A practical modulation recognition algorithm for HF signals and beyond

Automatic modulation recognition has been a popular subject of research over the last 20 years, particularly because of the signal processing capabilities of software defined radio systems. Literature discussing aspects of modulation recognition usually concentrates on signal features, probability density functions, estimation of signal-to-noise ratio (SNR), or classification methods. Few papers actually combine all the components into a coherent algorithm. This paper combines estimation of SNR, feature extraction, classification, and a classification confidence measure into a general modulation recognition algorithm that can be used to identify a constrained set of modulations. The algorithm can be used for any modulation type in any frequency band. Nevertheless, the exemplar presented here consists of a set of five modulations used in the HF band and the resulting performance of the algorithm against the modulations.