Offload Rethinking by Cloud Assistance for Efficient Environmental Sound Recognition on LPWANs

TL;DR

This paper tackles the challenge of environmental sound recognition on ultra-low-power, batteryless edge devices operating over LPWANs by introducing ORCA, a cloud-assisted framework that keeps the inference pipeline on the edge while leveraging cloud processing to identify the most informative frequency bands. A self-attention-based vision-transformer approach on a low-resolution cloud spectrogram generates a spectral attention mask that guides on-device refinement through multi-resolution spectrograms and a spectral encoding CNN, enabling high accuracy with drastically reduced uplink payloads. An adaptive, resource-aware scheduler balances energy and communication costs, allowing cloud-assisted inference to occur within a single power cycle even under variable LPWAN conditions and potential packet loss. Real-world LoRa testbeds show ORCA achieves up to 80x energy savings and 220x latency reduction with accuracy comparable to state-of-the-art methods, demonstrating the practical viability of cloud-assisted, batteryless environmental sensing at wide scales.

Abstract

Learning-based environmental sound recognition has emerged as a crucial method for ultra-low-power environmental monitoring in biological research and city-scale sensing systems. These systems usually operate under limited resources and are often powered by harvested energy in remote areas. Recent efforts in on-device sound recognition suffer from low accuracy due to resource constraints, whereas cloud offloading strategies are hindered by high communication costs. In this work, we introduce ORCA, a novel resource-efficient cloud-assisted environmental sound recognition system on batteryless devices operating over the Low-Power Wide-Area Networks (LPWANs), targeting wide-area audio sensing applications. We propose a cloud assistance strategy that remedies the low accuracy of on-device inference while minimizing the communication costs for cloud offloading. By leveraging a self-attention-based cloud sub-spectral feature selection method to facilitate efficient on-device inference, ORCA resolves three key challenges for resource-constrained cloud offloading over LPWANs: 1) high communication costs and low data rates, 2) dynamic wireless channel conditions, and 3) unreliable offloading. We implement ORCA on an energy-harvesting batteryless microcontroller and evaluate it in a real world urban sound testbed. Our results show that ORCA outperforms state-of-the-art methods by up to $80 \times$ in energy savings and $220 \times$ in latency reduction while maintaining comparable accuracy.

Figures (17)

• Figure 1: Comparisons of on-device inference, cloud offloading, and cloud assistance.
• Figure 2: Accuracy (left) and energy consumption (right) at various spectrogram resolutions.
• Figure 3: Accuracy of using high- and low-frequency band for ESC10 (left) and US8k (right).
• Figure 4: ORCA cloud-assisted design overview.
• Figure 5: Attention computation for attention rollout.