Recently, the associate professor Hao Shan in the Xinjiang Astronomical Observatory (XAO) of Chinese Academy of Sciences (CAS) implemented classification and recognition for 401 observed radio pulsar signals in the frequency range of 1.408-1.642GHz. This experiment was based on the pulsar features derived by the wavelet coefficients, and the clustering methods of fuzzy C-mean, Euclidean and Fisher criteria, and the minimum risk Bayesian decision were used. This is considered as a new proposed method for the important scientific issue—pulsar signal recognition, and it has significance for off-line pulsar signal processing. Internationally, the theories, methods and mathematical tools used are new and practical. The experimental results have been published in the core international journal "Astronomy and Computing."

A signal from a pulsar is a unique signature for a given pulsar, which can be used to differenciate from others. A reasonable and complete description can fully characterize the uniqueness of a pulsar signal. It can also be used to decide the pulsar name it belongs to. Hao Shan has applied the pulsar features based on the wavelet coefficients, and first introduced them into the field of pulsar signal description and recognition. He has initially completed the research task of pulsar signal recognition.

Wavelet features can make the inter-class distance of the classified signal feature samples relatively small, and the intra-class distance relatively large, i.e., the features are distributed on a direct line approximately. Hao Shan implemented the error analysis for 500 recognition experiments, and the results showed that wavelet features can achieve satisfying results in the aspect of pulsar recognition.

Fig.1. Feature classification results on three wavelet scales. Left: scale 1, middle: scale 2, and right: scale 3

Fig.2. Error curves for 500 experiments. The dotted lines are the classification error curves based on the shape parameters, the solid lines are the classification error curves based on the energy feature derived by wavelet coefficients. Red, green and blue represent the 1-D, 2-D and 3-D cases. We use the Euclidean criterion here.