ABSTRACT:Automatic modulation classification (AMC) has evolved significantly from traditional statistical methods to sophisticated machine learning approaches over the past two decades. This in-depth analysis explores the transition from likelihood-based and featurebased traditional techniques to recent deep learning architectures for radio signal modulation recognition. Traditional approaches, including Average Likelihood Ratio Test (ALRT), Generalized Likelihood Ratio Test (GLRT), and feature-based methods using statistical moments and spectral analysis, provided foundational understanding but suffered from computational complexity and manual feature engineering limitations. Machine learning, specifically deep learning, had transformed the area, with Convolution Neural Networks (CNNs) achieving 83.8% accuracy at high SNR, while advanced architectures such as ResNet (83.5%), DenseNet (86.6%), and Convolution Long Short-Term Deep Neural Networks (CLDNN) achieve state-of-the-art performance of 88.5% accuracy. Recent developments in transformerbased models and hybrid architectures show promising results for temporal modeling of radio signals. However, challenges remain in real- world deployment, including dataset domain gaps, computational constraints, and robustness under adverse conditions. This review synthesizes current methodologies, benchmarks, and identifies future research directions including federated learning, edge deployment optimization, and interpretability enhancement for critical communication systems.

KEYWORDS:machine learning, deep learning, radio signal processing, wireless communication.

A COMPREHENSIVE REVIEW ON OPTIMIZED COMPARATIVE ANALYSIS OF RADIO MODULATION TECHNIQUES USING MACHINE LEARNING APPROACHES FOR 5G SERVICES

MR. HARIS MAHMOOD QURAISHI
PROF. NILESH NAGRALE
DEPARTMENT OF INFORMATION TECHNOLOGY, TULSIRAM GAIKWAD COLLEGE OF ENGINEERING AND TECHNOLOGY MOHAGAON, WARDHA ROAD, NAGPUR-441108, INDIA