In view of the presence of feature extraction and grouping method used in multi-partial discharge source（multi-PDs）identification and noise suppression system realized by ultra-wideband detection with sampling rate of100 MS/s

this paper points out that one of the key technologies is the fast feature extractors used for the pulse se-quence waveforms. A general way to make feature extraction resorts to nonlinear mapping of the waveforms in timeand frequency（TF）domains

respectively

which are characterized by points in 2 D Cartesian parameters plane or 3 DCartesian parameters space（3 DCPS）

then multi-PDs and noise separation can be realized using the unsupervisedclustering analysis. According to the popular equivalent TF method（ETFM）

a TF autocorrelation method（TFACM）anda TF entropy method（TFEM）are proposed

and a separability criterion for those feature extractors is defined and ana-lyzed with pulse sequences generated with combination of exponential models and Monte Carlo simulation. The sepa-rability evaluation results show those TF operators with good performance

especially in 3 DCPS. The practicability ofthose TF operators is validated with the data from AC and DC PD tests. This study may provide experimental and the oretical basis for selecting feature extraction method in development of multi-PDs detection and identification system.