Fractional Fourier transform (FrFT) is widely utilized in nonstationary signal processing owing to its linearity and flexibility. Various discrete algorithms focus on the global discrete FrFT (DFrFT) points. However, only a part of the DFrFT points are considered because of a priori knowledge of applications and limited memory space of digital machines. Therefore, a local DFrFT (LDFrFT) was proposed in this study. LDFrFT for the first Q continuous DFrFT points is proposed, and its complexity was reduced from O(Mlog2(M)) to O(Mlog2(Q)) for an M-point sequence with Q being a factor of M. It is further extended to the LDFrFT for any Q continuous DFrFT points having complexity less than that of the DFrFT. These two algorithms were then generalized to local discrete linear canonical transform algorithms. The LDFrFT uses all sampling points, thus maintaining the resolution of the DFrFT. Moreover, LDFrFT is not limited by the starting point and is more flexible than the recently proposed local DFrFT algorithm. Finally, simulated frequency-modulated continuous-wave radar signals, gravitational waves, and real-world bat echolocation chirp data were utilized to demonstrate the effectiveness of the LDFrFT, which works well when the SNR is over -30 dB.
Read full abstract