Fine-grained recognition of error correcting codes based on 1-D convolutional neural network

Abstract

Forward error correction (FEC) codes are the most commonly used method in channel coding, and it is capable of automatically detecting and correcting errors at the receiving end during transmission. At present, the channel coding semi-blind recognition given known FEC codes type has been extensively studied. However, in cognitive radio and non-cooperative systems, the types of channel coding are unknown at the receiving end. Therefore, the ability to identify the type of error correcting codes without any a priori knowledge is essential. In this paper, we proposed a novel method based on convolutional neural network that can achieve fine-grained type recognition for error correcting codes in non-cooperative systems. The proposed algorithm classifies the incoming data symbols among hamming codes, Bose-Chaudhuri-Hocquenghem (BCH) codes, Reed-Solomon (RS) codes, Low-density Parity-check (LDPC) codes, turbo codes, polar codes, and convolutional codes. Further, to train a well-behaved CNN model, we constructed a deep fusion model based on 1-D convolutional layer and modified 1-D inception architecture that can achieve end-to-end extraction of features. Experimental results show that the proposed model earns an average recognition accuracy of roughly 99% under the condition of signal-to-noise ratio (SNR) ranging from 6 dB to 20 dB. In addition, we conduct a comprehensively and thoroughly investigation on the performance of convolutional neural network based code recognition for digital communications.