Abstract: Aim\ To study the methods of wavelet packet decomposition and noise suppression of finite length signals Methods By using the orthogonal guadrature filters, the constructing of the direct sum decomposition of finite dimension space with inner product and the discrete wavelet packets was presented The property of wavelet packet coefficients of noise was studied The method of reducing noise in wavelet packet domain was proposed Results and Conclusion The experimental results of noise reducing for a sine signal and a exponentially damped cosine signal support the efficiency of hard threshold method based on the optimum wavelet packet basis decomposition coefficients