Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhancing Automatic Modulation Recognition through Robust Global Feature Extraction

Yunpeng Qu, Zhilin Lu, Rui Zeng, Jintao Wang, Jian Wang

TL;DR

AMR benefits from global feature extraction rather than local patterns alone. The authors introduce TLDNN, a hybrid transformer-LSTM backbone with ReGLU and talking-heads attention, augmented by segment substitution to improve robustness to RF fingerprint and channel effects. Across RadioML2016.10a and RadioML2018.01a, TLDNN achieves state-of-the-art accuracy while reducing FLOPs by 80–90% compared with prior SOTA methods, and shows strong performance in low-SNR and few-shot settings. This approach suggests a scalable, dataset-agnostic backbone for AMR with practical implications for real-world, resource-constrained deployments.

Abstract

Automatic Modulation Recognition (AMR) plays a crucial role in wireless communication systems. Deep learning AMR strategies have achieved tremendous success in recent years. Modulated signals exhibit long temporal dependencies, and extracting global features is crucial in identifying modulation schemes. Traditionally, human experts analyze patterns in constellation diagrams to classify modulation schemes. Classical convolutional-based networks, due to their limited receptive fields, excel at extracting local features but struggle to capture global relationships. To address this limitation, we introduce a novel hybrid deep framework named TLDNN, which incorporates the architectures of the transformer and long short-term memory (LSTM). We utilize the self-attention mechanism of the transformer to model the global correlations in signal sequences while employing LSTM to enhance the capture of temporal dependencies. To mitigate the impact like RF fingerprint features and channel characteristics on model generalization, we propose data augmentation strategies known as segment substitution (SS) to enhance the model's robustness to modulation-related features. Experimental results on widely-used datasets demonstrate that our method achieves state-of-the-art performance and exhibits significant advantages in terms of complexity. Our proposed framework serves as a foundational backbone that can be extended to different datasets. We have verified the effectiveness of our augmentation approach in enhancing the generalization of the models, particularly in few-shot scenarios. Code is available at \url{https://github.com/AMR-Master/TLDNN}.

Enhancing Automatic Modulation Recognition through Robust Global Feature Extraction

TL;DR

AMR benefits from global feature extraction rather than local patterns alone. The authors introduce TLDNN, a hybrid transformer-LSTM backbone with ReGLU and talking-heads attention, augmented by segment substitution to improve robustness to RF fingerprint and channel effects. Across RadioML2016.10a and RadioML2018.01a, TLDNN achieves state-of-the-art accuracy while reducing FLOPs by 80–90% compared with prior SOTA methods, and shows strong performance in low-SNR and few-shot settings. This approach suggests a scalable, dataset-agnostic backbone for AMR with practical implications for real-world, resource-constrained deployments.

Abstract

Automatic Modulation Recognition (AMR) plays a crucial role in wireless communication systems. Deep learning AMR strategies have achieved tremendous success in recent years. Modulated signals exhibit long temporal dependencies, and extracting global features is crucial in identifying modulation schemes. Traditionally, human experts analyze patterns in constellation diagrams to classify modulation schemes. Classical convolutional-based networks, due to their limited receptive fields, excel at extracting local features but struggle to capture global relationships. To address this limitation, we introduce a novel hybrid deep framework named TLDNN, which incorporates the architectures of the transformer and long short-term memory (LSTM). We utilize the self-attention mechanism of the transformer to model the global correlations in signal sequences while employing LSTM to enhance the capture of temporal dependencies. To mitigate the impact like RF fingerprint features and channel characteristics on model generalization, we propose data augmentation strategies known as segment substitution (SS) to enhance the model's robustness to modulation-related features. Experimental results on widely-used datasets demonstrate that our method achieves state-of-the-art performance and exhibits significant advantages in terms of complexity. Our proposed framework serves as a foundational backbone that can be extended to different datasets. We have verified the effectiveness of our augmentation approach in enhancing the generalization of the models, particularly in few-shot scenarios. Code is available at \url{https://github.com/AMR-Master/TLDNN}.
Paper Structure (23 sections, 15 equations, 11 figures, 6 tables)

This paper contains 23 sections, 15 equations, 11 figures, 6 tables.

Table of Contents

  1. Introduction
  2. System Model
  3. Methods
  4. Feature Embedding Layer
  5. Convolutional Layer
  6. SE Block
  7. Transformer Encoder Layer
  8. Positional encoding
  9. Multi-head Attention
  10. Feed-Forward Network
  11. LSTM Layer and Classifier
  12. Segment Substitution
  13. Discrete Segment Substitution
  14. Continuous Segment Substitution
  15. Experimental Results
  16. ...and 8 more sections

Figures (11)

  • Figure 1: Different segments of a signal exhibit different modulation patterns on the constellation diagram.
  • Figure 2: Communication system model.
  • Figure 3: The diagram of our proposed architecture.
  • Figure 4: The mechanisms of transformer and LSTM for extracting global features.
  • Figure 5: The properties of modulation in constellation diagrams and transmission symbols
  • ...and 6 more figures