INCREMENTAL REFINEMENT OF DFT AND STFT APPROXIMATIONS

We present a class of multistage algorithms for carrying out increment al refinement of DFT and STFT approximations. Each stage is designed t o improve the previous stage's approximation in terms of frequency cov erage, frequency resolution, and SNR. These algorithms rely almost exc lusively on vector summation operations, and they can be designed to e xhibit a variety of tradeoffs between improvement in approximation qua lity and computational cost per stage. The performance of incremental STFT refinement on real data serves to illustrate the relevance of suc h algorithms to application systems with dynamic real-time constraints .