Blind Interleaver Recognition Using Deep Learning Techniques

In digital communication systems, channel encoders and interleavers play a crucial role in mitigating the random and burst errors introduced by noisy channels. While information about the encoder and interleaver is typically available in cooperative scenarios, non-cooperative military communication systems often lack such knowledge. This paper explores the application of deep learning to recognize four different interleavers such as block, convolutional, helical, and random especially in non-cooperative environments. By utilizing convolutional neural network and dilation residual network models, we perform the recognition task with input data for interleavers encoded using six different channel encoders such as block, convolutional, Bose-Chaudhuri-Hocquenghem, Reed-Solomon, low density parity check, and polar. Across all cases, both convolutional neural network and dilation residual network models consistently achieve classification accuracy exceeding 95% at varying signal-to-noise ratio values. In addition, we compare the classification accuracy and testing/training time of both our proposed models (i.e. convolutional neural network and dilation residual network) with the recurrent neural network, feedforward network, and autoencoder models available in the literature. Our findings reveal that the accuracy improves with high input sample length and yielding superior results for both the models. Notably, among the five models, dilation residual network outperforms other models in terms of classification accuracy and convolutional neural network outperforms dilation residual network in terms of training/testing time. Our results show that the convolutional neural network and dilation residual network models outperform the other methods. We also compare the classification accuracy of the proposed model with the existing rank-based approach proposed in the literature. Our findings indicate that at the same bit error rate values, our model demonstrates superior classification accuracy.