Real-time fault monitoring in cable networks with complex topologies is of great significance in improving the stability of power systems. In recent years, Time-Reversal Multiple Signal Classification (TR-MUSIC) has been proposed for soft fault location in cable networks. This method can use single-frequency data to calculate the pseudo-spectrum and the faults can be located by finding the maximum energy positions. However, due to the propagation of harmonic signals, there will be multiple peaks in the pseudo-spectrum in the high-frequency band, which are referred to as ghosting traces. This leads to an inability to accurately determine the location of the soft fault, causing the false peak to be referred to as a ghost fault. This paper analyses the frequency characteristics of the TR-MUSIC method and then introduces the idea of co-prime sampling for frequency sampling to eliminate the influence of ghost fault. The results show that co-prime sampling in TR-MUSIC can counteract the effects of ghosting traces and reduce the computational complexity.