Data-Driven Adaptive Modulation Classification Systems

The integration of machine learning into wireless communication systems has introduced innovative solutions for challenges such as modulation classification and interference mitigation. However, implementing deep learning models on edge devices with constrained computational capabilities poses significant obstacles. This study focuses on classifying Doppler velocities in Space-Time Orthogonal Frequency Division Multiplexing (ST-OFDM) systems using an adaptive approach enabled by 2D Convolutional Neural Networks (CNNs). By leveraging a realistic MATLAB-generated dataset, we simulate urban channel conditions with Doppler frequency ranging from 0 Hz to 500 Hz caused by relative velocity between mobile unit and base station. This Doppler frequency is considered as normalized frequency offsets at a 5 GHz carrier frequency. The dataset consists of QPSK-modulated signals transmitted across 512 subcarriers per OFDM block. Based on detected Doppler conditions, we introduce an adaptive mechanism to mitigate Doppler-induced intercarrier interference (ICI) by dynamically switching between two architectures: regular one-path Space-Time (ST)-OFDM with a higher data rate and two-path ST parallel cancellation (STPC)-OFDM with a lower bit error rate (BER) in mobile fading channels. The proposed model demonstrates high classification accuracy and robust performance, paving the way for real-world applications in dynamic and complex fading environments. By emphasizing adaptability and leveraging the efficiency of 2D CNNs, this work presents a novel framework for addressing Doppler-related challenges in ST-OFDM systems.