MATH 267 Due: March 21, 2012, in the class ASSIGNMENT # 8 You have FIVE problems to hand-in. Hand in written solutions for grading at the BEGINNING of the lecture on the due date. Illegible, disorganized or partial solutions will receive no credit. *Staple your HW. You will get F IV E marks OFF if you do not staple your HW! Note that the instructor will NOT provide stapler. Note: Throughout the assignment, the function u(t) denotes the unit step function: 1, t ≥ 0 u(t) = 0, t < 0 1. [NOT TO HAND-IN] [Scaling property of the delta function] (a) For α > 0, show that Z ∞ δ(αt)dt = −∞ 1 α (Hint: This is a one line proof. Use change of variables.) (b) For α < 0, show that Z ∞ 1 δ(αt)dt = − α −∞ (Hint: This is a one line proof. Use change of variables.) (c) From parts (a) and (b), we conclude that, for α 6= 0, Z ∞ 1 . δ(αt)dt = |a| −∞ Solution See Problem 2 in http://www.iam.ubc.ca/˜sospedra/a8MATH267-sol.pdf 2. [Delta function and integral] R∞ δ(t − π)eit/3 dt. √ R∞ Solution Since eit/3 is continuous, −∞ δ(t − π)eit/3 dt = eiπ/3 = 1/2 + i 3/2. R∞ d (b) Find the value of the integral −∞ sin(t + π2 ) dt u(t)dt. R∞ R∞ d d u(t)dt = −∞ sin(t + π2 )δ(t)dt = Solution Notice that dt u(t) = δ(t). Therefore, −∞ sin(t + π2 ) dt sin(0 + π/2) = 1. h i R∞ (c) Find the value of the integral −∞ δ(t − 1) u(1 − 2t ) + u(t − 1) + 2 dt. (Hint: Here applies the class R∞ example about −∞ δ(t)u(t)dt.) Solution We have to be carefulR because the step functions u(1 − 2t ), u(t − 1) are not continuous at ∞ t = 2, t = 1, respectively. Note −∞ δ(t − 1)u(1 − 2t )dt since u(1 − t/2) is continuous at t = 1 (t = 1 is the center of the delta function δ(t − 1). But, since u(t − 1) is discontinuous at the center t = 1 of R∞ the delta function u(t − 1), we have to do as in the class example, to get −∞ δ(t − 1)u(t − 1)dt = 1/2 (Here 1/2 is nothing but the average of the left and right limit of u(t − 1) at t = 1.) (a) Find the value of the integral Z −∞ ∞ h i t δ(t − 1) u(1 − ) + u(t − 1) + 2 dt 2 −∞ Z ∞ Z ∞ Z ∞ t = δ(t − 1)u(1 − )dt + δ(t − 1)u(t − 1)dt + δ(t − 1)2dt 2 −∞ −∞ −∞ 1 = u(1 − 1/2) + + 2. = 1 + 1/2 + 2 = 3 + 1/2 2 1 (d) Find the value of the integral ih i R∞ h u(t + 1) + rect(t) δ(1 − 2t) + 2δ(t + 1) dt. (Hint: The result of −∞ Problem 1 is relevant here.) Solution We can follow as in (c). Since rect(t) is not continuous at t = 1/2, −1/2, we need a little care. First,, Z ∞ rect(t)δ(t + 1)dt = rect−1 = 0. −∞ since rect(t) is continuous at t = −1. R∞ Now, to handle with −∞ rect(t)δ(1 − 2t)dt, note δ(1 − 2t) (or equivalently δ(2t − 1) (since δ(t) is an even function) and it has the center at t = 1/2. We