Hardware Implementation of Channel Encoder Classification Using Deep Learning

Channel encoders in digital communication systems are crucial for mitigating random errors introduced by noisy channels. While encoder information is usually accessible in cooperative scenarios, it is often unavailable in non-cooperative military communication systems. In this paper, we present a hybrid convolutional neural network (CNN)-support vector machine (SVM) model for the blind identification of forward error correction (FEC) encoders. Our work focuses on classifying different convolutional encoders with rates $1 / 2,1 / 3$, and $1 / 4$ from the message signals received over additive white Gaussian noise (AWGN) and Rayleigh fading channels. The proposed model surpasses a classification accuracy of 95% at 2 dB signal-to-noise ratio (SNR) for AWGN and at 5 dB SNR for Rayleigh fading channels with input lengths of $1 6, 3 8 4$. Additionally, our model achieves 100% accuracy for all input lengths beyond 10 dB SNR level under both AWGN and Rayleigh fading channel conditions. We validate our findings by comparing the results obtained using datasets generated through simulations with the real-time datasets generated via a GNU Radio-based universal software radio peripheral (USRP) B210 device, revealing that the real-time results are comparable with the simulations results. Moreover, we have compared our model’s performance with the existing models in the literature under Rayleigh fading channel condition and observed that our proposed hybrid CNN-SVM model significantly outperforms them in terms of accuracy, particularly at 5 dB SNR level. These results highlight the effectiveness and robustness of our model, demonstrating its superior performance in practical wireless communication scenarios.