A signal analysis method of non-stationary wideband noise

The signal in reality is mostly non-stationary signal. In current, this type of signal analysis focus on time-frequency (or time scale) domain, which has also made a series of results. This paper presents a reversible transform method based on a bit of time series from another point. Using this method, we analyze non-stationary broadband noise signal according to the different characteristics between noise and speech signals (such as random degree). Experiment results show that this method can reverse random sequence of disorder time-domain to generally orderly time series. The distinction between broadband noise characteristics and voice characteristics is reversely clear, the method principle is simple, and the algorithm complexity is small which can be real-time processed.