A Light Weight Neural Network for Automatic Modulation Classification in OFDM Systems

TL;DR

This paper introduces a lightweight neural network approach for automatic modulation classification in OFDM systems, combining subcarrier selection with RNNs to improve accuracy while reducing computational complexity.

Contribution

It presents a novel lightweight neural network method that efficiently classifies subcarriers in OFDM systems, enhancing performance with lower computational demands.

Findings

Improved modulation classification accuracy

Reduced computational complexity

Effective subcarrier prediction

Abstract

Automatic Modulation Classification (AMC) is a vital component in the development of intelligent and adaptive transceivers for future wireless communication systems. Existing statistically-based blind modulation classification methods for Orthogonal Frequency Division Multiplexing (OFDM) often fail to achieve the required accuracy and performance. Consequently, the modulation classification research community has shifted its focus toward deep learning techniques, which demonstrate promising performance, but come with increased computational complexity. In this paper, we propose a lightweight subcarrier-based modulation classification method for OFDM systems. In the proposed approach, a selected set of subcarriers in an OFDM frame is classified first, followed by the prediction of the modulation types for the remaining subcarriers based on the initial results. A Lightweight Neural Network (LWNN) is employed to identify the initially selected set of subcarriers, and its output is fed into a Recurrent Neural Network (RNN) as an embedded vector to predict the modulation schemes of the remaining subcarriers in the OFDM frame.