Research On Signal Detection Algorithm On The Basis Of Bispectrum And The Complex Number Spectrum

This thesis researches the signal stream detection in strong background noise environment. The traditional detection algorithm, based on zero-crossing or energy, will not acquire ideal effect when the Signal-to-Noise is low or the signal is weaker. Therefore, to resolve the real problem in the real environment that all kinds of random noise and speech signal exit together, some new algorithm must be put forward.Account for the complexity of real noise, we integrate the wavelet transform and high-order statistics and advance a new algorithm; the algorithm can effectively separate the speech signal and the non-Gauss noise. The statistic of wavelet transform coefficient algorithm can solve the periodic noise, high-energy noise and some non-Gauss noise simply and effectively; bi-spectrum can acquire more information from the original signal than power-spectrum, detect more information except from range and restrain the Gauss noise. Short-time speech signal can be considered as stationary and with periodic non-Gauss signal, so we can make use of bi-spectrum to obtain the speech character and separate the speech and noise and detect Morse telegraph signal ; complex number spectrum variance algorithm is put forward based on the deeply observing speech data, it is a new algorithm, experiment show that it is simple, effective.