Digital Signal Processing (EE313): DFT as a filtering tool Krishnan C.M.C Assistant Professor, E&E, NITK Surathkal 2 Significance of Linear Convolution Filter π₯(π) β(π) π¦(π) Linear Filtering π¦ π = π₯ π ∗β π π¦ π = ΰ· π₯ π β(π − π) π DTFT π₯(π) π(π) π»(π) I DTFT × EE313, Dept. of E & E, NITK Surathkal π(π) π¦(π) 3 Linear Convolution with DFT DTFT π(π) π₯(π) π»(π) I DTFT × π¦(π) = ΰ· π₯ π β(π − π) π(π) π DFT π₯(π) π(π) π»(π) I DFT × π(π) EE313, Dept. of E & E, NITK Surathkal = ΰ· π=0→π−1 π₯ π β π−π π 4 Linear Convolution Linear convolution of two sequencesβ π = 1↑ , −1,2 (impulse response), π₯ π = 0↑ , 1,2,3,4,5 (ramp input) demonstrated graphically π₯(π) 0 1 2 π 4 5 π¦ π = ΰ· π₯ π β(π − π) (π = 0) π β(−π) 2 −1 (π = 1) β(1 − π) 2 1 π¦(0) = 0 × 1 = π −1 π¦ 1 = 0 × −1 + 1 × 1 = π 1 (π = 6) 2 β(6 − π) −1 1 π¦ 6 = 4 × 2 + 5 × −1 = π 2 π¦ 7 = 5 × 2 = ππ (π = 7) β(7 − π) π₯(π) π¦(π) 0 1 2 π 4 5 0 1 1 π 5 7 EE313, Dept. of E & E, NITK Surathkal −1 1 3 10 Total length (6+3-1=8 samples) Tail (3-1=2 samples) 5 Linear convolution using circular convolution β π = 1↑ , −1,2 π₯ π = 0↑ , 1,2,3,4,5 π(π) 6 − point × β(π) DFT π»(π) π₯(π) π₯(π) 0 1 2 π 4 5 π¦(π) 0 1 1 π 5 7 6 − point I DFT π(π) 3 10 π¦1 (π) 0 1 1 π 5 7 3 10 6 samples 0 1 1 π 5 7 3 10 3 11 1 π 5 7 EE313, Dept. of E & E, NITK Surathkal 0 1 1 π 5 7 6 Linear convolution using circular convolution β π = 1↑ , −1,2 π₯ π = 0↑ , 1,2,3,4,5 π₯(π) 0 1 2 π 4 5 π¦(π) 0 1 1 π 5 7 π ππ 1 π 5 7 3 10 (8-point) π¦πππππ (π) Points to note: (6-point) Option-1 • For a circular convolution (DFT multiplication) to mimic linear convolution (DTFT multiplication) ο§ Increase the N of the N-point DFT to π = πΏ + π − 1 (here 6 + 3 − 1 = 8) ο§ By giving “space” (i.e., π ≥ πΏ + π − 1) you avoid time domain alias. Option-2 • If you want to keep π = πΏ, the output will have Time domain alias. ο§ First M-1 samples of output π¦ π will be wrong ο§ The contribution of the last M-1 samples of input π₯ π will be forgotten. EE313, Dept. of E & E, NITK Surathkal