Automatic modulation classification via symbolic representations of complex time series data

Automatic modulation classification (AMC) is of interest for spectrum sensing and network coexistence applications, and it is desirable to find AMC techniques that work with limited computational complexity and memory. This work is an initial exploration of how to apply existing dimensionality reduction symbolic representation approaches for real-valued time series, that claim computational and memory efficiency in data mining and other applications, to feature-based automatic modulation classification of complex-valued time series data. This initial investigation shows that it is feasible to apply these symbolic representation techniques to automatic modulation classification problems with promising performance. Future work needs to be done to refine the approach and to compare the computational complexity and storage efficiency to other techniques.