also use change of variables as in Problem 1: Namely, Z ∞ Z ∞ Z 1 ∞ rect(t)δ(2t − 1)dt = rect(t)δ(1 − 2t)dt = rect(s/2)δ(s − 1)ds 2 −∞ −∞ −∞ 1 1 = × 2 2 (The value 1/2 comes from the average of the left and right limit of rects/2 at s = 1. ) Using above and proceeding as in part (c), we see Z ∞h ih i u(t + 1) + rect(t) δ(1 − 2t) + 2δ(t + 1) dt −∞ Z ∞ Z ∞ Z = u(t + 1)δ(1 − 2t)dt + 2u(t + 1)δ(t + 1)dt + −∞ −∞ ∞ −∞ Z ∞ rect(t)δ(1 − 2t)dt + 2rect(t)δ(t + 1)dt −∞ = u(1/2) + 2 × 1/2 + 1/2 × 1/2+) = 1 + 1 + 1/4 = 2 + 1/4 (e) Find the value of the integral Solution Notice that d dt u(t) Z ∞ R∞ h 3δ(t − 2) + −∞ = δ(t), therefore, d dt u(t d dt u(t ih i + 1) u(t + 1) − u( 2t − 1) dt. + 1) = δ(t + 1). We proceed similarly as part (d) to see ih i d t u(t + 1) u(t + 1) − u( − 1) dt dt 2 −∞ Z ∞h i ih t = 3δ(t − 2) + δ(t + 1) u(t + 1) − u( − 1) dt 2 −∞ Z ∞ Z ∞ t = 3δ(t − 2)u(t + 1)dt + −3δ(t − 2)u( − 1)dt 2 −∞ −∞ Z ∞ Z ∞ t + δ(t + 1)u(t + 1)dt + −δ(t + 1)u( − 1)dt 2 −∞ −∞ 1 1 1 1 = 3u(2 + 1) − 3 × × 1 + × 1 − u(−1/2 − 1) = 3 − 3 × × 1 + × 1 − 0 2 2 2 2 =2 h 3δ(t − 2) + 3. [Delta function and Fourier transform] (a) Let f (t) = δ(t − 1) + 3 + δ(1 − 2t) + ei2t . Find fb(ω). Solution F[f (t)](ω) = F[δ(t − 1)](ω) + F[3[(ω) + F[δ(2(1/2 − t))](ω) + F[ei2t ](ω) 1 = e−iω + 3 × 2πδ(ω) + e−iω/2 + 2πδ(ω − 2) 2 1 −iω/2 −iω =e + 6πδ(ω) + e + 2πδ(ω − 2) 2 2 (b) Let gb(ω) = δ(2ω − 1) + 1 + δ(2 − 2ω) + eiω . Find g(t). Solution F −1 [g(t)](ω) = F −1 [δ(2(ω − 1/2))](t) + F −1 [1[(t) + F −1 [δ(2(1 − ω))](t) + F[eiω ](t) 1 1 = eit/2 F −1 [δ((ω))](t/2) + δ(t) + eit F −1 [δ(ω)](t/2) + δ(t + 1) 2 2 1 it/2 1 1 it 1 = e + δ(t) + e + δ(t + 1) 2 2π 2 2π 1 1 = eit/2 + δ(t) + eit + δ(t + 1) 4π 4π (c) Let h(t) = u(t + 1) + u(2t + 1). Find the values b h(π) and b h(π/2). (Hint: Here, the class example d u(t) = δ(t), so about u b(ω) applies.) Solution Note that dt d h(t) = δ(t + 1) + 2δ(2t + 1) dt = δ(t + 1) + δ(2(t + 1/2)). Now, F[ d h(t)](ω) = iωF[h(t)](ω) dt on one hand, and on the other hand, we have F[ 1 d h(t)](ω) = F[δ(t + 1) + δ(2(t + 1/2))](ω) = eiω + eiω/2 dt 2 (here, of course, we used the time-shift and scaling property). So, we get 1 iωF[h(t)](ω) = eiω + eiω/2 2 In particular, 1 iπb h(π) = eiπ + eiπ/2 = −1 + i/2 2 √ √ 1 iπ/4 2 2 iπ/2 b iπ/2h(π/2) = e + e =i+ +i 2 4 4 Thus, −1 + i/2 b h(π) = iπ √ √ i(2 + 2/2) + 2/2 b . h(π/2) = iπ 4. [Discrete Fourier transform] (In the following, we do not distinguish between “length= N discrete-time signals” and “N -periodic discrete-time signals”.) For the given discrete-time signal x, find x b, i.e. its discrete Fourier transform (or in other words, discrete Fourier series). (a) x = [0, 1, 0, 0]. Solution Note e−i2π/4k = e−ikπ/2 = 1, −i, −1, i for k = 0, 1, 2, 3, respectively. Therefore, 1 1 1 i (0 + 1 + 0 + 0) = x b[1] = (0 + (−i) + 0 + 0) = − 4 4 4 4 1 1 1 i x b[1] = (0 + (−i)2 + 0 + 0) = − x b[3] = (0 + (−i)3 + 0 + 0) = . 