Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BiLSTM and Attention-Based Modulation Classification of Realistic Wireless Signals

Rohit Udaiwal, Nayan Baishya, Yash Gupta, B. R. Manoj

TL;DR

This work tackles automatic modulation classification (AMC) under realistic wireless conditions by introducing QSLA, a quad-stream architecture that extracts spatial features from four representations (IQ, Aφ, I, Q) and models temporal dependencies with a BiLSTM followed by an attention layer. The approach achieves state-of-the-art performance on the realistic RML22 dataset, reaching about 99% accuracy at high signal-to-noise ratios while maintaining a favorable resource footprint compared with several benchmarks. The paper demonstrates that early fusion of spatial features coupled with a single BiLSTM-attention temporal block yields superior accuracy and efficiency over dual-stream or multi-stage temporal fusion methods. These results have practical implications for real-world AMC, enabling accurate, efficient modulation recognition on edge devices and in spectrum monitoring tasks.

Abstract

This work proposes a novel and efficient quadstream BiLSTM-Attention network, abbreviated as QSLA network, for robust automatic modulation classification (AMC) of wireless signals. The proposed model exploits multiple representations of the wireless signal as inputs to the network and the feature extraction process combines convolutional and BiLSTM layers for processing the spatial and temporal features of the signal, respectively. An attention layer is used after the BiLSTM layer to emphasize the important temporal features. The experimental results on the recent and realistic RML22 dataset demonstrate the superior performance of the proposed model with an accuracy up to around 99%. The model is compared with other benchmark models in the literature in terms of classification accuracy, computational complexity, memory usage, and training time to show the effectiveness of our proposed approach.

BiLSTM and Attention-Based Modulation Classification of Realistic Wireless Signals

TL;DR

This work tackles automatic modulation classification (AMC) under realistic wireless conditions by introducing QSLA, a quad-stream architecture that extracts spatial features from four representations (IQ, Aφ, I, Q) and models temporal dependencies with a BiLSTM followed by an attention layer. The approach achieves state-of-the-art performance on the realistic RML22 dataset, reaching about 99% accuracy at high signal-to-noise ratios while maintaining a favorable resource footprint compared with several benchmarks. The paper demonstrates that early fusion of spatial features coupled with a single BiLSTM-attention temporal block yields superior accuracy and efficiency over dual-stream or multi-stage temporal fusion methods. These results have practical implications for real-world AMC, enabling accurate, efficient modulation recognition on edge devices and in spectrum monitoring tasks.

Abstract

This work proposes a novel and efficient quadstream BiLSTM-Attention network, abbreviated as QSLA network, for robust automatic modulation classification (AMC) of wireless signals. The proposed model exploits multiple representations of the wireless signal as inputs to the network and the feature extraction process combines convolutional and BiLSTM layers for processing the spatial and temporal features of the signal, respectively. An attention layer is used after the BiLSTM layer to emphasize the important temporal features. The experimental results on the recent and realistic RML22 dataset demonstrate the superior performance of the proposed model with an accuracy up to around 99%. The model is compared with other benchmark models in the literature in terms of classification accuracy, computational complexity, memory usage, and training time to show the effectiveness of our proposed approach.
Paper Structure (14 sections, 4 equations, 5 figures, 2 tables)

This paper contains 14 sections, 4 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Proposed Architecture
  3. Signal Pre-processing Block
  4. Spatial Feature Extraction
  5. Temporal Feature Extraction
  6. BiLSTM
  7. Attention
  8. Fully Connected Layer and Softmax Classifier
  9. Experiments and results
  10. Experimental Setup
  11. Results and Discussions
  12. Classification Performance
  13. Computational Complexity
  14. Conclusions

Figures (5)

  • Figure 1: Proposed architecture: Quad-Stream BiLSTM-Attention (QSLA) network.
  • Figure 2: Accuracy comparison of QSLA vs benchmark models.
  • Figure 3: Classwise PR curves for QSLA model across all SNRs.
  • Figure 4: Confusion matrices at $6$dB: (a) Proposed QSLA model and (b) DSBA model dual-coms-2023.
  • Figure 5: Performance of the proposed model with different layers for temporal features extraction.