# Blind recognition of channel codes based on dual-branch feature fusion convolutional neural networks

## TL;DR

This paper introduces a new neural network framework to accurately identify different channel coding schemes in dynamic spectrum environments.

## Contribution

The novel DBFCNN framework combines multi-scale and statistical features for improved blind channel coding identification.

## Key findings

- DBFCNN improves identification accuracy by about 5% over prior methods.
- The framework effectively identifies seven common channel-coding schemes.
- The dual-branch design captures both long-range dependencies and algebraic characteristics.

## Abstract

Facing heterogeneous signals increasing in dynamic spectrum, cognitive radio urgently needs blind channel coding identification. This technology addresses the core challenge of unknown coding schemes in non-cooperative communications. Existing methods are typically restricted to specific coding types and suffer from poor identification accuracy and robustness. To mitigate this constraint, we propose a Dual-Branch Feature Fusion Convolutional Neural Network (DBFCNN) framework for fine-grained identification among seven common channel-coding schemes. The network adopts a two-branch architecture. One branch employs multi-scale dilated convolutions to extract long-range dependencies in the received bit sequence, the other is a statistical branch that extract descriptors such as run length, entropy values, coding depth and so on to expose code-specific algebraic characteristics. The fused representation is fed to a fully connected classifier to jointly identify the seven code types. Extensive simulations demonstrate that DBFCNN improves identification accuracy by about 5% (absolute) over a strong prior baseline under comparable settings, proving the feasibility and effectiveness of the method.