CNN-LSTM Hybrid Architecture for Over-the-Air Automatic Modulation Classification Using SDR

TL;DR

This work presents a CNN-LSTM architecture for automatic modulation classification (AMC) that is deployed on a Software Defined Radio (SDR) platform and evaluated with over-the-air (OTA) signals. By converting I/Q data into a 3-channel 224x224 representation (amplitude, phase, I/Q) and employing a modified AlexNet CNN followed by an LSTM over eight windows, the model captures both spatial and temporal signal features. The approach achieves 93.48% accuracy across nine modulation schemes over a challenging SNR range (0–20 dB OTA), with near-perfect ROC for most classes, and demonstrates that preserving the full LSTM sequence outperforms a temporal attention compression. These results highlight the method’s robustness and potential for real-world spectrum management and cognitive radio applications in contested RF environments.

Abstract

Automatic Modulation Classification (AMC) is a core technology for future wireless communication systems, enabling the identification of modulation schemes without prior knowledge. This capability is essential for applications in cognitive radio, spectrum monitoring, and intelligent communication networks. We propose an AMC system based on a hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) architecture, integrated with a Software Defined Radio (SDR) platform. The proposed architecture leverages CNNs for spatial feature extraction and LSTMs for capturing temporal dependencies, enabling efficient handling of complex, time-varying communication signals. The system's practical ability was demonstrated by identifying over-the-air (OTA) signals from a custom-built FM transmitter alongside other modulation schemes. The system was trained on a hybrid dataset combining the RadioML2018 dataset with a custom-generated dataset, featuring samples at Signal-to-Noise Ratios (SNRs) from 0 to 30dB. System performance was evaluated using accuracy, precision, recall, F1 score, and the Area Under the Receiver Operating Characteristic Curve (AUC-ROC). The optimized model achieved 93.48% accuracy, 93.53% precision, 93.48% recall, and an F1 score of 93.45%. The AUC-ROC analysis confirmed the model's discriminative power, even in noisy conditions. This paper's experimental results validate the effectiveness of the hybrid CNN-LSTM architecture for AMC, suggesting its potential application in adaptive spectrum management and advanced cognitive radio systems.

Figures (11)

• Figure 1: Dataset generation and validation pipeline implemented in GNU Radio for creating both clean and AWGN-corrupted signal samples across multiple modulation schemes
• Figure 2: Input preprocessing pipeline showing transformation of I/Q samples into amplitude, phase, and I/Q representations for CNN-LSTM model input
• Figure 3: Preprocessed RGB image of Amplitude, I/Q and Phase
• Figure 4: CNN-LSTM hybrid architecture: AlexNet-based feature extractor, LSTM for temporal modeling, and classification head
• Figure 5: Training and validation curves over 10 epochs for the optimized CNN-LSTM model