4 4 4 4 x b[0] = 3 Thus, 1 i 1 i x b = [ ,− ,− , ] 4 4 4 4 Note that this is nothing but x b[k] = eikπ/2 /4 for k = 0, 1, 2, 3. (b) x = [1, 1, 1, 1]. Solution Note e−iπ/4k = e−ikπ/2 = 1, −i, −1, i for k = 0, 1, 2, 3, respectively. Therefore, 1 1 (1 + 1 + 1 + 1) = 1 x b[1] = (1 + (−i) − 1 + i) = 0 4 4 1 1 x b[1] = (1 + (−i)2 + (−1)2 + (i)2 ) = 0 x b[3] = (1 + (−i)3 + (−1)3 + i3 ) = 0. 4 4 x b[0] = (You can also use ‘orthogonality’ to see the above immediately.) Thus, x b = [1, 0, 0, 0] N entries z }| { Remark Let us consider more general case: x = [1, 1, · · · , 1]. By definition, x b[k] = Clearly x b[0] = x b[k] = 1 N [N ] i (N −1)k k 2k 1h 1 + e−2πi N + e−2πi N + · · · + e−2πi N N = 1. For 1 ≤ k ≤ N − 1, let w = wN = e 2πi N to get i 1h 1 1 − w−N k 1 + w−k + w−2k + · · · + w−(N −1)k = =0 N N 1 − w−k since w−kN = e−2πki = 1 Hence x b = [1, 0, · · · , 0] (N entries). (c) x = [1, −1, 1, −1]. Solution Note e−i2π/4k = e−ikπ/2 = 1, −i, −1, i for k = 0, 1, 2, 3, respectively. Therefore, 1 1 (1 − 1 + 1 − 1) = 0 x b[1] = (1 − (−i) − 1 − i) = 0 4 4 1 1 2 2 2 x b[1] = (1 − (−i) + (−1) − i ) = 1 x b[3] = (1 − (−i)3 + (−1)3 − i3 ) = 0. 4 4 x b[0] = (You can also use ‘orthogonality’ to see the above immediately, since x[n] = (−1)n for n = 0, 1, 2, 3.) Thus, x b = [0, 0, 1, 0] (d) x = [1, 2, 3, 4]. Solution Note ei2π/4k = eikπ/2 = 1, i, −1, −i for k = 0, 1, 2, 3, respectively. Therefore, 1 1 −1 + i (1 − 2 + 3 − 4) = −1/2 x b[1] = (1 + 2(−i) − 3 + 4i) = 4 4 2 1 1 1 −1 − i x b[1] = (1 + 2(−i)2 + 3(−1)2 + 4i2 ) = − x b[3] = (1 + 2(−i)3 + 3(−1)3 + 4i3 ) = . 4 2 4 2 x b[0] = Thus, 1 −1 + i 1 −1 − i x b = [− , ,− , ] 2 2 2 2 4 (e) x = [1, 31 , 312 , · · · , 3110 , 3111 ]. Solution Let us consider more general case: x = [1, r, r2 , · · · , rN −1 ]. (In the problem, N = 12, r = 1/3. ) Let w = wN = e 2πi N Then, x b[0] = i 1 1 − rN 1h 1 + r + r2 + · · · + r(N −1) = N N 1−r (note that we could use the geometric sum for the case r 6= 1) and x b[k] = 2 (N −1) i 1h 1 − (rw−k )N 1 − rN 1 + rw−k + rw−k + · · · + rw−k = = N N (1 − rw−k ) N (1 − rw−k ) for 1 ≤ k ≤ N − 1. (Here, we used that wN = 1. ) Back to our specific case, we see that x b[k] = 1 − ( 31 )12 12(1 − 13 e−iπk/6 ) for k = 0, 1, 2, · · · , 11. 