A Machine Learning Approach to Modulation Detection in Wireless Communications

Autonomous Modulation Detection (AMD) in the presence of unknown channel conditions is critical in spectrum sensing and channel estimation. Existing works achieve AMD by processing raw spectrum measurement data. However, such approaches are computationally complex and time-consuming in practice. We propose a novel AMD as a conglomeration of symbol level extraction and a simplified machine learning classifier. The salient feature of our work is the development of a symbol level extraction scheme that processes physical layer spectrum measurements. The extracted features further simplify the machine learning model design that not only reduces computational complexity but also increases classification performance. The proposed scheme was evaluated on both software as well as hardware testing platforms. The evaluation results were discussed using different metrics and a comprehensive study on the performance of this scheme is also included in this work.