DESIGN OF FIR DIGITAL FILTERS Solution 17.1 Rectangular window: N-1 WR (ej) e-jwn n=-N+l 2N-2 = e-j(n-N+l) n=O 2N-2 e -jw(-N+l) n=0 jw(-N+l) e- 1-ejw(2N-l) 1-e~jW sin ((2N-1),,) 2 WRW WR ( e W) sin 4 The width of the main lobe is which, for N>> 1 is approximately 2N-1 27. N Bartlett window: 0 < n Define w R(n) = 1 (n - B N + 1) = -j )=e (N-i) sin si sin Therefore (e) WB W (e W wN _K (with N = M + 1). -N ­ 2 2 2 sin 1 N) 1 1~ w (n) * w R (n) R N From Eq. 7.76 of the text WR(e - otherwise = 0 Then w < 'N __ (Sin -T In this case the width of the main lobe is N which is twice that for the rectangular window. S17.1 Raised cosine window: N-l e-jon a + Cos (Tr)n ) WH(e -(N-l) N-1 N-1 jwn + e c~2 e-jn(w -(N-l) sin N-1 e -jn ) + (w- T N-1 -(N-l) 2N-l (o (2N-1 sin sin[ sin 2 2N-l (-iTV (w---) 1 Assuming that N >> 1, this can be rewritten as sin (Nw +T) sinwN sin (Nw-TV) + )=a_ sin WH(e WH~ec 2 + 2 2N Lsin-(Nw+7) sin (Nw-Tr) This is the superposition of three terms as sketched in Figure S17.1-l. sin W N sin -­ 2 (NO+-T) sin s sin (NW- 2N1 sin TV -N 7T) (Nw-T) 2TV N Figure S17.1-1 which From this figure we observe that the first values of w for Consequently for this the superposition will be zero are w = ± -. window also, the width of the main lobe isN. S17.2 ) Solution 17.2 (a) Since h1 (n) and h2 (n) are related by a circular shift, their DFTs are related by H2 (k)2 = (k) W 4kH 8 H1 ( (-l)kH1 (k) k Thus, their magnitudes are equal. Since the DFT corresponds to samples of the Fourier transform, the values of H1 (k) are the samples of H1 (ejW) indicated in (b) Figure S17.2-1 H (e )W DFT values -7T 7T 7 Figure S17.2-1 From (a), H2 (k) = (-1)kH (k). Thus the values H2 (k) are as indicated Since these alternate in polarity, the continuous frequency response of which these are samples, must go through zero in between these samples as indicated. in Figure S17.2-2. S17.3 H2 (ejW , ' I \ I -T TI Figure S17.2-2 Thus h2 (n) obviously does not correspond to a good low pass filter, even though the magnitude of its DFT values are identical to those of the low pass filter h (n). Solution 17.3 (a) Since E(eJW) H (eJW) d transform of E(e W), e(n) = hd(n) - = H(e 3), and e(n) is the inverse Fourier - h(n). Tr (b) e2 JE(ejw)1 2 dw 1 -1? 7T +00 1 f[ .i n=-o +00 E e(n)e(k)ej wnejwk dw k=-0 Interchanging the order of integration and summation e 2= 1 S17.4 +00 +00 E X 1 7 e (n) e (k) J ejw (k-n) dw =0 k =27 k = n n Thus, c 2 (c) 2 e n=-co = E:= 2 (n) From the results of parts (a) and (b), E+o [hd(n)-h(n)] 2 n=-o or, since h(n) = 0, n < 0 and n > N, 2 2 N-1 E2 = EQ [hd(n)-h(n)]2 n=0 + -l *O 2 E h d(n) + n=N n=-o h hd(n). Clearly, the choice of h(n) cannot affect the last summations. The first is non-negative and consequently its minimum value is zero which is achieved for h(n) = hd(n). It should be stressed that although a rectangular window minimizes the mean square error, it does not generally result in the best frequency characteristic in terms, for example of passband or stopband ripple. SL7. 5 MIT OpenCourseWare http://ocw.mit.edu Resource: Digital Signal Processing Prof. Alan V. Oppenheim The following may not correspond to a particular course on MIT OpenCourseWare, but has been provided by the author as an individual learning resource. For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.