A Feature Weighted Hybrid ICA-SVM Approach to Automatic Modulation Recognition

Automatic modulation recognition is a topic of interest in many fields including signal surveillance, multi-user detection and radio frequency spectrum monitoring. A major weakness of conventional modulation recognition algorithms is their reliance on high SNR environments and favorable statistics. In this paper an algorithm is developed using elements of cyclo-spectral analysis, ICA and SVM algorithms to distinguish between different modulation types. By first estimating the cyclic spectrum and then analyzing statistical features of the spectrum using machine learning techniques, the particular modulation type can be determined over a wide range of SNR values. This can further be enhanced by employing ICA algorithms to remove feature redundancy. To demonstrate this; simulations are constructed which illustrate the efficiency of the algorithm using digital phase and amplitude modulation. The algorithm's performance is tested over a wide range of SNR values.