Abstract: Multipath fading is the main feature of underwater acoustic channel. Considering the chaotic behavior of multipath fading channel, a new method was proposed to resist the multipath interference. Firstly, an underwater acoustic channel model was established based on chaotic phase space. On the basis of this model, through kernel principal component analysis(KPCA) with nuclear techniques <i<φ</i<(<i<x</i<)·<i<φ</i<(<i<y</i<)=<i<k</i<(<i<x</i<, <i<y</i<), nonlinear kernel principal component could be extracted from the spatial data, which cumulative contribution rate could achieve 90%. And then, the kernel principal component was used to reverse project the spatial data back to the original phase space, and the noise component was removed to improve the noise ratio of acoustic signal. The simulation results and the measured acoustic data show that this method can improve the noise ratio of acoustic signal to 7.76 dB, and can achieve the goal of enhancement of multipath fading underwater acoustic signal.