Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FCOS: A Two-Stage Recoverable Model Pruning Framework for Automatic Modulation Recognition

Yao Lu, Tengfei Ma, Zeyu Wang, Zhuangzhi Chen, Dongwei Xu, Yun Lin, Qi Xuan, Guan Gui

TL;DR

FCOS presents a novel two-stage pruning framework for automatic modulation recognition that combines channel-level pruning via similarity-based channel fusion with a layer-collapse diagnosis step using linear probes. This fine-to-coarse approach mitigates the model-collapse risk common in aggressive channel pruning, enabling extreme reductions in FLOPs and parameters while preserving accuracy across multiple AMR benchmarks. Empirical results demonstrate superior compression and robustness compared with existing channel and layer pruning baselines, including substantial improvements on CNN1D, ResNet56, and SigNet50 across several datasets. The method supports efficient AMR deployment on edge devices, advancing real-time spectrum analysis in constrained environments.

Abstract

With the rapid development of wireless communications and the growing complexity of digital modulation schemes, traditional manual modulation recognition methods struggle to extract reliable signal features and meet real-time requirements in modern scenarios. Recently, deep learning based Automatic Modulation Recognition (AMR) approaches have greatly improved classification accuracy. However, their large model sizes and high computational demands hinder deployment on resource-constrained devices. Model pruning provides a general approach to reduce model complexity, but existing weight, channel, and layer pruning techniques each present a trade-off between compression rate, hardware acceleration, and accuracy preservation. To this end, in this paper, we introduce FCOS, a novel Fine-to-COarse two-Stage pruning framework that combines channel-level pruning with layer-level collapse diagnosis to achieve extreme compression, high performance and efficient inference. In the first stage of FCOS, hierarchical clustering and parameter fusion are applied to channel weights to achieve channel-level pruning. Then a Layer Collapse Diagnosis (LaCD) module uses linear probing to identify layer collapse and removes the collapsed layers due to high channel compression ratio. Experiments on multiple AMR benchmarks demonstrate that FCOS outperforms existing channel and layer pruning methods. Specifically, FCOS achieves 95.51% FLOPs reduction and 95.31% parameter reduction while still maintaining performance close to the original ResNet56, with only a 0.46% drop in accuracy on Sig2019-12. Code is available at https://github.com/yaolu-zjut/FCOS.

FCOS: A Two-Stage Recoverable Model Pruning Framework for Automatic Modulation Recognition

TL;DR

FCOS presents a novel two-stage pruning framework for automatic modulation recognition that combines channel-level pruning via similarity-based channel fusion with a layer-collapse diagnosis step using linear probes. This fine-to-coarse approach mitigates the model-collapse risk common in aggressive channel pruning, enabling extreme reductions in FLOPs and parameters while preserving accuracy across multiple AMR benchmarks. Empirical results demonstrate superior compression and robustness compared with existing channel and layer pruning baselines, including substantial improvements on CNN1D, ResNet56, and SigNet50 across several datasets. The method supports efficient AMR deployment on edge devices, advancing real-time spectrum analysis in constrained environments.

Abstract

With the rapid development of wireless communications and the growing complexity of digital modulation schemes, traditional manual modulation recognition methods struggle to extract reliable signal features and meet real-time requirements in modern scenarios. Recently, deep learning based Automatic Modulation Recognition (AMR) approaches have greatly improved classification accuracy. However, their large model sizes and high computational demands hinder deployment on resource-constrained devices. Model pruning provides a general approach to reduce model complexity, but existing weight, channel, and layer pruning techniques each present a trade-off between compression rate, hardware acceleration, and accuracy preservation. To this end, in this paper, we introduce FCOS, a novel Fine-to-COarse two-Stage pruning framework that combines channel-level pruning with layer-level collapse diagnosis to achieve extreme compression, high performance and efficient inference. In the first stage of FCOS, hierarchical clustering and parameter fusion are applied to channel weights to achieve channel-level pruning. Then a Layer Collapse Diagnosis (LaCD) module uses linear probing to identify layer collapse and removes the collapsed layers due to high channel compression ratio. Experiments on multiple AMR benchmarks demonstrate that FCOS outperforms existing channel and layer pruning methods. Specifically, FCOS achieves 95.51% FLOPs reduction and 95.31% parameter reduction while still maintaining performance close to the original ResNet56, with only a 0.46% drop in accuracy on Sig2019-12. Code is available at https://github.com/yaolu-zjut/FCOS.

Paper Structure

This paper contains 13 sections, 12 equations, 3 figures, 8 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Works
  3. Method
  4. Channel Pruning via Similar Channel Fusion
  5. Layer Collapse Diagnosis via Linear Probing
  6. Experiments
  7. Models and Datasets Used
  8. Implementation details
  9. Baselines
  10. Results and Analysis
  11. Additional Experiments and Analyses
  12. Ablation Study
  13. Conclusion

Figures (3)

  • Figure 1: The pipeline of FCOS. In the first stage, hierarchical clustering and parameter fusion are applied to channel weights to achieve channel-level pruning. Then a layer collapse diagnosis (LaCD) module uses linear probing to identify layer collapse and removes the collapsed layers due to high channel compression ratio.
  • Figure 2: Visualization of the changes in layers and channels during pruning of CNN1D on RML2016.10a.
  • Figure 3: Comparison of the accuracy curve of the model obtained by FCOS and a model of the same size trained from scratch. Experiments are conducted on RML2016.10a using CNN1D.