An Ensemble Deep Learning Model for Automatic Modulation Classification in 5G and Beyond IoT Networks

Area of Science:

• Artificial Intelligence
• Wireless Communications
• Machine Learning

Background:

• Deep learning (DL) techniques are increasingly vital for automatic modulation classification (AMC) in artificial intelligence (AI) and machine learning (ML).
• AMC is crucial for Internet of Things (IoT)-assisted wireless systems, demanding efficient and accurate classification methods.
• Existing AMC models often face challenges with lightweight architectures and performance across diverse signal-to-noise ratios (SNRs).

Purpose of the Study:

• To introduce a novel, lightweight ensemble deep learning model for AMC.
• To evaluate the proposed model's effectiveness on a standard dataset covering multiple modulation types.
• To demonstrate the model's superiority compared to existing approaches, particularly in varying SNR conditions.

Main Methods:

• Development of a deep recurrent convoluted network with an additional gated layer (DRCaG), integrating convolution, Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU) layers.
• Testing the DRCaG model on the RadioML2016(b) dataset, which includes eight distinct modulation types (BPSK, QPSK, 8-PSK, 16-QAM, 4-PAM, CPFSK, GFSK, WBFM).
• Performance evaluation through extensive simulations, analyzing training loss, accuracy, and confusion matrices across a wide range of SNRs (-20 dB to +20 dB).

Main Results:

• The DRCaG model achieved high accuracy in classifying various modulation types.
• The model demonstrated robust performance across a broad spectrum of signal-to-noise ratios, from -20 dB to +20 dB.
• Simulations confirmed the superiority of the DRCaG model compared to existing AMC techniques.

Conclusions:

• The proposed DRCaG model offers an effective and lightweight solution for AMC in IoT wireless systems.
• The ensemble approach combining convolutional and recurrent layers provides enhanced classification accuracy and robustness.
• DRCaG represents a significant advancement in deep learning-based modulation classification for modern wireless communication.