南京理工大学数字信号处理英文课件 Lecture23.ppt

314 * Time-Dependent Fourier Transform Motivation Definition ? Application Lecture 23 Discrete Fourier Transform ( VI ) If the signal is not stationary, frequency contents varies over time ? Need for a time-varying frequency representation ? Fourier transform of a windowed portion of the signals ? Sliding window concept A. Principle Time-Dependent Fourier Transform (TDFT) or Short-Time Fourier Transform (STFT) ? n is a discrete time-index ? λ is a continuous frequency variable ? w[m] is a (given) window ? Fourier transform of the signal x[n + m]w[m] Gives the frequency contents at frequency λ of the signal around time n. B. Definition If the DTFT is implemented via an L-point DFT: requires only that L ≥ N , the length of the window in order to be invertible ? If invertible: C. Sampling in frequency ? If invertible: ? from the STFT at time n, one can recover x[n], . . . , x[n + N － 1] It is thus possible to down sample in time by a factor R ≤N , so that: X[rR, k], r = － ∞, . . . , ∞, and 0 ≤ k ≤ L － 1 completely characterizes x[n], n ? It is also possible to have overlapping sliding windows D. Sampling in Time Let a signal have a varying frequency: x[n]=cos (?0-?[n])n, ?[n] = 4×10－6n2 + 1.4 ×10－4 Spectrogram: picture showing the evolution of |X[n, k]| with time In the choice of the window: trade o. between frequency and time resolution E. Examples E.1 Example : Spectrogram ( I ) Example : Spectrogram ( II ) * * *