3.07 IMPERIAL COLLEGE LONDON at 18E3.11 DEPARTMENT OF ELECTRICAL AND ELECTRONIC ENGINEERING EXAMINATIONS 2005, EEE/ISE PART llVV: MEng, BEng and ACGI DIGITAL SIGNAL PROCESSING Thursday, 28 April 10:00 am Time allowed: 3:00 hours There are SIX questions on this paper. Answer FOUR questions. Corrected Copy All questions carry equal marks ‘Any special instructions for invigilators and information for candidates are on page 1. Examiners responsible First Marker(s) P.A. Naylor Second Marker(s): D.P. Mandic. © Unversity of London 2005E3.07/AUVISES.11 Special Instructions for Invigilators: None Information for Candidates: ‘Sequence z-transform (n) 1 i a"u(n) 1-az™ Table 1 : ztransform pairs ‘5(n) Is defined to be the unit impulse function. 1u(n) is defined to be the unit step function. Numbers in square brackets against the right margin of the following pages are a guide to the marking scheme. Digital Signal Processing Page 1 of 5(a) (b) (o) x(n) E3.07/AUVISE3.11 Write down the Type 1 polyphase form of afiter H(z) Consider the system shown in Figure 4 in which P and Q represent linear operators with L inputs and L outputs. () What function would this system typically perform? (li) For this typical function, describe in words the relationship of v(m) 1=0,1..,L-1 to x(n). (ii) Determine expressions for the transfer functions from the input signal x(n) to the signals v,(n) i=0,1,...,L—I. Consider a lowpass fitter H/,(z) with impulse response ,(n) and normalized cut- off frequency 7/2 . Derive a highpass fiter H,(z) and give its impulse response ‘A,(m) such that H,(z) and H,(z) are Quadrature Mirror Filters. Sketch an example of the magnitude frequency responses of the filters for the frequency range -2n 10 2 te ad tt} vn) Q z tt}-3 fi] vis(n) Liar 7 we) Figure 1. Digital Signal Processing Page 2 of 5 (4) [2] 3] {51 [6]E3.07/AUISE3.11 Consider a discrete-time sequence {x(n)} (a) (b) (©) (6) (2) Give the definition of the z-transform X(z) Explain what is meant by the Region of Convergence for z-ransforms. 3 State briefly any significant similarities and differences between the z-transform and ) the Laplace transform. ilustrate your answer using sketches in the z and s domains. Comment on the way in which the spectrum of discrete-time signals is represented in the z-domain 2ez+l Find the inverse z-transform of H(2)=—= 243242 1<|z]<2 {10} Let X(k) be the DFT of a discrete-time signal x() of length N samples. (a) () ©) Consider vectors x and X representing the time domain and frequency domain m7 data respectively, and a matrix y known as the DFT matrix. The DFT operation can be expressed in the following matrix form X=Dyx ‘Write out in full the vectors X and x, and the DFT matrix Dy . showing their elements in terms of x(n), X(k) and the term Wy =e7/27/" When x(n) is complex it can be written a x(n) = a(n) + JA). Show that GR) => X (ky) *X"(Cky)) and 2 1 ; yylte-* (ky) Hk where G(k) is the DFT of g(n), H(k) is the DFT of A(n), X” is the complex conjugate of X and the subscript NV indicates modulo N indexing. Given two 4-point real sequences p(n) = {1, 2,0, 1} and g(n) = {2,2,1,1},usethe 5) ‘method of part (b) to formulate, and write out in full, the matrix form for this particular case including the matrix elements. Hence find the DFTs of p(n) and q(n). Digital Signal Processing Page 3 of 5ES.07/AUVISES.11 (2) Draw the block diagram of a two-point DFT process and hence show with the aid of [6] appropriate diagrams how a four-point decimation-in-time FFT operation can be implemented efficiently. (b) Discuss briefly the features of a DSP microprocessor which would facilitate efficient [4] computation of an FFT algorithm. (©) The system shown in Figure 2 represents after with input X(2) and output ¥(2)- 4g) Deduce the important characteristics ofthis fiter and find expressions for any zeros andior poles. Figure 2. Digital Signal Processing Page 4 of 5E3.07/AUVISE3.11 A finite impulse response digital fiter has an output y(n) for an input x(1) y(n) = x(n) + x(n 3) where nis the discrete time index. (a) (b) (a) (o) (©) () Find the poles and zeros associated with this fiter and sketch a plot of them on the zplane. ‘Write an expression and sketch a plot (in dB) for the filter's magnitude response and hence determine the gain of the fier at a frequency of 7/8. Mark this frequency and corresponding gain on your plot. Define the group delay of a digital fiter and state the units in which group delay is measured. If the digital filter above operates with a sampling period of