Partial discharge (PD) is a common reason that causes electrical breakdown in high voltage underground XLPE cables. This paper proposes a concept of how to build an on-line, on-site system that is able to diagnose the severity of PD activities in XLPE cable as well as differentiate different types of PD signals. The system consists of magnetic probes, low noise amplifier, 3GSPS analog to digital converter (ADC) and a field programmable gate array (FPGA) board. The energy of PD signals is used to assess the severity of PD activities and artificial neural network (ANN) is used to classify different types of PD waveforms. In addition, wavelet transform is used to clean the time-resolved input signals and statistical method is used to extract important features of PD signals to fetch into neural network. The training of ANN is done on personal computer. The prototype and results of the research is elaborated in this paper.