position to cut down the beam‑training overhead of millimeter‑wave (mmWave) systems. The resulting sensing‑and‑inference pipeline, however, can be energy‑hungry, motivating the exploration of energy‑efficient spiking neural networks (SNNs) as an alternative to conventional artificial neural networks (ANNs). In this work we construct several SNN architectures that ingest position‑only, vision‑only, and multimodal inputs, and we benchmark them against compact ANNs. Results show that for position‑only sensing, ANNs consistently outperform SNNs. In contrast, with vision or multimodal data, SNNs achieve higher top‑1 accuracy at comparable energy budgets and require up to an order of magnitude less energy for the same accuracy. Notably, the multimodal SNN reaches 58.08% top‑1 accuracy, surpassing the closest‑energy ANN by 18.11 percentage points. Further analysis reveals that SNN accuracy is limited by the binary communication between layers rather than by model size. These findings highlight the promise of SNNs for energy‑efficient sensing‑aided beam prediction while pinpointing binary inter‑layer signaling as the main accuracy bottleneck.
Review