J” seconds, what is the group delay of the filter at a frequency of a What isthe relationship between system function and frequency response of a discrete-time system? ‘A. communications channel can be represented by a linear shift invariant system defined by the difference equation ¥(H) = x(n) —0.8x(n 1) + 0.8x(n~ 2). ¥(z) X(z) zeros and region of convergence on the z-plane. Write down the system function, H(z . for this system and plot its poles, Itis desired to equalise the channel in (b) by passing the signal y(n) through a fiter G(e) such that the signal x(n) is recovered exactly. Determine the required system function G(z) for the equaliser and plot the poles and zeros on the z-plane. Describe with reference to the z-plane what is meant by a non-minimum phase channel. State what problem occurs when attempting to equalize such a channel. ‘Suggest possible approaches to this problem. Digital Signal Processing Page Sof 5 (4) (6) 3 7 8) (6) (6) 15)Secu a) b) ° b) °) a "4 Wy & peews o WEE TAL Wi bwAL He eCe Wes Sauiises.t1 Vor H(2)= Sz" B(e") with E@)=De(mz" and e(n)= h(n +), Subband analysis and synthesis. The signals vi(n) represent the subband division (or analysis) of x(n). They ere formed from the effective bandpass filtering of x(n). 7 1 Al) Plz") Ryla") Pua) 1 Hy (2) | Poi") Phat 2 en h(n) =hlny-D" ‘Sketch should show symmetry points at +/- pil2. The effective frequency translation of HO to form H1 must be made clear for full marks. ROC encloses the range of z for which the z-transform expression converges. Outside the ROC the z-transform does not exist in any meaningful way. ‘The z-transform in the discrete-time case corresponds to the Laplace transform in the continuous time case. Sketches should show the significance of the unit circle vs. the jw axis. The frequency response of discrete-time signals is periodic. These are compactly represented by multiple rotations around the unit circle in z. (1) = 0.58(n) = (-1)" un) =1.5(-2)" a(n =) Digital Signal Processing Page 1 of 6E3.07/AUWISES.11 (a) (0) x0)... NDE XO) XU). XIN =DE 1 1 1 Low om D.=|1 WWE Lowe wee" ) x(n) = gy jn) XUk) = F(gin) + fmye?" x(k Sita) — shlye™"" Yk) = Liglor)— flyer?" ( 0.5(X(k)+ X°(-K)) =05| Lacon + SDA +Y ginny -iZhowr} = Lawn =G(k) Similar expressions follow for H1(K) (o) Form the complex sequence from the 2 real sequences as x(n) = pl) + jg(n) then iS 11 ot 1 i427] (446; x@|_jiog -t -sf2+2e)_] 2 | XQ) fl -b or -tf a |- 2 X@J LL FH A FILM] Ls Xj ={4-76 2 -2 -j2 W(-R)= {4-76 -j2 -2 2 Ptky={4 1-j -2 +i} Olk)={6 1-7 0 1+s} Digital Signal Processing Page 2 of 64 (2) [Bookwork} (o) bit-reversed addressing fast MAC (o) Ny =X -ac 1X, V,=X,-de" X,=X,(e+2") Y =bX, +e21X, “15d + aeye eae Yb (hd-+c0)z"+ 027 _ + (bi +c¢)2! + be" avd +aeyz 42 The filter is therefore 2™ order IIR. (bd + ce) + ce ‘The zeros are given by Hid sce) tld + c0)' — the 2c ~(d +ae)+ f+ aey — the poles are given by iach tes ooh te 2a and Digital Signal Processing E3,07/AUMISE3.11 Page 3 of 63.07/AUVISES.11 5 (a) H(z)=1 Considering the numerator, the solutions of =’ = ~1 give zeros at -1, 0.5 + j0.866, 0.5000 - j0.866. Consider the denominator, the solutions of 2° give 3 poles at the origin. 1 08 ‘ ost | o4 | 02) Imaginary Part oa J 06 | 08 wo | 4 () Magnitude (48) & 8 o 8 8 0.2 04 05 Normalized Frequency (xnrad/sample) Frequency response is given by Hel )=14 0" =14 c0s3e~ jsin3o. {ue |= J+ 0830) + inde)? = V2 + 20830 66=4.4dB For w= m8, |H(e)|=,)2+2005(344 Digital Signal Processing Page 4 of 6E3,07/AUUISE3.11 (©) Group delay is the negative derivative of phase wet frequency. Units of seconds, (d) For FIR filters such as this, the phase response is linear ~ only need to find the gradient of the linear function. Two frequency points are sufficient. aC y= tan Se 1+ 00830 / For w=0, 2=0 For w=2/6, 4=tan™ sing (2 \_ (Tr eosz/2 Therefore the group delay = 1.5T seconds at al frequencies. 6. (a) Frequency response is the z-transform evaluated on the unit circle in the z-plane, tb) AC with roots at 0.4 +/-(0.8, Imaginary Part se 08) 08 Real Part (©) Equalizer must have poles to cancel the zeros giving a)- = oy 0.825 40.8 Digital Signal Processing Page 5 of 6E3,07/AUWISES.11 os x os) oa. 02) Imaginary Part 02! al 26 os ; (4) In anon-minimum phase channel, the zeros are outside the unit circle and therefore give rise to unstable poles in the equalizer. A possible approach is to reflect the ‘unstable” poles inside the unit circle which does not affect the magnitude response of the equalizer. Digital Signal Processing Page 6 of 6