An efficient method for computing a very high resolution DFT of a short sequence

Many signal processing analysis and design problems require the computation of the DFT of a time sequence composed of some number of non-zero samples followed by some, usually larger, number of zero valued samples. Recently, a novel solution to this problem was presented in [1]. We reexamine this technique and propose two modifications which, for a common class of problems, reduces significantly the number of multiplications.