Non-stationary signal spectrum analysis improving maximum entropy estimation error

A novel approach is presented for short-time spectrum estimation of nonstationary signals. A two-dimensional spectrum is defined for a discrete-time sequence. One dimension is the conventional axis, and the other is a new frequency axis that corresponds to the time variation of frequency of the signal spectrum. Nonstationary behavior is thought to occur along the new axis, and a set of lowpass filters is used to eliminate the undesired variations. The approach has been applied to the conventional maximum entropy method (MEM). A novel approach to nonstationary signal detection is also presented, using information entropy. These approaches are found to be effective for nonstationary signal analysis