5. [Inverse discrete Fourier transform] (a) x b = [0, 0, 3, 0]. Solution Note eiπ/4k = eikπ/2 = 1, i, −1, −i for k = 0, 1, 2, 3, respectively. Therefore, x[1] = 0 + 0 − 3 − 0 = −3 x[0] = 0 + 0 + 3 + 0 = 3 2 x[1] = 0 + 0 + 3(−1) + 0 = 3 x[3] = 0 + 0 + 3(−1)3 + 0 = −3. Thus, x = [3, −3, 3, −3] (b) x b = [1, 1, 1, 1]. Solution Note eiπ/4k = eikπ/2 = 1, i, −1, −i for k = 0, 1, 2, 3, respectively. Therefore, x[0] = 1 + 1 + 1 + 1 = 4 2 2 x[1] = 1 + i − 1 − i = 0 2 x[1] = 1 + i + (−1) + (−i) = 0 x[3] = 1 + i3 + (−1)3 + (−i)3 ) = 0. (You can also use ‘orthogonality’ to see the above immediately.) Thus, x = [4, 0, 0, 0] N entries z }| { Remark Let us consider more general case: x b = [1, 1, · · · , 1]. By definition, n 2n x[n] = 1 + e2πi N + e2πi N + · · · + e2πi Clearly x[0] = N . For 1 ≤ n ≤ N − 1, let w = wN = e x[n] = 1 + wn + w2n + · · · + w(N −1)n = Hence x = [N, 0, · · · , 0], (N entries). 5 2πi N (N −1)n N to get 1 − wN n =0 1 − wn since wnN = e2πni = 1 (c) x b = [1, 41 , 412 , · · · , 4110 , 4111 ] Solution Let us consider more general case: x b = [1, r, r2 , · · · , rN −1 ]. (In the problem, N = 12, r = 1/4. ) 2πi Let w = wN = e N . Then, 1 − rN 1−r (note that we could use the geometric sum for the case r 6= 1) and 2 (N −1) i 1 − (rwn )N 1 − rN x[n] = 1 + rwn + rwn + · · · + rwn = = (1 − rwn ) 1 − rwn x[0] = 1 + r + r2 + · · · + r(N −1) = for 1 ≤ n ≤ N − 1. (Here, we used that wN = 1. ) Back to our specific case, we see that x[n] = 1 − ( 41 )12 1 − 14 eiπn/6 for k = 0, 1, 2, · · · , 11. 6. [Discrete complex exponentials] (a) Compute 10 h 2π i X 2π 2π 2π 2π e−i 11 2n + ei 11 3n ei 11 2n + e−i 11 3n + ei 11 8n n=0 P1 Solution Note that in the following sum n=0 0 three are N = 11 terms. This matches well with 2π e0i 11 kn to apply orthogonality of discrete complex exponentials. 10 h X 2π 2π e−i 11 2n + ei 11 3n 2π 2π 2π ei 11 2n + e−i 11 3n + ei 11 8n i n=0 = 10 X 2π 2π e−i 11 2n ei 11 2n + 2π 2π e−i 11 2n e−i 11 3n + n=0 n=0 + 10 X 10 X 2π 2π ei 11 3n ei 11 2n n=0 10 X 2π 2π e−i 11 2n ei 11 8n n=0 10 X 2π 2π ei 11 3n e−i 11 3n + n=0 10 X 2π 2π ei 11 3n ei 11 8n n=0 = 11 + 0 + 0 + 0 + 11 + 11 In the last line we have used the orthogonality of discrete complex exponentials. (For the last term, notice that 3 + 8 = 11 an integer multiple of 11.) So, the answer is 33 (b) Compute 10 h 2π i X 2π 2π 2π 2π e−i 10 2n + ei 10 3n ei 10 2n + e−i 10 3n + ei 10 7n n=0 P10 2π Solution Notice that the sum n=0 has 11 terms, but, we have complex exponentials e−i 10 kn with N = 10. Because of this we decompose the sum as 10 h X 2π 2π e−i 10 2n + ei 10 3n 2π 2π 2π ei 10 2n + e−i 10 3n + ei 10 7n i n=0 = 9 h X 2π 2π e−i 10 2n + ei 10 3n 2π 2π 2π ei 10 2n + e−i 10 3n + ei 10 7n i n=0 2π 2π 2π 2π 2π + e−i 10 2×10 + ei 10 3×10 ei 10 2×10 + e−i 10 3×10 + ei 10 7×10 6 For the summation P9 n=0 part, we use the orthogonality of discrete complex exponentials, and see 9 h 2π i X 2π 2π 2π 2π e−i 10 2n + ei 10 3n ei 10 2n + e−i 10 3n + ei 10 7n n=0 = 9 X 2π 2π e−i 10 2n ei 10 2n + n=0 + 9 X 2π 2π e−i 10 2n e−i 10 3n + n=0 9 X 2π 2π ei 10 3n ei 10 2n + n=0 9 X 2π 2π 2π 2π e−i 10 2n ei 10 7n n=0 9 X 2π 2π ei 10 3n e−i 10 3n + n=0 9 X ei 10 3n ei 10 7n n=0 = 10 + 0 + 0 + 0 + 10 + 10 = 30 For the remaining part, 2π 2π 2π 2π 2π e−i 10 2×10 + ei 10 3×10 ei 10 2×10 + e−i 10 3×10 + ei 10 7×10 (since ei2πk = 1 for integer k). = (1 + 1)(1 + 1 + 1) = 6. Therefore the final answer is 36. 7. [NOT TO HAND-IN] Two N -periodic (or length= N ) signals x, y and their discrete Fourier transforms x b, yb are given. Prove: np f [n] = e2πi N x[n] f [n] = 1 N N −1 X x[m] y[n − m] =⇒ fb[k] = x b[k − p] (frequency shifting) =⇒ fb[k] = x b[k] yb[k] (convolution) =⇒ fb[k] = m=0 f [n] = x[n] y[n] N −1 X x b[m] yb[k − m] (multiplication) m=0 (Notice some typos are corrected in red color above!!!) Solution Use the definition of the discrete fourier transforms for the three properties. kp • • Frequency Shifting: Given that f [n] = e2πi N x[n], substitution yields N −1 N −1 N −1 (k−p)n kn kn 1 X 1 X 2πi kp 1 X fb[k] = f [n]e−2πi N = e N x[n]e−2πi N = x[n]e−2πi N = x b[k − p] N n=0 N n=0 N n=0 • • Convolution: Substituting f [n] = N −1 1 X x[m] y[n − m] N m=0 into N −1 kn 1 X fb[k] = f [n]e−2πi N N n=0 gives N −1 N −1 N −1 N −1 kn kn 1 XX 1 X X fb[k] = 2 x[m] y[n − m] e−2πi N = 2 x[m] y[n − m]e−2πi N N n=0 m=0 N m=0 n=0 Make a change of variables replacing the summation variable n with r = n − m: N −1 N −1−m N −1 N −1−m k(r+m) kr 1 X X 1 X 1 X −2πi N −2πi km b N f [k] = 2 x[m] y[r]e = x[m]e y[r]e−2πi N N m=0 r=−m N m=0 N r=−m The summand of the r sum is periodic of period N . That is, replacing r by r + N has no effect on the summand. So all domains of summation consisting PN −1−m PN −1 of a single full period give the same sum and we may replace the sum r=−m by the sum r=0 , which gives fb[k] = x b[k] yb[k]. 7 • • Multiplication: Substituting x[n] = N −1 X x b[m]e2πi mn N into m=0 N −1 N −1 kn kn 1 X 1 X fb[k] = f [n]e−2πi N = x[n]y[n]e−2πi N N n=0 N n=0 gives NX N −1 −1 kn 1 X 2πi mn b N e−2πi N f [k] = y[n] x b[m]e N n=0 m=0 = N −1 N −1 (k−m)n 1 X X y[n] x b[m]e−2πi N N n=0 m=0 = N −1 N −1 (k−m)n 1 X X y[n] x b[m]e−2πi N N m=0 n=0 = N −1 X m=0 = N −1 X x b[m] N −1 (k−m)n 1 X y[n]e−2πi N N n=0 x b[m] yb[k − m] m=0 *Staple your HW. You will get F IV E marks OFF if you do not staple your HW! Note that the instructor will NOT provide stapler. 8