Because of some disturbances, all experimental data are to some extent contaminated by noise, which will affect and abate the efficiency of chaos time series analysis. So studying noise reduction of time series is very important. The spectral content of nonlinear time series contains strong broadband contributions, therefore, linear filtering in the Fourier domain fails in many cases, e.g. it cannot restrain noises, on the contrary, may lead to signal distortion. This paper introduces a nonlinear state space projection technique developed for noise reduction in chaotic signals, and applies this technique to the analysis of ship radiated signal. The experiment results reveal that the method is feasible and effective.