Semantic split inference improves remote modulation recognition accuracy cost
Semantic Split Inference for Remote Modulation Recognition
Networking and Internet Architecture
Summary
Recognizing signal types remotely is a balance between how much work the sensing device does, how much data it sends, and how accurate the result is. The authors propose a method where the sensing device sends a summarized report over a noisy connection, and a nearby server finishes identifying the signal type. They show that splitting the work between the device and server in the middle produces the best mix of accuracy and resource use. Their approach lets designers choose how much processing happens at the device and how much data it sends, training the system end to end with the communication channel in mind.
What this means in practice
- •For wireless network operators: Optimize sensing node design to balance energy use and identification accuracy of wireless signals at edge servers.
- •For remote sensing system engineers: Develop efficient signal modulation recognition systems that split processing between sensing nodes and edge servers to reduce latency and data transmission costs.
Authors
Samer Lahoud, Kinda Khawam
Abstract
Remote automatic modulation recognition balances sensing-node complexity, reporting cost and accuracy. To address this trade-off, we propose channel-aware semantic split inference: a sensing node sends a semantic report over a noisy link and the edge server completes recognition. In our model, split depth and report length are independent design variables, with end-to-end training through the channel. We compare the resulting design against basic split placements, which run inference at the edge server or at the sensing node, and against a state-of-the-art collaborative scheme. We assess sensing-node model size, computation, latency and energy against recognition accuracy. We show that intermediate splits give the best accuracy-cost trade-off.