short time fourier transform overlap
Short-time fourier transsform is a analysis tool. The result may not be used to resconstruct the signal. Your time-frequency frame (sometime it is called Garber Frame) have to satisfy unertianty princitple. While DCT, wavelet, KLT is a signal approximation. You can also use it to analize signal also but your transformed signal has not much meaning in physic. IN the case of signal compression, who care? we just need the one that give the lowest entropy. But in signal analysis, it is crucial. how to reconstruct the signal is not importance. Fourier transform is not so good in signal compression since it may map from real/integer to complex numer so u have twice as much as data. To get a real specrum, you need a self-adjoint operator which DCT is one of them. That why they use DCT for image but DCT has it own problem. The image is an integer. if we map from interger to real, your entropy increase. Hence we need a quantizer. Wavelet is much better, besides it is a self adjoint operator, it have a property such as vanishing moment, so you can deaign you wavelet so that you series is not too long since after a certain value, all coefficients will be zero. There is also an implaementation of the wavelet transform that can map from interger to integer (lifting scheme) so you dont loss any information in quantizattion process. KLT is optimal since it use the signal itself to construct the basis; hence you get the real signal space from the undelying signal. It is not so efficient since you have to compute the basis everytimg u need to do a transformation.