With fixed windows or kernels

Time-frequency representation（TFR）performs well only for limited signals

however representation with signal depended kernels can overcome this limitation. The TFR developed here is based on signal depended tilted Gaussian kernel. In order to find the optimal tilted Gaussian kernel we introduce expectation maximization（EM）algorithm. We can easily get the optimal Gaussian kernel from the estimated parameters of the auto-component by simply transplanting.