Semantic Split Inference for Remote Modulation Recognition

2026-10-05

AI Summary

This study addresses the trade-off among sensing node complexity, reporting overhead, and recognition accuracy in remote modulation recognition by proposing a channel-aware semantic split inference framework. The proposed method decouples the splitting depth from the reporting length as independent design variables, enabling sensing nodes to transmit compact semantic features over noisy links while delegating final classification to an edge server. Furthermore, channel-aware modeling is incorporated to facilitate end-to-end joint training optimization. Experimental results demonstrate that the proposed intermediate-layer splitting scheme achieves an optimal balance among model size, computational overhead, latency-energy consumption, and recognition accuracy. Consequently, this work provides an efficient paradigm for distributed intelligent sensing in resource-constrained scenarios.

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.

Problem

• Remote Modulation Recognition
• Semantic Split Inference
• Complexity-Accuracy Trade-off
• Edge Computing

Innovation

• Semantic Split Inference
• Automatic Modulation Recognition
• Channel-aware End-to-end Training
• Edge Computing
• Split Depth Optimization