Blind parameter estimation of 5G channel coding schemes

Abstract

This project solves the issue of blind parameter estimation and type classification of LDPC and Polar codes, two central schemes of forward error correction in modern wireless communication standards like 5G. The work introduces and builds machine learning-based paradigms to correctly detect code type and code parameters from the received signals, without any initial knowledge of the encoder setup. Large-scale datasets were created for both LDPC and Polar codes with different SNR levels and codeword lengths, and several classification models-CNN, DRN, and hybrid CNN-SVM-were learned and tested. Classification accuracy of the learned models was high, and identification of Polar codes was more than 97%, while they performed well over codeword lengths and SNRs. Experimental verification was performed with a USRP-based Soft- ware Defined Radio platform, and it was shown that the methods presented have solid accuracy and reliability in actual hardware environments, closely replicating MATLAB simulation results. Comparative evaluation against current literature verifies that Polar codes have better robustness and classification accuracy at low SNRs, and LDPC codes have high accuracy at larger block lengths and higher SNRs. The results demonstrate the efficacy of deep learning methods for blind code parameter estimation and justify the real-world application of such methodologies in adaptive, intelligent wireless communication systems.