Hardware Implementation of LDPC Encoder Classification for Wi-Fi and 5G via Deep Learning

Abstract

Channel encoders are essential for mitigating errors in digital communication

however, their blind identification remains a significant challenge in non-cooperative systems. Further, the blind recognition of channel encoders plays a significant role in the design of blind receivers for future-generation communication systems. This paper proposes a capsule network (CapsNet)-based model for the blind recognition of low-density parity-check (LDPC) codes, designed to distinguish between wireless fidelity (Wi-Fi) and fifth generation (5G) new radio (NR) signals. The model successfully classifies Wi-Fi-based and 5G-NR-based LDPC encoders, under both additive white Gaussian noise (AWGN) and Rayleigh fading channels. The proposed approach achieves over 95% classification accuracy at a signal-to-noise ratio (SNR) value of 5 dB under AWGN channel condition and exceeds 85% accuracy beyond 5 dB SNR under Rayleigh fading conditions. Furthermore 100% accuracy is achieved in both the scenarios at the SNR levels of 10 dB (AWGN) and 20 dB (Rayleigh). Real-time validation using GNU Radio platform and universal software radio peripheral (USRP) B210 devices confirms the alignment of experimental results with simulations, demonstrating the practical feasibility of the proposed model. Comparative analysis shows that the proposed model outperforms the existing methods, particularly under Rayleigh fading, making it a promising solution for next-generation wireless communication systems. © 2025 IEEE.