Automatic Modulation Recognition Based on a DCN-BiLSTM Network.

Area of Science:

• Electrical Engineering
• Computer Science
• Signal Processing

Background:

• Automatic modulation recognition (AMR) is crucial for noncooperative wireless communications.
• Traditional Convolutional Neural Networks (CNNs) struggle with phase information loss, impacting recognition accuracy.
• Deep Complex Networks (DCNs) offer a potential solution by preserving phase and amplitude data.

Purpose of the Study:

• To propose a novel deep complex network cascading bidirectional long short-term memory (DCN-BiLSTM) for enhanced AMR.
• To address the limitations of traditional CNNs in capturing signal phase information.
• To improve the accuracy and robustness of modulation recognition in wireless systems.

Main Methods:

• Feature extraction using a Deep Complex Network (DCN) to capture phase and amplitude information.
• Cascading Bidirectional Long Short-Term Memory (BiLSTM) layers to process sequential signal data and capture contextual dependencies.
• Classification using a fully connected layer followed by a softmax classifier.

Main Results:

• The proposed DCN-BiLSTM algorithm demonstrates superior performance compared to existing neural network recognition algorithms.
• Achieved a recognition rate of over 90% for 11 modulation signals when the signal-to-noise ratio (SNR) exceeds 4 dB.
• The model effectively extracts signal features and contextual information, leading to high recognition accuracy.

Conclusions:

• The DCN-BiLSTM model is an effective approach for automatic modulation recognition in noncooperative wireless systems.
• This method overcomes the limitations of traditional CNNs by preserving crucial phase information.
• The proposed model offers a significant advancement in modulation recognition technology, achieving high accuracy at moderate SNRs.